The Pennsylvania State University

The Graduate School

Department of Physics

ATOMIC LAYER DEPOSITION OF HIGH-K DIELECTRICS ON GERMANIUM AND

TRANSITION METAL DICHALCOGENIDE

A Dissertation in

Physics

by

Yuanxia Zheng

2017 Yuanxia Zheng

Submitted in Partial Fulfillment

of the Requirements

for the Degree of

Doctor of Philosophy

December 2017

ii

The dissertation of Yuanxia Zheng was reviewed and approved* by the following:

Roman Engel-Herbert

Professor of Materials Science and Engineering

Dissertation Co-Advisor

Co-Chair of Committee

Nitin Samarth

Professor of Physics

Dissertation Co-Advisor

Co-chair of Committee

Head of the Department of Physics

Mauricio Terrones

Professor of Physics

Adri van Duin

Professor of Mechanical & Nuclear Engineering

*Signatures are on file in the Graduate School

iii

ABSTRACT Two topics related to atomic layer deposition (ALD) have been studied in this thesis. One

is the challenging task of integrating high-k dielectric on germanium (Ge) surface. The other is

utilizing an ALD approach to synthesize transition metal dichalcogenide (TMD) of 1T-TaS2.

The surface preparation primarily using in-situ H2 plasma to obtain pristine Ge surfaces

has been investigated. The reaction mechanism and the resultant material properties have been

examined carefully using in-situ and ex-situ metrologies. An optimized process has been proposed

and resulted in an oxygen-free and atomically flat Ge surface.

The nucleation behavior of Al2O3 ALD was investigated on Ge surfaces with two different

chemicals states, hydrogenated and oxidized. The growth mode and the resultant dielectric/Ge

interface properties have been clarified using in-situ and ex-situ metrologies. By comparing the

experimental results with an atomic scale simulations (from collaborators), the reaction mechanism

as well as the thermodynamic properties have been identified.

A trilayer dielectric gate stack of HfO2/Al2O3/GeOx has been used to electrically test the

aforementioned mechanisms of dielectric ALD on Ge. The optimum process has yielded a highly

scaled Ge MOSCap device with superior interface qualities.

1T-TaS2 thin films has been synthesized using TaCl5 and H2S as the precursors in a home-

made ALD system. A strong temperature dependence has been identified. A use of ultrathin Ta2O5

seed layer has been found beneficial to facilitate the nucleation of 1T-TaS2. ALD growth at a high

temperature of 480 °C has yielded the optimum results.

Ferroelectric HfO2 has also been synthesized as the gate insulator for the future transistor

fabrication using 1T-TaS2 as the channel material. A process flow for Al-doped HfO2 primarily

using ALD approach in conjunction with magnetron sputtering has been developed. The electrical

properties for various doping levels have evaluated using electrical polarization measurements.

iv

TABLE OF CONTENTS

List of Figures .......................................................................................................................... vi

List of Tables ........................................................................................................................... xv

Acknowledgements .................................................................................................................. xvi

Chapter 1 Introduction ............................................................................................................. 1

1.1 A Brief Review of CMOS Technology ...................................................................... 1 1.2 Advantages and Challenges of Germanium Transistors ............................................ 8 1.3 Properties of 2D Layered Material 1T-TaS2 .............................................................. 17 1.4 A Brief Review of Transition Metal Dichalcogenide Synthesis ................................ 27 1.5 Thesis Organization ................................................................................................... 33

Chapter 2 Experimental Techniques ........................................................................................ 34

2.1 Atomic Layer Deposition ........................................................................................... 34 2.1.1 Principles of Atomic Layer Deposition ........................................................... 35 2.1.2 Hardware Information for High-k ALD .......................................................... 38 2.1.3 Hardware Information for TMD ALD ............................................................ 40

2.2 Spectroscopic Ellipsometry ........................................................................................ 44 2.2.1 Principles of Ellipsometry ............................................................................... 45 2.2.2 Data Analysis .................................................................................................. 49 2.2.3 Material Parameterization ............................................................................... 54 2.2.4 Hardware Information ..................................................................................... 55

2.3 Admittance Characterization ...................................................................................... 57 2.3.1 Principles of MOSCap Admittance ................................................................. 57 2.3.2 High Frequency Admittance Measurement ..................................................... 62 2.3.3 Other Electrical Characteristics ....................................................................... 66 2.3.4 Data Analysis .................................................................................................. 71 2.3.5 Hardware Information ..................................................................................... 72

2.4 X-ray Photoelectron Spectroscopy ............................................................................. 74 2.4.1 Principles of X-ray Photoelectron Spectroscopy ............................................. 74 2.4.2 Data Analysis .................................................................................................. 75 2.4.3 Hardware Information ..................................................................................... 79

Chapter 3 Atomic Layer Deposition of High-k Dielectrics on Germanium ............................ 81

3.1 Introduction ................................................................................................................ 81 3.2 Preparing Pristine Ge Surfaces................................................................................... 82

3.2.1 Experimental details ........................................................................................ 83 3.2.2 XPS Studies ..................................................................................................... 85 3.2.3 AFM Studies ................................................................................................... 87 3.2.4 Raman Studies ................................................................................................. 92 3.2.5 Discussions and summary ............................................................................... 95

v

3.3 Al2O3 ALD Nucleation Mechanism on Ge ................................................................ 97 3.3.1 Baseline for ALD growth ................................................................................ 98 3.3.2 Al2O3 ALD on H-terminated Ge surface ......................................................... 100 3.3.3 Al2O3 ALD on oxidized Ge surface ................................................................ 107

3.4. Electrical characterization of Ge MOSCap devices .................................................. 114 3.5 Summary .................................................................................................................... 119

Chapter 4 Atomic Layer Deposition Synthesis of 1T-TaS2 ..................................................... 121

4.1 Introduction ................................................................................................................ 121 4.2 ALD Growth of 1T-TaS2 ............................................................................................ 122

4.2.1 Experimental details ........................................................................................ 122 4.2.2 ALD growth at low-mid temperatures ............................................................ 123 4.2.3 ALD growth at high temperatures ................................................................... 128 4.2.4 Synthesis of ferroelectric HfO2 ....................................................................... 133

Chapter 5 Conclusions, Future Work and Outlook .................................................................. 143

Reference ................................................................................................................................. 148

vi

List of Figures

Figure 1.1. Structural schematics for (a) NPN bipolar transistor, (b) p-MOSEFT and (c)

CMOS. Figure (a) is taken from Ref. [2]. ....................................................................... 5

Figure 1.2. A brief chronology of the major milestones in the semiconductor industry. The

data is mainly from Ref. [1]. ........................................................................................... 5

Figure 1.3. Moore’s law from 1970 to 2017. The VLSI standard requires >104

transistor/chip, which occurred after ~1977. The background figure (year ≤ 2015) is

taken from Ref. [3], while the red star is the 5nm node result announced by IBM in

June, 2017. [4] .................................................................................................................. 6

Figure 1.4. (a) Technology node size for semiconductor manufacturing processes (1971-

2017). [4] The red star is the 5 nm node announced by IBM in June, 2017.[4] (b) Gate

leakage current increase as a result of gate oxide (SiO2) thickness (tox) downscaling

(36-15 nm). [5] (c) Drain/source leakage current at off state increase as result of

transistor physical length downscaling. [6] ...................................................................... 6

Figure 1.5. The structures of (a) high-k/metal gate planar MOSEFT, [7] (b) tri-gate

FET, [8] and (c) gate-all-around FET. [4] ........................................................................ 7

Figure 1.6. (a) The miniaturization of transistors in VLSI is gradually approaching the

physical limit. [9] (b) Replace the low mobility Si channel in P-MOS with a high

mobility semiconductor of Ge. ......................................................................................... 7

Figure 1.7. (a) Unit cell of Ge crystal structure. [34] (b) Band structure of Ge at room

temperature. [32] .............................................................................................................. 14

Figure 1.8. (a) Electron mobility e and (b) hole mobility h for various semiconductors as

functions of band gap EG. [32] ......................................................................................... 14

Figure 1.9. (a) Interface reaction between GeO2 and Ge substrate. (b) Interfacial reaction

between HfO2 and Ge substrate. (c) A schematic for the design of using interlayer

between high-k and Ge to form stable dielectric/Ge interface. (d) TEM images for the

interfaces after plasma post oxidation on Al2O3/Ge at 300 C and room temperature.

Figure (d) is copied from Ref. [29]. ................................................................................. 15

Figure 1.10. Process optimization for integrating high-k dielectrics on Ge surface.

Compared to the traditional black box mode, an application of in-situ metrology in

conjunction with atomic scale simulations provides a more straightforward route to

process optimization. [35] ................................................................................................ 16

Figure 1.11. (a) Resistivity as a function of temperature for bulk 1T-TaS2, reproduced from

Ref. [41]. Schematic views for device switching performance in (b) Boltzmann and

(c) Landau switch, respectively........................................................................................ 22

Figure 1.12. (a) Crystal structure of 1T-TaS2 at T> 850 C. The structural parameters are

from Ref. [47]. (b) Crystal reconstruction in CCDW phase (T< TC); the dash line

vii

indicates the supercell by 13𝑎×13𝑎 reconstruction; the arrow indicates the

displacement of Ta atoms with respected to the un-reconstructed phase at T> 850 C;

the corresponding shrinkage for A-B and B-C distances is 6.4 and 3.2%,

respectively. [60] .............................................................................................................. 22

Figure 1.13. STM images for (a) CCDW [48], and (b) NCCDW [61] phase of 1T-TaS2,

respectively. (c) A schematic view for phase transition from commensuratenearly

commensurateincommensurate CDW phase of 1T-TaS2 with increasing

temperature; the dark region represents David-star reconstruction, while the

reconstruction in the gray region is considerably suppressed or removed; the red arrow

represents the current flow in NCCDW. [42] .................................................................. 23

Figure 1.14. Ta3d orbital splitting by the crystal field. [50] (b) The Brillouin zone (BZ) of

1T-TaS2 at CCDW phase (T<TC); the larger BZ (solid) corresponds to the un-

reconstructed phase (T>850C), while the smaller BZ (dash) is for the supercell after

reconstruction at CCDW phase. [60] (c) Band structure of CCDW phase with

considering spin-orbital coupling for one monolayer 1T-TaS2; the band gap

EG≈0.2eV. [50] (d) Density of state (DOS) contribution from a, b, and c sites of Ta

in David-star (Fig. 2b) at CCDW phase. [50] (e) Density of state (DOS) contribution

from different Ta3d-orbitals at CCDW phase. [50] ......................................................... 24

Figure 1.15. Schematic views for (a) metallic narrow band (U=0), (b) Mott insulator (U>W,

at low temperature), and (c) Mott insulator melted by thermally excited free electrons,

respectively. EF is the Fermi level, U is the Coulombic interaction between electrons,

and W is the band width. .................................................................................................. 25

Figure 1.16. (a) Schematic structure of ionic field effect transistor (iFET) made of 10nm

1T-TaS2. [54] (b) Resistance R vs gate voltage VG at T= 325 K of the 1T-TaS2 iFET

in (a).Results are from Ref. [54]. ..................................................................................... 25

Figure 1.17. (a) Schematic demonstration of charge control in La1-xSrxMnO3/La1-

xCaxMnO3/BaTiO3/La1-xSrxMnO3 tunneling hetero-junction. (b) Pulsed voltage

control of the resistivity of the hetero-junction in (a). The response of the hetero-

junction in (a) to the time-dependent control voltage. Figures (a)-(c) are from

Ref. [56]. (d) The phase diagram for La1-xCaxMnO3 system. The figure (d) is from

Ref. [57]. .......................................................................................................................... 26

Figure 1.18. (a) Resistivity vs temperature for Se-doped 1T-TaS2. [58] (b) ID-VG

characteristics of Hf1-xSixO2-FeFET on n-Si with 500nm gate length. [59] (c) A

proposed design of 1T-TaS2 FeFET. ................................................................................ 26

Figure 1.19. Schematics of growth setup for (a) CVT, [75,76] (b) CVD, [63] (c) van der

Waals epitaxy, [77] and (d) solvothermal method. [66] .................................................. 30

Figure 1.20. (a) Schematic illustration of one growth cycle of an ALD MoS2 film on c-

sapphire. (b) SEM and AFM images for MoS2 after 800C anneal. The results are

from Ref. [78]. ................................................................................................................. 31

viii

Figure 1.21. Controlled synthesis of ultrathin 1T-TaS2 crystals via a CVD method. (a)

Schematic of CVD setup for the growth of 1T-TaS2 on a SiO2/Si substrate with

tantalum pentachloride powder and sulfur powder used as the precursors. (b, c) Crystal

structure of distorted 1T-TaS2 on a SiO2/Si substrate. (d, e) Optical images of 1T-TaS2

ultrathin flakes. (f) The controlled thicknesses of 1T-TaS2 at different growth times.

The inset shows an octahedral arrangement of the central Ta atom coordinated with S

atoms. (g−l) AFM images and their corresponding height profiles of various 1T-TaS2

at different growing times. The scale bars are 10 μm in parts d and e and 2 μm in parts

g, h, and i. These results are from Ref. [74]. .................................................................... 32

Figure 2.1. A schematic for the sequential process for growing Al2O3 by ALD using TMA

and H2O as the precursors. ............................................................................................... 38

Figure 2.2. A schematic for the ALD system of ALD-150LX. The figure was originally

plotted by Jason Lapano from Roman Engel-Herbert group at MatSE, Penn State

University. ........................................................................................................................ 40

Figure 2.3. A schematic of the home-made ALD system for synthesizing transition metal

dichalcogenides. ............................................................................................................... 42

Figure 2.4. The structure of the showerhead used in the home-made ALD system. ............... 43

Figure 2.5. The structure of the heater used in the home-made ALD system. ......................... 43

Figure 2.6. A schematic for the setup of the ellipsometry measurement. ................................ 48

Figure 2.7. Reflection and transmission of an incident light at the boundary between a

homostructure sample and air/vacuum in the SE measurement. ...................................... 48

Figure 2.8. (a) A schematic for the backside reflection of the sample and resultant

interfering signal in the SE measurement of non-absorbing substrate. (b)

Depolarization of the SE signals for single- and double-side polished Si substrates. ...... 48

Figure 2.9. Multi-reflection inside the thin film for measuring a film/substrate

heterostructure by SE. ...................................................................................................... 52

Figure 2.10. The evolution of ellipsometric angles / with respect to an increasing

thickness of SiO2 thin film (d=0~284.2 nm, n1=1.457, k1=0) on a Si substrate

(n2=3.8812, k2=0.0195). The incident light is HeNe laser with a wavelength =632.8

nm at an incident angle of 70 °. (a) /evolution as functions of the SiO2 thickness.

(b) /evolution trajectory as the SiO2 thickness increases from 0 to 284.2 nm. .......... 52

Figure 2.11. / evolution trajectory for growing of a-Si:H on a single crystal Si substrate.

The refractive index of a-Si:H is n=5-i0.85 at 500 nm. This figure is reproduces from

Ref. [97]. .......................................................................................................................... 53

Figure 2.12. Schematic for the bottom-up strategy of characterizing multi-layer structure

by SE. ............................................................................................................................... 53

ix

Figure 2.13. Dielectric function models used in ellipsometry data analysis. Figure is copied

from Ref. [97]. HOA stands for harmonic oscillator approximation, and MDF for

model dielectric function.................................................................................................. 53

Figure 2.14. Examples for the dispersion of (a) Lorentz, (b) Cauchy, and (c) Tauc-Lorentz

models used in the ellipsometry data analysis. ................................................................ 54

Figure 2.15. The optic configuration for M-2000U J.A.Woollam spectroscopic

ellipsometer. ..................................................................................................................... 56

Figure 2.16. (a) Equivalent circuit for the admittance measurement of a MOSCap device.

(b) The contributions to the capacitance from the gate oxide and semiconductor. .......... 60

Figure 2.17. Simulated band diagrams, stack capacitance Cstack, and semiconductor

capacitance Cs for (a) accumulation, (b) flat band, (c) depletion, and (d) inversion

regimes of an ideal MOSCap made of Cu(3nm)/HfO2(3nm)/p-Si (Na=3.3e17 cm-3).

The work function of Cu is m=4.5 eV, dielectric constant of HfO2 is k=25, and

dielectric constant of Si is k=11.7. The calculated Debye length is LD 7.1 nm. The

threshold voltage is Vth= 0.368 V. .................................................................................... 61

Figure 2.18. (a) The capacitance contributions for a MOSCap device with interface trap

states (Dit). Cit represents the capacitance from the interface trap states Dit. (b) Stretch-

out effect in C-V characteristics caused by the interface trap states Dit. .......................... 65

Figure 2.19. (a) Charge exchange between an interface trap state (Eit) and the majority-

carrier band (valence band for p-type semiconductor). (b) Frequency dispersion in C-

V characteristics caused by the interface trap states Dit inside the band gap. .................. 66

Figure 2.20. (a) Charge exchange between a border trap state (Nbt) and the accumulation

charge carrier (holes for a p-type semiconductor) at the semiconductor surface near

the Fermi level (EF). (b) The equivalent circuit for a MOSCap device with border trap

states; this schematic is reproduced from Ref. [104]. ...................................................... 66

Figure 2.21. (a) Hysteresis (VFB) in a bi-directional C-V measurement of MOSCap. (b)

The response of mobile oxide charge to the gate voltage modifies the electric field

inside the gate dielectric, and therefore modifies the band bending (dash). .................... 70

Figure 2.22. (a) MOSCap using metal/high-k/Si structure. The figure is released by

Intel. [108] (b) Benchmarking the band gap vs the dielectric constant for various

dielectrics. The figure is copied from Ref. [109].In particular, the band gap and

dielectric constant for GeO2 are EG=6.1 eV [110] and k~4.5. [16] .................................. 70

Figure 2.23. (a) Schematic and (b) equivalent circuit for measuring JG-V measurement of

MOSCap. (c) Example of leakage current as a function of gate voltage. ........................ 70

Figure 2.24. (a) Principles of XPS measurements: interaction between the X-ray photon

and inner core electrons. (b) An XPS survey scan for a GeO2(10 nm)/Ge sample. ......... 75

x

Figure 2.25. (a) XPS signal from a homogenous material. (b) XPS signal from a GeO2(10

nm)/Ge sample. ................................................................................................................ 78

Figure 2.26. A high resolution XPS measurement of Ge3d for a GeO2(10 nm)/Ge sample.

The take-off angle of the measurement is 90 °. ............................................................... 79

Figure 2.27. Calibrating SiO2 thickness on Si substrate using angle resolved XPS. This

figure is reproduced from Ref. [116]. .............................................................................. 79

Figure 2.28. The hardware setup for an XPS measurement. The figure is take from

Ref. [117]. ........................................................................................................................ 80

Figure 3.1. A schematic for the remote inductively coupled H2 plasma source. An induction

coil is used to generate AC electromagnetic field inside quartz tube. The flowing gas

molecules are ionized or dissociated and then react with the sample surface. ................. 85

Figure 3.2. High resolution XPS measurements of Ge 3d at a 90 ° take-off angle for

GeO2(10 nm)/Ge samples (a) before H2 plasma, (b) after H2 plasma at T=110 °C, and

(c) after H2 plasma at T=330 °C. The numerical deconvolution is based on

References [112–114] with a Tougaard background profile. A Voigt profile (0.667

branching ratio) is used to describe all the peak shapes. The Ge0 3d5/2 peak is

referenced to 29.3 eV. Binding energy shift for Ge 3d3/2 with respect to Ge 3d5/2 is

0.58eV, and the areal ratio between Ge 3d3/2 and Ge 3d5/2 is fixed at 0.67. The core

level shifts for +1, +2, +3, and +4 are 0.8,1.8,2.7,3.3 eV, respectively. .......................... 87

Figure 3.3. (a) A schematic for using a shadow mask to create a step during the H2 plasma

clean. AFM line scans across the height steps and the corresponding schematics for

the resultant structures created in the H2 plasma at (b) 110 °C and (c) 330 °C, with the

left side of the steps covered by the shadow mask. .......................................................... 90

Figure 3.4. (a) In-situ SE monitoring of H2 plasma clean of GeO2(~1 nm)/Ge(100) at 110

and 270 °C. (b) The surface morphology of Ge(100) substrate after H2 plasma clean.

(c) The surface roughening as a function of temperature. ................................................ 91

Figure 3.5. Raman measurements for GeO2(10 nm)/Ge samples (a) before H2 plasma, (b)

after H2 plasma at T=110 °C, and (c) after H2 plasma at T=330 °C. The plasma treated

samples were subsequently in-situ capped with 2 nm Al2O3 by ALD in order protect

the surface from air contamination during the sample transfer. (d) Areal percentage of

poly-Ge signal for H2-plasma treated GeO2(10 nm)/Ge samples as a function of the

process temperature. The poly-Ge percentage is normalized to the Ge bulk signal. A

linear background profile was used for the numerical deconvolution. A Voigt profile

(0.98 branching ratio) is used to describe all the peak shapes. The Ge bulk peak was

referenced to 300.9 cm-1. .................................................................................................. 94

Figure 3.6. Ellingham diagram calculation for the decomposition of GeH4. The same

calculation was also done for SiH4 as a comparison. The calculation was based on the

thermochemical data from Reference [33]. ...................................................................... 94

xi

Figure 3.7. The optimized cleaning process results in a pristine Ge surface free of oxygen,

with minimized formation of nanocrystalline Ge, atomically flat surface and a well-

defined (2×1) surface reconstruction. .............................................................................. 96

Figure 3.8. Al2O3 ALD on Al2O3(30 nm)/SiO2(25 nm)/Si. (a) In-situ SE monitoring of real-

time thickness. (b) Enlarged section for Figure (a) with details of the sequential

precursor doses, TMA adsorption and GPC. (c) Extracted ratio of GPC to TMA

adsorption as a function of growth cycle. Results published in Ref. [35]. ...................... 100

Figure 3.9. Al2O3 ALD on Ge: H surface. (a) In-situ SE monitoring of growth curve. (b)

TMA dose rise and extracted growth per cycle (GPC) as a function of ALD cycles;

the blue dash line is the base growth rate for baseline Al2O3 ALD (~0.86 Å/cycle). (c)

GPC: TMA adsorption ratio as a function of ALD cycles; the blue line is the baseline

ratio of ~0.404. AFM images for different A2lO3 ALD stages: (d) Ge: H surface before

ALD, (e) 15 ALD cycles, and (f) 27 ALD cycles, respectively. Surface roughness is

represented by root-mean-square (RMS) of height. Results published in Ref. [35]. ...... 104

Figure 3.10. Reaction profiles obtained by the ReaxFF-nudged elastic band scheme for

Al2O3 ALD on H-terminated Ge(100). (a) TMA and (b) H2O adsorptions at H-

terminated site. (c) TMA adsorption on a Ge-dangling bond. (d) H2O removes one

CH4 group from Ge-Al(CH3)2* site. (e) H2O removes CH4 group from Ge-

Al(CH3)(OH)* site. (f) TMA adsorption at Ge-Al(OH)2* site. The insets correspond

to the atomic structures for the stage at the solid points in each profile, while the

neighboring number denotes the corresponding energy. Results published in

Ref. [35]. ......................................................................................................................... 105

Figure 3.11. Results of MD simulations at 500 K for (a) 80 TMA molecules on a H-

terminated Ge(100) surface with a single Ge-dangling bond; (b) 100 H2O molecules

on the H-terminated Ge(100) surface with a Ge-Al(CH3)2* site. ReaxFF-MD results

confirm that TMA and H2O molecules preferably adsorb on the Ge-dangling and Ge-

Al(CH3)2* sites, respectively, rather than H-terminated Ge sites. Results published in

Ref. [35]. ......................................................................................................................... 106

Figure 3.12. In-situ SE monitoring for three GeOx growth modes at 270 C: (a) molecular

O2 (pO2≈33 mTorr/2min), (b) continuous O*-plasma (8 sec), and (c) sequence of O*-

plasma pulses (1.75 sec/pulse). Note the shorter time scale in (b). Results published

in Ref. [118]. .................................................................................................................... 110

Figure 3.13. (a) In-situ SE monitoring for Al2O3 ALD on GeOx(5 Å)/Ge(100). Inset is the

AFM image right after Al2O3 ALD; RMS=0.285 nm. (b) TMA dose rise and extracted

GPC as a function of ALD cycles; the ALD base GPC (blue dashi line) is ~0.86

Å/cycle. (c) Al2O3-growth/TMA-dose ratio extracted from ellipsometry; the blue line

of 0.4 is the typical value for Al2O3 ALD. (d) XPS measurements (Ge 3d at 90 ° take-

off angle) after 1 nm Al2O3 ALD on different starting surfaces, Ge: H, GeOx(3

Å)/Ge(100), and GeOx(5 Å)/Ge(100), respectively. Inset is the XPS measurements at

15 ° take-off angle. Results published in Ref. [35]. ........................................................ 111

Figure 3.14. Ellingham calculations for As2O3, GeO2, Ga2O3 and Al2O3, respectively.

Results published in Ref. [35]. ........................................................................................ 112

xii

Figure 3.15. ReaxFF-MD simulations of Al2O3 ALD using TMA+H2O dose on

GeOx(5Å)/Ge(100) surface at 1000 K. (a) Final snapshots of MD simulations for the

sequential precursor dose; the MD-NVT simulations were performed for each dose

up to 500 ps; the highlighted molecules 1 and 2 are byproducts of methane (CH4) and

ethane (H3C-CH3), respectively. (b) ReaxFF-NEB scheme for TMA adsorption on

GeOx/Ge surface. (c) Number of oxygen atoms coordinated to Ge during Al2O3 ALD.

(d) Oxygen origin of Al2O3 grown on GeOx/Ge surface after Al2O3 ALD. Results

published in Ref. [35]. ..................................................................................................... 112

Figure 3.16. ReaxFF-MD simulations for oxygen diffusion from GeOx into underlying Ge

subsurface for GeOx/Ge(100) (red) and [Al2O3/GeOx]/Ge(100) (blue) interfaces,

respectively. To accelerate the oxygen diffusion effect within a limited time frame

(2000 ps), a high temperature (800 K) was assigned to oxygen atoms, while the other

systems were kept at 300 K with the NVT-ensemble. The portion of oxygen diffusion

was normalized by the total number of oxygen atoms in GeOx layer. ............................. 113

Figure 3.17. A structural schematic for a MOSCap device using [Al2O3/GeOx] as the

interlayer between high-k and Ge. ................................................................................... 116

Figure 3.18. In-situ SE monitoring of the fabrication process: (a) H*-plasma clean of native

GeOx, (b) Ge passivation by pulsed O*-plasma, (c) Al2O3 thermal ALD, and (d) HfO2

thermal ALD, with all oxides described as Cauchy model. Results published in

Ref. [118]. ....................................................................................................................... 117

Figure 3.19. C-V characteristics of HfO2(24 Å)/Al2O3(10 Å)/GeOx/p-Ge(100) MOSCAPs

with varying GeOx thickness: (a) 0.0, (b) 2.5, and (c) 5.0 Å. (d) The effect of GeOx

thickness on MOSCAP performance. VFB is the flat-band voltage, and EV is the

valence band edge. Dit@Ev were extracted using conductance method. [169] ................ 117

Figure 3.20. C-V of HfO2(24 Å)/Al2O3/GeOx(5 Å)/p-Ge MOSCAPs with Al2O3 of (a) 0.0

Å (0 cycle), (b) 2.2 Å (4 cycles), (c) 4.9 Å (9cycles), and (d) 10.1 Å (15 cycle); (e) the

corresponding gate leakages. ........................................................................................... 118

Figure 3.21. Electrical characterization of Ni(60 nm)/HfO2(24 Å)/[Al2O3(varied

cycles)/GeOx(5 Å)]/p-Ge(100) MOSCaps. (a) Capacitance-voltage (C-V) and gate

leakage-voltage (JG-V) for the device with 9 Al2O3-ALD cycles. The Al2O3-ALD

dependence for (b) maximum capacitance (Cmax) for C-V at f=1.5 MHz, (c) gate

leakage current (JG) at over-driving V=VFB-1V (VFB is the flat-band voltage), (d) C-V

hysteresis (flat-band voltage shift, VFB), (e) interface trap density (Dit) at valence

band edge (E=EV). The Dit was obtained by analyzing frequency-dependent C-V

characteristics based on conductance method. [169] ....................................................... 118

Figure 3.22. Gate leakage (JG@VFB-1V) vs. EOT benchmark of Ge MOSCAPs using

HfO2(24 Å)/Al2O3/GeOx(5 Å) gate stacks with 0, 4, 9, and 15 cycles of ALD Al2O3

(indicated by dash arrow). EOT is calculated using Cmax@1.5 MHz. .............................. 120

Figure 4.1. The working principle of QCM for monitoring growth processes. [170] ............. 123

Figure 4.2. The cleaning process for cleaning a sapphire substrate. ........................................ 123

xiii

Figure 4.3. (a) Real-time thickness measured by in-situ QCM for ALD growth of 1T-TaS2

at low-mid temperatures (220-380 °C). (b) An enlargement for the growth at 380 °C

indicated by the black box in Figure (a). (c) 100-cycle growth represented by QCM

signals as a function of temperature. (d) Temperature dependence for QCM signal of

100 ALD cycles. .............................................................................................................. 126

Figure 4.4. A thermodynamic calculation of the ALD reaction. The critical temperature is

estimated to be 453 °C. .................................................................................................... 126

Figure 4.5. AFM measurements for TaS2 ALD growth at T=380 °C. (a) Surface

morphology at difference ALD cycles. A line scan (white dash) is shown as an inset

in each AFM image. (b) The surface coverage as a function of ALD cycle, extracted

by image processing of the AFM morphology using ImageJ software. (c) Raman

spectrum for the sample with 1000 ALD cycles. ............................................................. 127

Figure 4.6. AFM image after 500 cycles of TaCl5+ H2S ALD growth on a c-sapphire

substrate. .......................................................................................................................... 131

Figure 4.7. AFM image after 500 cycles of TaCl5+ H2S ALD growth on a Ta2O5/c-sapphire

substrate. .......................................................................................................................... 131

Figure 4.8. Raman spectrum for the sample with 500 cycles of TaCl5+H2S ALD on a

Ta2O5/sapphire substrate. ................................................................................................. 132

Figure 4.9. (a) Setup for H2S annealing the sample with 500 cycles of TaCl5+H2S ALD on

a Ta2O5/sapphire substrate. The sample photo on the left is before anneal and the one

on the right is after anneal. (b) The Raman spectrum of the sample after anneal. (c)

Raman spectrum for TaS3. [176] ...................................................................................... 132

Figure 4.10. Electric current switching of MIT transition in (a) 10 nm thick 1T-TaS2, [198]

and (b) monolayer 1T-TaS2. [178] ................................................................................... 137

Figure 4.11. Bulk carrier density of 1T-TaS2 as a function of temperature. [199] The table

on the right compares the monolayer carrier density for various materials. .................... 137

Figure 4.12. The mechanism of forming ferroelectric HfO2 at room temperature. ................. 138

Figure 4.13. The process flow for developing Al doped HfO2 ferroelectrics. ......................... 138

Figure 4.14. Characterizations of TiN sputtering. The sputtering parameters are: CMS18-

#1 sputter system in PSU Nanofab, 300Watt DC/15Watt AC, Ar:N2= 14:4.5 sccm, 5

mTorr, 3000 sec. (a) AFM measurement of the thickness step created by using a

shadow mask during the sputtering. (b) Surface morphology of sputtered TiN films

(56.23 nm) on c-plane sapphire (RMS 0.1 nm). (c) The dielectric function extracted

from spectroscopic ellipsometry, assuming a thickness of 56.23 nm. (d) XPS depth

profiling of TiN film deposited on a p+-Si substrate. ....................................................... 139

Figure 4.15. In-situ SE monitoring of 300 °C ALD growth for (a) Al2O3 calibration, (b)

HfO2 calibration, and (c) Al doped HfO2 with 1:20 cycle ratio. ...................................... 140

xiv

Figure 4.16. Process of defining electrode pattern for measuring MIM. ................................. 140

Figure 4.17. Ex-situ SE monitoring the SC1 treatment on TiN/Si samples. (a) as a

function of treatment time. (b) as a function of treatment time. (c) A schematic for

SC1 treatment. (d) at E=2.5 eV as a function of treatment time. ............................. 141

Figure 4.18. Electrical polarization as a function of electric field for MIM structures using

an insulator of (a) purge HfO2, (b) 6.6% Al doped HfO2, and (c) 12.3% Al doped

HfO2. (d) Dielectric constants for the insulators as a function of Al doping. .................. 142

xv

List of Tables

Table 1.1. Scaling MOSFET device and circuit parameters with keep a constant electric

field. This table is mainly reproduced from Ref. [1]. ...................................................... 4

Table 1.2. Basics parameters of single crystal Ge and Si. Most of the parameters are taken

from Ref. [32]. ................................................................................................................. 13

Table 2.1. Parameters for the precursors used in thermal ALD. .............................................. 39

Table 2.2. Parameters for the plasma treatments. .................................................................... 39

Table 2.3. The dielectric function of Ge substrate at T=270 °C is fitted with generalized

oscillators (Gen-Osc). inf is the contribution from the optical transitions at higher

energies, is the peak broadening, E0 is the peak transition energy, A is the transition

amplitude, and Eg is the band gap of the corresponding optical transition. ..................... 55

Table 2.4. The refractive index (n) of GeOx, Al2O3 and HfO2 deposited by ALD at T=270

°C are fitted with Cauchy model. Cn term has trivial contribution, so is not included

in the fitting. Since the band gaps of the three oxides are beyond the spectrum range

(1.24~5.18 eV), the oxides are considered as transparent with extinction coefficient

k=0. ................................................................................................................................... 55

Table 2.5. XPS parameters for Ge3d, O1s and Al2p orbitals used in this work. The calculation

of the inelastic mean free path is performed based on Ref. [115]. The XPS light source

is Al K (h= 1486.70 eV). ............................................................................................. 78

Table 2.6. Some basic properties of materials used in this work. ............................................ 78

Table 3.1. Comparison between the calculated and measured step heights created by the H2

plasma clean at 110 and 330 °C. The starting sample before the H2 plasma was using

a GeO2(10.1 nm)/Ge structure. The calculation was assuming (i) reaction mechanisms

of routes (I) and (II) for 110 and 330 °C, respectively, and (ii) 3.06 nm GeOx formation

by air oxidation on a plasma cleaned Ge surface. A density of 5.32 g/cm3 was used for

both the Ge bulk and Ge overlayer created by the H2 plasma, while 4.25 g/cm3 for

GeO2 and native GeOx formed by air exposure. .............................................................. 89

Table 3.2. De-convoluted peak areas of XPS measurements for samples after Al2O3 ALD

on Ge: H, GeOx(3 Å)/Ge(100), and GeOx(5 Å)/Ge(100), respectively. The peak area

is normalized by the area sum of Ge0 3d3/2 and Ge0 3d5/2. Results published in

Ref. [35]. ......................................................................................................................... 110

Table 4.1. Parameters for ALD growth of 1T-TaS2. ................................................................ 136

Table 4.2. Growth rates for Al2O3 and HfO2 at 300 °C. ........................................................... 136

xvi

ACKNOWLEDGEMENTS

I want to gratefully thank my advisors, Dr. Roman Engel-Herbert, and my committee

member Dr. Ari van Duin, for their diligent guidance and direction of my efforts and energy over

the years. It has been greatly fortunate to work with awesome colleagues in the lab, including Dr.

Matt Brahlek, Dr. Craig Eaton, Dr. Lei Zhang, Dr. Ryan Haislmaier, Haitian Zhang, Jason Lapano,

and Joseph Roth. Their passion for science and kind help with research have facilitated me to grow

into a researcher and engineer. Special thanks to my collaborators, Dr. G. Bruce Rayner, Jr. and

Dr. Sungwook Hong for experimental and theoretical assistance and discussions. I would like to

also thank my other friends at Penn State, Dr. Renzhong Du, Dr. Wenqing Dai, Dr. Shiming Lei,

Xiaoyu Ji, Yakun Yuan, Chenjin Zhang for all their help. Thanks to many staffs of Material

Research Institute for their experimental assistance, including Jacob Lyons, Charles Cole, Bangzhi

Liu, Bill Drawl, Tim Klinger, Andrew M. Fitzgerald, Shane P. Miller, Sarah Eichfeld, Tim Tighe,

Maria DiChol, Amanda Baker, and Jeff Long. Many thanks to the funding agents, Intermolecular

Inc. and National Science Foundation (EFRI) for making this research financially possible. Finally,

a big thank to my family, parents and brother, for their love and support.

Chapter 1

Introduction

1.1 A Brief Review of CMOS Technology

A transistor is a three-terminal semiconductor device that can regulate electronic signals

(current or voltage) and act as a switch or gate in integrated circuits (ICs). There are two basic

requirements for transistors: (i) tunability of output by a small electrical input, and (ii) saturation

of output at certain threshold input. The first transistor was the bipolar transistor invented in 1947.

A bipolar transistor utilizes a small input current to control an amplified output (see Fig. 1.1a) and

until now the bipolar circuits still remain the fastest at the individual-circuit level, but the large

power consumption has limited their application for a high-density integration (<104

circuits/chip). [1] Instead, the idea of modulating the surface conductance of the semiconductor

channel using electric field was firstly realized in 1960, by using a metal-oxide-semiconductor

field-effect transistor (MOSFET) structure (see Fig. 1.1b). In 1963, the invention of using a

complementary metal-oxide-semiconductor (CMOS) structure (see Fig. 1.1c) made possible a

negligible standby power consumption because there is always one MOS out of the pair at the OFF

state. This breakthrough soon became the foundation of the so-called “very-large-scale-integration”

(VLSI) manufacturing in semiconductor industries (see Fig. 1.2). The VLSI campaign among

semiconductor industries was mainly driven by the famous “Moore’s law”, namely, the areal

density of transistors on ICs doubled every year. In reality, the pace was slightly slow down to a

doubling rate of every 18 months. But amazingly, even after 50 years or so, this law is still working

as a guideline for the semiconductor industries (see Fig. 1.3). Beginning from 1970, the transistor

density has increased significantly from ~104/chip to >109/chip.

2

The key of keeping the Moore’s law alive for such a long time is miniaturization of the

transistor dimensions. Tab. 1.1 shows the scaling effect on MOSEFT parameters following a

constant field rule or Dennard scaling rule, in order to minimize the short channel effect when

scaling down the gate length. [1] The downscaling of transistor dimensions allows more devices

per chip operating at a faster speed, so the calculation number per chip is significantly increased.

Fig. 1.4 shows the trend of technology node size. But Dennard scaling cannot keep going forever.

The first roadblock was that when scaling down the gate oxide thickness to <2 nm, which was SiO2

(dielectric constant k=3.9) in early generations of MOSEFTs, the quantum tunneling effect

becomes large, and as a result the power consumption from the leakage is ineligible and the

reliability of the device performance is compromised. To address this issue, Intel introduced

Hafnium-based high-k dielectric in junction with metallic gate and successfully produced the 45

nm node in 2008. However, as further downscaling the node size, another roadblock became

challenging, which was the leakage between drain and source caused by the drain-induced barrier

lowering (DIBL) effect in a short channel node. [1] An overcome of this technical issue requires a

better gate control of the channel. A further downscaling of gate oxide thickness even if using high-

dielectrics did not appear applicable due to the increase of gate leakage issue. Instead of

downscaling the planar transistor dimensions, in 2012, Intel firstly started using a 3D tri-gate

structure or FinFET to achieve a more effective gating and meanwhile allowing for further scaling

the node size down to ~ 7nm. Up to now, a most recent result announced by IBM in June, 2017

unveiled 5nm node with a gate-all-around FET (GAAFET) structure (see Fig. 1.5c), which is able

to squeeze ~3×1010 transistors onto a 50 mm2 sized chip!

From above, state-of-the-art technologies have been developed in the semiconductor

industries. Now the question is: what is the next? Is there still any room to move even further?

People believing in Moore’s law will generally answer: yes. But the answer is not straightforward,

because a continuing dimension scaling will eventually come to an end due to the physical limit

3

(see Fig. 1.6a). For example, in each 5 nm node, if counting the lateral dimension in Si atoms, it is

only a ~140 atom array. The device miniaturization of VLSI is now counting a handful number of

atoms. As revealed above, a useful strategy of tackling the scaling difficulty the semiconductor

industries have applied before is to introduce new materials, for example, straining Si to obtain

higher mobility channel, and substituting SiO2 with high-k dielectric to improve the gate

capacitance and suppress the gate leakage. Using the same strategy, the substitution of Si with high-

mobility semiconductors like Germanium (Ge) and III-V compound semiconductors is a promising

route to further push the development of VLSI. For example, the relatively low p-MOS Si channel

can be replaced with Ge so that the p-MOS channel mobility can pair up with that of the n-MOS

(see Fig. 1.6b). Part of this thesis will focus on Ge as the channel material of semiconductor devices,

as will be discussed later.

4

Table 1.1. Scaling MOSFET device and circuit parameters with keep a constant electric field. This table is

mainly reproduced from Ref. [1].

MOSFET or Circuit Parameters Multiplicative Factor (>1)

Scaling assumptions

Device dimensions (tox, L, W, xj) 1/

Device area(A) 1/ 2

Doping concentration (Na, Nd)

Voltage (V) 1/

Derived scaling behavior of device parameters

Electric filed () 1

carrier velocity (v) 1

Depletion width (Wd) 1/

Capacitance (C=A/tox) 1/

Inversion charge density (Qi) 1

Drift current (I) 1/

Channel resistance (Rch)

Gate leakage [JG~exp(-∙tox)]

1

Power of

Derived scaling behavior of circuit parameters

Circuit delay time (~CV/I) 1/

Power dissipation per circuit (P~VI) 1/2

Power-delay product per circuit (P) 1/3

Circuit density (~1/A) 1/2

Power density (P/A)

Circuit speed (1/)

Calculations per chip (~1/A)

1

3

5

Figure 1.1. Structural schematics for (a) NPN bipolar transistor, (b) p-MOSEFT and (c) CMOS. Figure (a) is

taken from Ref. [2].

Figure 1.2. A brief chronology of the major milestones in the semiconductor industry. The data is mainly

from Ref. [1].

6

Figure 1.3. Moore’s law from 1970 to 2017. The VLSI standard requires >104 transistor/chip, which occurred

after ~1977. The background figure (year ≤ 2015) is taken from Ref. [3], while the red star is the 5nm node

result announced by IBM in June, 2017. [4]

Figure 1.4. (a) Technology node size for semiconductor manufacturing processes (1971-2017). [4] The red

star is the 5 nm node announced by IBM in June, 2017.[4] (b) Gate leakage current increase as a result of

gate oxide (SiO2) thickness (tox) downscaling (36-15 nm). [5] (c) Drain/source leakage current at off state

increase as result of transistor physical length downscaling. [6]

7

Figure 1.5. The structures of (a) high-k/metal gate planar MOSEFT, [7] (b) tri-gate FET, [8] and (c) gate-all-

around FET. [4]

Figure 1.6. (a) The miniaturization of transistors in VLSI is gradually approaching the physical limit. [9] (b)

Replace the low mobility Si channel in P-MOS with a high mobility semiconductor of Ge.

8

1.2 Advantages and Challenges of Germanium Transistors

Elemental Germanium (32Ge) by its nature forms a diamond-like cubic structure with a

lattice constant of 5.858 Å (see Fig. 1.7a). Some other basic parameters of single crystal Ge are

shown in Tab. 1.2 and compared to Si. It can been that Ge does not have a lot in common with Si,

except for the same crystal structure, the same in-direct type of band gap, and close electron

affinities (). The most important advantage of Ge over Si is its high mobility for both electron and

holes; particularly, Ge has the highest hole mobility among all group IV and III-V compound

semiconductors, ≤1900 cm2/V/s. Fig. 1.8 shows the mobility of electron, and the band gap for

various semiconductors. While III-V compounds are like “king” of semiconductors in terms of the

electron mobility, with values way higher than group IV semiconductors, Ge is outstanding for its

hole mobility with a moderate band gap EG. This high mobility allows for higher saturated drift

current of MOSEFTs which is proportional to the carrier mobility, namely

drift ox GS th carrierI C V V . Also, the smaller band gap of Ge (0.661 eV) compared to that of

Si (1.12 eV) is advantageous in terms of scaling the power voltage (VDD) in CMOS structures,

which is beneficial for the power saving. Another important advantage of Ge is its lower melting

point than Si, which results in a low temperature dopant activation and therefore allows for shallow

junction formation. [10]

However, replacing Si with Ge is not a simple task because of a few material and technical

challenges. Firstly, material-wise, the supply of large scale Ge wafer (200, 300 mm) is insufficient

to cover the market needs of the whole semiconductor industries. As a consequence, the

manufacturing cost at the very first stage is going to be high, namely Ge is too expensive to afford

compared to Si. This issue may be addressed by using thin films of Ge only. Then there comes

another roadblock, the large lattice mismatch between Ge and Si (~4 %, see Tab. 1.2), which makes

the integration of Ge films on Si substrate technically challenging. Second, the smaller band gap of

9

Ge leads to a larger leakage current between the drain and source at the stand-by state, due to a

lower tunneling barrier height in-between. Third, the larger dielectric constant of Ge enhances the

unwanted coupling between the drain and source, which consequently results in a more prominent

short channel effect, namely a loss of turning off the channel by gating. Last but maybe the most

important, the physicochemical properties of Ge surface are distinct from that of Si, which will be

discussed in details as follows.

When the world’s first bipolar transistor was invented in 1947, it was made of Ge. But not

for long, Ge was proven not suitable for the new technologies of MOSEFT and CMOS, because

the low quality of native oxide GeO2 was not able to effectively passivate the surface states of Ge

substrate, namely making the surface states inactive in the electrical characteristics of the devices.

Fortunately, people found a thankful alternative, Si, whose native oxide SiO2 showed a far better

behavior than GeO2. Using this combination of SiO2/Si, the semiconductor industries had been

prosperous for more than 50 years. Until now, people start looking back at Ge again for its high

mobility. There is a reason why GeO2/Ge did work in the early VLSI era, which is the

thermodynamically unstable nature of the GeO2/Ge interface. Previous reports have pointed out

that right after an ozone annealing, the formed GeO2/Ge interface yielded superior electrical

characteristics. [11] However, in an oxygen-free environment at elevated temperatures of ≥ 400 °C

(for example, dopant activation process), the following reaction occurs at the interface during the

fabrication of Ge MOSFETs (also see Fig. 1.9a): [12–15]

Ge(s) + GeO2(s) 2GeO(g). Eq. 1.1

This unstable nature can be understood as a result of the relative weak bonds of Ge-Ge and Ge-O

(see Tab. 1.2), which is not an issue for the stable SiO2/Si interface. As will be discussed later in

this thesis, the volatile product of GeO will introduce electrically active defects to both the interface

and the deep dielectric bulk.

10

From the standpoint of the device dimension downscaling, GeO2 is not a favorable gate

dielectric, due to its low dielectric constant (k~4.5). [16] However, a direct deposition of high-k

dielectrics on Ge has resulted in poor electrical characteristics. [17–19] For example, the growth of

HfO2 by atomic layer deposition (ALD) on a hydrogen (H) terminated Ge surface showed a long

inhibition during the initial growth, and also resulted in surface roughening. [20] The electrical

characterization revealed that MOS capacitors (MOSCap) using HfO2/Ge yielded a high defect

density both at the interface and the dielectric, [19] as well as a high gate leakage current. [14] The

mechanism behind this is also related to the interface reaction of high-k/Ge and also involves the

GeO formation and diffusion (see Fig. 1.9b). [10] Since an integration of high-k dielectric in Ge

MOSFETs is inevitable due to the stringent scaling requirement, the task then becomes how to

stabilize the interface properties.

The pioneers of Ge MOSFETs fabrication came up with an idea of using a dielectric

interlayer between high-k and Ge (see Fig. 1.9c). This interlayer should satisfy the following

requirements: (i) the formed interlayer/Ge has a high quality; (ii) the interlayer/Ge interface has to

be thermodynamically stable at elevated temperatures; (iii) the interlayer shows a good film

morphology with a low interface and film roughness; (iv) a moderate band gap and band alignment

to Ge substrate; (v) a downscaling of the interlayer thickness does not result in significant

degradation of interface properties (usually interlayer material is low-k dielectric, so needs

downscaling to improve the device capacitance). Various materials have been used as the interlayer

between high-k and Ge substrate, including GeON, [21,22] GeAlON, [23–25] GeAlO, [24,26,27]

GePO, [25] GeZrO, [14] and even GeN. [11,28] It can be seen that the principle of the materials

choosing for interlayer is incorporating other elements (metal or nitrogen) into GeO2 for partial

substitution (complete substituting O with N in GeN). Among all the interlayers, Al incorporation

into GeO2 (GeAlO) has proven most successful. For example, Takagi group has utilized a room

temperature plasma post oxidation (PPO) on ALD grown Al2O3/Ge structure to form a GeAlO/Ge

11

interface. The as-formed interface shows a state-of-the-art interface quality an atomically flatness

and sharpness (see Fig. 1.9d), as well as superior electrical characteristics with a high capacitance

density, low defect state density, and low gate leakage current. [26,29]

Material-wise, the gate dielectrics used in Ge MOSFET are mainly grown by ALD because

of the high quality films grown by ALD can meet the stringent demand for the device scaling (more

details about ALD mechanism will be discussed in Chapter 2). [30,31] During the fabrication of

Ge MOSFET, the ALD process optimization is carried out mainly through a feedback loop between

the process and post-process analysis, while the process itself remains as a black box, unavailable

for a direct detection (see Fig. 1.10). The back-and-forth in this loop is simple for most of the time,

because the researchers only need to list all the process parameters and make a so-called “design

of experiments” or DOE so as to map out all the possible combinations and thereafter pick out the

one yielding the best results. However, when dealing with multiple parameters that are entangled

with each other, the roadmap to a highly scaled and electrically reliable Ge devices is not

straightforward due to a lack of in-situ probing metrology and atomic scale mechanism of the

process. Particularly in the process of high-k ALD on Ge, the growth kinetics play an important

role in determining the dielectric properties and the interface characteristics, and therefore an

understanding of the kinetics becomes crucial for improving the device performance.

In this thesis, an in-situ spectroscopic ellipsometry (SE) was used to real-time monitor the

fabrication process of Ge MOSCaps, in conjunction with ex-situ metrologies and a ReaxFF reactive

force field simulation (in collaboration with Prof. Adri van Duin group at Penn State University)

of the ALD growth kinetics. The objectives are as follows:

1) Investigate and optimize a cleaning process to in-situ prepare a pristine Ge

surface, free of oxygen and atomically flat; a pristine starting surface can exclude

the unwanted inference from the extrinsic factors;

2) Investigate the nucleation mechanism of ALD on Ge surface;

12

3) Optimize the gate dielectric stack structure on Ge and evaluate the electrical

consequences.

13

Table 1.2. Basics parameters of single crystal Ge and Si. Most of the parameters are taken from Ref. [32].

Property Si Ge

Lattice constant (Å) 5.431 5.658

Atomic density (cm-3) 5.01E+22 4.41E+22

Dielectric constant k 11.7 16

Bond length (Å) 2.3 2.41

Bond strength (kJ/mol) 222 188

Band gap EG (eV) 1.124 0.661

Direct band gap (eV) 4.200 0.800

Conduction band maximum EC or electron affinity χ (eV) 4.05 4.00

Valence band maximum EV (eV) 5.17 4.66

Intrinsic carrier concentration ni (cm-3) 1.0E+10 2.0E+13

Effective density of states (cm-3) Conduction band NC 2.9E+19 1.0E+19

Valence band NV 3.1E+19 5.0E+18

Mobility (cm2/V/s) Electron e ≤1400 ≤3900

Holeh ≤450 ≤1900

Thermal velocity vt (cm/s) Electron vt,e 2.30E+05 3.30E+05

Hole vt,h 1.65E+05 1.90E+05

Electron effective mass me (m0) Longitudinal mle 0.98 1.6

Transverse mte 0.19 0.08

Hole effective mass mh (m0) Light hole mlh 0.16 0.043

Heavy hole mhh 0.49 0.33

Melting point Tm (°C) 1415 938.2

Native oxide

Formula SiO2 GeO2

Formation energy (kJ/mol) [33]

−910.86 -579.90

Dielectric constant k [16] 3.9 4.5

14

Figure 1.7. (a) Unit cell of Ge crystal structure. [34] (b) Band structure of Ge at room temperature. [32]

Figure 1.8. (a) Electron mobility e and (b) hole mobility h for various semiconductors as functions of band

gap EG. [32]

15

Figure 1.9. (a) Interface reaction between GeO2 and Ge substrate. (b) Interfacial reaction between HfO2 and

Ge substrate. (c) A schematic for the design of using interlayer between high-k and Ge to form stable

dielectric/Ge interface. (d) TEM images for the interfaces after plasma post oxidation on Al2O3/Ge at 300 C

and room temperature. Figure (d) is copied from Ref. [29].

16

Figure 1.10. Process optimization for integrating high-k dielectrics on Ge surface. Compared to the traditional

black box mode, an application of in-situ metrology in conjunction with atomic scale simulations provides a

more straightforward route to process optimization. [35]

17

1.3 Properties of 2D Layered Material 1T-TaS2

The recently developed exfoliation method has re-triggered the research interests on mono-

layer 2-dimensional layered materials (2DLM) or van der Waals (vdW) materials. [36] 2DLMs

have demonstrated diverse fundamental physics, ranging from conductors (e.g. graphene [37]),

semiconductors (e.g. transition metal dichalcogenide (TMD) [38]), wide band-gap insulators (e.g.

hexagonal Boron Nitride (hBN) [39]), and superconductors (e.g. Niobium Selenide (NbSe2) [40]).

Among all, layered Tantalum disulfide (1T-TaS2) has been particularly attractive due to its unique

and strong electron-correlation which results in a metal-to-insulator transition (MIT). As shown in

Fig.1.11a, an abrupt change of resistivity (~10×) occurs across the MIT transition (Tc≈ 200 K). [41]

It has been proven that this phase transition can be tuned not only by temperature, but also

pressure [42], chemical doping [43], and disorders by electron irradiation. [44] A more important

implication is that a reversible electric-field control of this correlation phenomenon makes possible

a new switch type, Landau switch. In traditional Boltzmann switch using group IV and III-V

compound semiconductor as the channel (see Fig. 1.11b), there is a fundamental physical limit,

which is the switching slope > ln(10)kBT/q= 60 mV/dec. The abrupt transport change during the

MIT transition potentially provides a solution to this physical limitation, with a much faster

switching swing (see Fig. 1.11c). [45] This switching mechanism is plausible in layered 1T-TaS2

because: (i) the electron-electron correlation is stronger than the electron-lattice interaction

(Ueff/W=1.5>1.3) [46], (ii) the nanoscale thickness of the materials makes possible a field-effect

control even in correlated systems, whereas the high electron density (~1021 cm-3) suppresses the

screening length down to ~nm. [45]

TaS2 has a few polymorphs, including 1T-, 2H-, 3R- and 6R-TaS2, all of which are vdW

materials. [47] The prefix for each structure represents a short description of the structure; taking

1T-TaS2 for example, ‘1’ means the stacking period along c-direction is one TaS2 slab, and ‘T’

18

represents ‘trigonal’; in other cases, ‘H’ is ‘hexagonal’, and ‘R’ is ‘rhombohedral’. Among all these

polymorphs, only 1T-TaS2 shows a MIT transition, and therefore becomes the material of interest

in our research. The parental crystal structure of 1T-TaS2 is a Cd(OH)2 type (space group= 𝑃3̅𝑚1)

at elevated temperatures (T> 850 C), as shown in Fig. 1.12a; in the slab plane, each Ta (green) is

six-fold coordinated by S atoms (yellow), equally locating on the top of and underneath Ta sub-

layer, while each S coordinates to three Ta atoms; in-between the slabs, there is no strong bonding,

or nearly van der Waals bonding. The low temperature phases are basically a distortion of this

parental structure. At Mott insulator state (T< TC≈ 200 K), a commensurate charge density wave

(CCDW) is developed. [42] To accommodate this charge localization, the lattice reconstructs by

forming a David-star pattern, with 12 Ta atoms at the star corners displaced toward the center Ta

(Fig. 1.12b), as revealed by previous scan tunneling microscopy (STM) (Fig. 1.13a) [48]; as a

consequence, a √13𝑎×√13𝑎×13𝑐 supercell is derived. [49] At medium temperatures (TC< T< 350

K), the CDW phase is partially melted, whereas the David-star pattern is sustained locally in a

domain form (Fig. 1.13b), i.e. a nearly commensurate charge density wave (NCDW). The gray

region developed in-between domains (Fig. 1.13c) forms conducting channels for electrons, and

therefore results in a significant reduction of resistivity, as mentioned above (Fig. 1a). At even

higher temperatures (T> 350 C), the David-star pattern completely disappears (Fig. 1.13c), and is

replaced by an incommensurate charge density wave (ICDW), leaving a pure metallic phase.

The above-mentioned MIT transition in 1T-TaS2 originates from the strong electron-

electron correlation. The system Hamiltonian is typically described by a Hubbard model [42]:

, , ,

. .i j ii ii j i i

H t c c h c U n n n

. Eq. 1.2

i jc c

is the creation (annihilation) operator of single electron on ith site with spin,

i i in c c

is the corresponding electron number operator, t is the electron wavefunction overlap

or hopping integral between the neighboring sites, U is the Coulombic interaction between

19

electrons, and is the chemical potential. Tight-binding calculation shows that at CCDW phase,

the David-star reconstruction leads to a gapping of the twelve Ta3d electrons from the Ta atoms at

the David-star outer-shell (Fig. 1.14d), i.e. B and C sites in Fig. 1.12b; the thirteenth electron from

the center Ta atom (a site) form a narrow band (band with W≈80 meV) around EF. [50] This narrow-

band feature is typical in d orbitals of transition metal, whereas the neighboring hopping t is small

in Eq. 1.2; when the hopping term t is comparable to the Coulombic interaction U, a MIT transition

may occur. [51] The physical picture of Mott insulator state can be understood as Fig. 1.15. [45]

Without the electron-electron interaction U, the narrow band is half-filled, forming a metallic state

(Fig. 1.15a). The electron-electron interaction U breaks the degeneracy in the narrow band, and

therefore split it into two sub-bands, upper and lower Hubbard bands (UHB and LHB), as shown

in Fig. 1.15b; with LHB fully occupied, the system undergoes a Mott-Hubbard transition, leading

to a high resistivity. A transition between the metallic and insulating states requires a tuning of U/t

ratio [51], by either tuning U or t alone or both. [42] For example, an elevated temperature results

in more free electrons by thermal excitation, and results in a screening of the Coulombic interaction

U [52], i.e. a reduction of effective U. So beyond certain temperature, the UHB bottom would meet

with LHB top, and they merge into one band as a whole (Fig. 1.15c). The criterion for melting Mott

insulating phase is also pointed out by Mott [52]:

1/3 0.25c Hn a . Eq. 1.3

nc is the critical free electron density, and aH is an effective Bohr radius of the isolated (localized-

electron) center. Bohr radius aH is usually described as [52,53]:

2

* 2

4Ha

m e

. Eq. 1.4

is the background dielectric constant, and m* is the effective mass of electrons in the conduction

band.

20

As mentioned above, in order to manipulate the MIT transition in 1T-TaS2 by electric field,

a charge density control of the material is necessary. Due to the lack of information for effective

mass of 1T-TaS2, the Bohr radius is not calculated using Eq. 1.4; instead, we can estimate using the

Shannon radius for Ta4+, ~7.5Å. With this approximation, the critical free electron density is

19 33.7 10cn cm . Actually, this is fairly close to a previous Hall measurement of bulk 1T-TaS2,

whereas the carrier concentration right before the MIT transition (CCDW phase at T≤200K) is

19 3~ 4 10 cm . In the same paper, the Hall measurement also showed that right after the MIT

transition (T≥200K), the carrier concentration at NCCDW phase skyrockets to21 3~ 5 10 cm . So

to transit from NCCDW to CCDW phase, one needs to deplete a charge density of

21 19 21 35 10 4 10 5 10 cm in 1T-TaS2. Assuming the carrier concentration of monolayer 1T-

TaS2 is similar to bulk (this may be a bad assumption, considering the strong thickness dependence

of the resistivity [54]), the sheet charge density to be depleted is

21 3 15 2~ 5 10 5.9A 3 10cm cm . This enormous amount of charge depletion is very

challenging if using traditional metal-oxide-semiconductor (MOS) structure; a typical achievable

capacitance density of traditional MOS capacitor is ~3C/cm2, so to deplete15 23 10 cm sheet

charge, a voltage of ~160V is required, which is not applicable in the modern transistor

technologies. A possible solution to this challenge is to use ionic field-effect mechanism, whereas

the high charge density in the electrolyte increases the charge controllability by about two orders

of magnitude, compared to the traditional oxide-based capacitor; [45] a successful resistivity

control up to ~50% at room temperature in 1T-TaS2 has been achieved by such technique (Fig.

1.16). [54] The problem with this method resides in the scalability limitation and low switching

speed due to the solid electrolyte used as the capacitor material. Alternatively, another route of

tuning MIT is to heavily dope the channel material (typically 10-30% or even more [55]) such that

the carrier concentration is close to the boundary of phase transition, and then to use a ferroelectric

21

(FE) capacitor material to gate the channel; note that a typical achievable capacitance density of

FE material like Pb(Zr,Ti)O3 (PZT) is about 30C/cm2, one order of magnitude higher than

traditional gate oxide. This strategy has been successful in La1-xSrxMnO3/La1-

xCaxMnO3/BaTiO3/La1-xSrxMnO3 hetero-junctions (Fig. 1.17a-c), [56] whereas La1-xCaxMnO3 is

the functional MIT material whose conductance can be tuned by electric gating. [57] Analogous to

this, a ferroelectric field effect transistor (FeFET) based on doped 1T-TaS2 is plausible: (i) its MIT

transition is susceptible to chemical doping, and can be even strongly suppressed (e.g. Se-doped

1T-TaS2 in Fig. 1.18a [58]); (ii) a simple atomic-layer-deposition (ALD) method can be applied in

the integration of ferroelectric materials (e.g. Si FeFET in Fig. 1.18b [59]), so that the fabrication

process of FeFET should potentially be more compatible with Si-CMOS technologies. Based on

these footstones, a proposed design of 1T-TaS2 based FeFET is shown in Fig. 1.18c. This is

basically a back-gated thin film transistor (TFT), with the conductive substrate acting as the back-

gate, ferroelectric layer as the capacitor material, appropriately doped 1T-TaS2 as the channel, top

capping layer for the surface protection, and top electrodes of matched work function as the

source/drain (S/D). More details will be discussed below.

22

Figure 1.11. (a) Resistivity as a function of temperature for bulk 1T-TaS2, reproduced from Ref. [41].

Schematic views for device switching performance in (b) Boltzmann and (c) Landau switch, respectively.

Figure 1.12. (a) Crystal structure of 1T-TaS2 at T> 850 C. The structural parameters are from Ref. [47]. (b)

Crystal reconstruction in CCDW phase (T< TC); the dash line indicates the supercell by √13𝑎×√13𝑎

reconstruction; the arrow indicates the displacement of Ta atoms with respected to the un-reconstructed phase

at T> 850 C; the corresponding shrinkage for A-B and B-C distances is 6.4 and 3.2%, respectively. [60]

23

Figure 1.13. STM images for (a) CCDW [48], and (b) NCCDW [61] phase of 1T-TaS2, respectively. (c) A

schematic view for phase transition from commensuratenearly commensurateincommensurate CDW

phase of 1T-TaS2 with increasing temperature; the dark region represents David-star reconstruction, while

the reconstruction in the gray region is considerably suppressed or removed; the red arrow represents the

current flow in NCCDW. [42]

24

Figure 1.14. Ta3d orbital splitting by the crystal field. [50] (b) The Brillouin zone (BZ) of 1T-TaS2 at CCDW

phase (T<TC); the larger BZ (solid) corresponds to the un-reconstructed phase (T>850C), while the smaller

BZ (dash) is for the supercell after reconstruction at CCDW phase. [60] (c) Band structure of CCDW phase

with considering spin-orbital coupling for one monolayer 1T-TaS2; the band gap EG≈0.2eV. [50] (d) Density

of state (DOS) contribution from a, b, and c sites of Ta in David-star (Fig. 2b) at CCDW phase. [50] (e)

Density of state (DOS) contribution from different Ta3d-orbitals at CCDW phase. [50]

25

Figure 1.15. Schematic views for (a) metallic narrow band (U=0), (b) Mott insulator (U>W, at low

temperature), and (c) Mott insulator melted by thermally excited free electrons, respectively. EF is the Fermi

level, U is the Coulombic interaction between electrons, and W is the band width.

Figure 1.16. (a) Schematic structure of ionic field effect transistor (iFET) made of 10nm 1T-TaS2. [54] (b)

Resistance R vs gate voltage VG at T= 325 K of the 1T-TaS2 iFET in (a).Results are from Ref. [54].

26

Figure 1.17. (a) Schematic demonstration of charge control in La1-xSrxMnO3/La1-xCaxMnO3/BaTiO3/La1-

xSrxMnO3 tunneling hetero-junction. (b) Pulsed voltage control of the resistivity of the hetero-junction in (a).

The response of the hetero-junction in (a) to the time-dependent control voltage. Figures (a)-(c) are from

Ref. [56]. (d) The phase diagram for La1-xCaxMnO3 system. The figure (d) is from Ref. [57].

Figure 1.18. (a) Resistivity vs temperature for Se-doped 1T-TaS2. [58] (b) ID-VG characteristics of Hf1-

xSixO2-FeFET on n-Si with 500nm gate length. [59] (c) A proposed design of 1T-TaS2 FeFET.

27

1.4 A Brief Review of Transition Metal Dichalcogenide Synthesis

Various methods have been developed to growth 2-D transition metal dichalcogenides

(TMD), and mostly they are vapor phase deposition processes. [62] Fig. 1.19a shows the growth

setup for chemical vapor transport (CVT). This method is particularly useful when the precursor is

low vapor pressure solid powder. Utilizing a particular carrier gas called “transport agent”, the solid

powder becomes more volatile, like dissolving in the agent, and then the flow of the agent carries

the precursor vapor from a high temperature zone to cold zone, where the decreased temperature

results in the cold condensation of the precursor and then crystallizes. So this growth mechanism

is essentially just “transport”, without real chemical reactions. CVT has been used for growing

large size single crystals of TMD, but it is not suitable for growing thin films.

Chemical vapor deposition (CVD) has been widely used in semiconductor industries to

grow thin films, and turns out also an effective method for growing ultrathin layers of TMD. Fig.

1.19b shows a typical setup for CVD growth of TMD. The low melting point precursor A (S/Se

powder is placed in the upstream of the inert carrier gas flow, which is kept at a relatively low

temperature but sufficient to generate enough vapor pressure A. The low vapor pressure precursor

B (MoO3) is placed in the midstream which is at a high temperature for effectively vaporizing B.

In the downstream of the flow, the two precursors mix and reacts with each other, producing

deposition on the substrate sample. This method can yield relatively large crystal size (~×10 m)

of TMD, [63] and sometimes can even grow wafer-scale thin films. [64]

Another interesting synthesis approach is called “van der Waals epitaxy” or vdWE. While

the mass transport can be realized only by CVD but also physical vapor deposition (PVD) like

magnetron sputtering, the “epitaxy” in this method is also slightly different from the traditional

understanding of epitaxial growth, like molecular beam epitaxy (MBE). [62] While a single crystal

substrate is also required in vdWE to act as a template for epitaxial growth of TMD, the requirement

28

of lattice mismatch is much more tolerant due to the weak van der Waals coupling between the film

and substrate. [65] This approach allows for a directional growth of TMD materials, namely the as

grown TMD flakes show a highly oriented crystal alignment.

Solution phase synthesis has also been used to grow TMD nanocrystals. For example, a

solvothermal method, which is similar to the hydrothermal synthesis, has been used to grow InSe

(see Fig. 1.19d). [66] In this method, a N2 gas was used as the reaction solvent for precursors at

elevated temperatures. The high pressure and high temperature environment inside the reactor

facilitates the reaction process. This approach is suitable for producing nanowires and nanosheets,

but not appropriate for growing ultrathin layers of TMD due to the lack of thickness control.

Atomic layer deposition (ALD) is well-known for precise control of deposition thickness,

so can be used for growing ultrathin layers of TMD materials. Fig. 1.20a shows the ALD growth

for MoS2 using MoCl5 and H2S as precursors. The self-limiting nature of ALD growth allows for a

precise thickness control from one to multiple monolayers by counting the growth cycles. More

importantly, the high conformity of ALD growth provides a large scale deposition for 2DLM

materials. However, one issue with ALD growth of TMD is that the resultant thin films are usually

amorphous due to the relatively low growth temperature, so a post-deposition annealing at high

temperature is necessary to crystallize the grown films (see Fig. 1.20b).

As for preparing ultrathin layers of 1T-TaS2, the primary method used by researchers is

still mechanical exfoliation using Scotch tape from a single crystal bulk, [54,67–70] which is grown

by a CVT method. [67,71–73] This method is simple and the as obtained flakes (< 20 m) show a

low density of defects since they are extracted directly from the fresh bulk 1T-TaS2 surface. But

the problem is that one can barely control the thickness and the shape of the flakes, which is not

favorable for the application in device manufacturing. The thickness control of 1T-TaS2 is

particularly important, because previous results have revealed that its electrical properties is

sensitive to the thickness, with a critical thickness ~ 10 nm, corresponding to ~13 monolayer, which

29

coincides with the superlattice dimension along c-axis, namely √13𝑎×√13𝑎×13𝑐 supercell at the

CDW phase. [49] On the other hand, CVD method has been successfully used to grow 1T-TaS2

flakes (~ 10 m) on SiO2/Si substrates with controlled thickness (~ nm resolution), using TaCl5 and

S as the precursors in conjunction with a H2(10%)/Ar carrier gas (see Fig. 1.21a). [74] The reaction

of growth kinetics is supposed to be as:

5/2H2(g) + 2S(g) + TaCl5(g) TaS2(s) + 5HCl(g). Eq. 1.5

The as-grown flake crystal is ~10 mm in lateral size with a 1T-phase structure. This result is

encouraging, but the TMD material coverage on the substrate is still too low to become a continuous

thin film.

Combining all the previous results, we propose in this thesis that wafer scale 1T-TaS2 thin

films can be grown using ALD approach using TaCl5 and H2S as the precursors. The reaction

mechanism is similar to that of ALD growth of MoS2 using MoCl5 and H2S, namely as follows:

2H2S(g) + TaCl5(g) TaS2(s) + 4HCl(g) + 1/2Cl2(g). Eq. 1.6

Also, regarding the similarity between Eq. 1.6 and 1.5, it is reasonable to believe that this ALD

approach of growing 1T-TaS2 is plausible. The objectives of the research on 1T-TaS2 in this thesis

are:

1) Design and build an ALD system with the following features: (i) compatible with the

hazardous H2S precursor; (ii) a capability of in-situ monitoring mechanism; (iii) a high

temperature capability of the substrate heater.

2) Develop and optimize a process of 1T-TaS2 ALD growth using TaCl5 and H2S as

precursors; investigate the nucleation mechanism.

3) Develop and optimize a process of ferroelectric HfO2 using ALD growth, which is

intended for the application of gate dielectrics in the fabrication of ferroelectric field

effect transistor (FeFET) with 1T-TaS2 as the channel.

30

Figure 1.19. Schematics of growth setup for (a) CVT, [75,76] (b) CVD, [63] (c) van der Waals epitaxy, [77]

and (d) solvothermal method. [66]

31

Figure 1.20. (a) Schematic illustration of one growth cycle of an ALD MoS2 film on c-sapphire. (b) SEM

and AFM images for MoS2 after 800C anneal. The results are from Ref. [78].

32

Figure 1.21. Controlled synthesis of ultrathin 1T-TaS2 crystals via a CVD method. (a) Schematic of CVD

setup for the growth of 1T-TaS2 on a SiO2/Si substrate with tantalum pentachloride powder and sulfur powder

used as the precursors. (b, c) Crystal structure of distorted 1T-TaS2 on a SiO2/Si substrate. (d, e) Optical

images of 1T-TaS2 ultrathin flakes. (f) The controlled thicknesses of 1T-TaS2 at different growth times. The

inset shows an octahedral arrangement of the central Ta atom coordinated with S atoms. (g−l) AFM images

and their corresponding height profiles of various 1T-TaS2 at different growing times. The scale bars are 10

μm in parts d and e and 2 μm in parts g, h, and i. These results are from Ref. [74].

33

1.5 Thesis Organization

This thesis is organized in the following chapters involving experimental techniques used

in this thesis, ALD of high-k dielectrics on Ge, ALD synthesis of 1T-TaS2 and ALD synthesis of

ferroelectric HfO2.

Chapter 2 Various key experimental techniques will be discussed in details,

including the working principle, hardware information, and data analysis. The techniques that will

be discussed are: atomic layer deposition (ALD), spectroscopic ellipsometry (SE), admittance

characterization, and X-ray photoelectron spectroscopy (XPS). Other experimental techniques like

atomic force microscopy (AFM) and Raman spectroscopy will not be discussed in details, but the

experimental parameters will be included when the corresponding data are discussed.

Chapter 3 In this chapter, a cleaning mechanism of removing native GeOx on Ge by

H2 plasma will be firstly addressed in details. It will be shown that the process temperature plays

an important role to obtain pristine Ge surfaces. Then, the nucleation mechanism of Al2O3 on two

different Ge surfaces (H-terminated and oxidized) are studied by in-situ spectroscopic ellipsometry

and ex-situ characterizations. The experimental results are compared with the ReaxFF simulations

from the collaborators (Prof. Adri van Duin group at Penn State University) to establish a structure-

property relationship so as to figure out the atomic scale mechanism. Finally, with the

understanding of the ALD mechanism on Ge, a tri-layer gate dielectric stack is proposed and

optimized to form Ge MOSCap devices. The dielectric/Ge interface qualities are evaluated by using

admittance characterizations.

Chapter 2

Experimental Techniques

This chapter will discuss about the principles of the primary experimental techniques used

in the research of this thesis, including atomic layer deposition, spectroscopic ellipsometry,

admittance measurement, and X-ray photoelectron spectroscopy. Other experimental techniques

are also used, but will only be introduced together the corresponding results.

2.1 Atomic Layer Deposition

Atomic layer deposition (ALD) technology was originally discovered developed

independently with two different names: “molecular layering” in the Soviet Union during 1960s,

and much better known as “atomic layer epitaxy (ALE)” as early as 1974 in Finland. [31,79,80] A

good reading source about the history of ALD evolution is a volunteer-based effort called virtual

project on the history of ALD (VPHA) launched in 2013. Ever since the discovery of ALE, the

technique has been further optimized and matured into an advanced deposition technology that is

widely used in semiconductor industries, particularly growing high-permittivity (high-k) dielectrics

for the gate insulators in the metal-oxide-semiconductor field effect transistor (MOSFET)

structures and for copper diffusion barriers in backend interconnects. [31,81] The self-limiting

reaction mechanism enables an atomic level control of deposition thickness by ALD, which has

significantly boosted the miniaturization of the device dimensions. More importantly, the high

conformality of ALD-grown films allows the application in depositions on high-aspect ratio

structures, e.g. a 3D structure of FinFET transistors. This advantageous feature is not achievable in

other high vacuum deposition techniques like molecular beam epitaxy (MBE) and thermal/e-beam

35

evaporation, wherein the mass transport of materials is a line-of-sight mode and therefore highly

directional. Also, the relatively low temperature required for ALD growth results in continuous,

amorphous and pin-hole-free materials in a wafer scale, which are necessary for the semiconductor

devices. Last but not the least, the low temperature is beneficial to the industry manufacturing in

terms of saving energy cost.

2.1.1 Principles of Atomic Layer Deposition

ALD growth is a surface reaction process by sequentially dosing different reactants and

separating the doses with sufficient purging. Even if using the same precursors, ALD is distinct to

chemical vapor deposition (CVD), wherein all the reacting precursors are dose simultaneously and

the reaction is driven by the high temperature near the surface in a gas phase mode. CVD is known

for fast growth, but not for a good control of thickness, surface roughness or material continuity.

The working principles of ALD is usually demonstrated by using Al2O3 growth with tri-

methyl-aluminum [Al(CH3)3, TMA] and water (H2O) precursors as an example. This reaction has

been comprehensive studied both experimentally [35,82–84] and theoretically, [35,85,86] because

Al2O3 has been extensively used as a high quality gate dielectric with a low leakage current and

relatively high permittivity, [87] as well as a high temperature and highly corrosion resistant

coating. [88] A well accepted scenario for a reaction cycle of Al2O3 ALD is shown in Fig. 2.1. A

good starting surface for Al2O3 ALD is terminated with hydroxyl groups, i.e. -OH*, where the

asterisk superscription denotes surface species. TMA molecules are transported together by inert

carrier gas (like N2, Ar) and dosed into the system and react with the –OH* groups by exchanging

one methyl group (-CH3) in TMA with the H atom in –OH*, and subsequently forming CH4

byproduct, as follows:

-OH* + Al(CH3)3 -O-Al(CH3)2* + CH4. Eq. 2.1

36

In this scenario, the reaction only occurs on the surface sites with –OH* groups, and the resultant

surface is decorated with di-methyl-aluminum (DMA) groups, namely -Al(CH3)2*. A complete

reaction requires a sufficient amount of TMA and certain time for the precursor to uniformly

distribute across the whole sample. The actual consumption of TMA is way less than the required

dosage, due to the low sticking coefficient of TMA (~10-3-10-4). [84] On the other hand, an over

dose of TMA molecules will not result in further surface reactions, because TMA is chemically

unreactive to DMA. So the chemical reaction is limited to at most one monolayer on the surface,

which is well-known as a self-limiting effect. The next step is to purge the system with inert gas to

remove the CH4 byproduct as well as the unreacted gas phase TMA molecules and any possible

physisorption of TMA on the surface, so that there will be no direct intermixing and reaction

between TMA gas and the subsequently dosed oxidant precursor H2O. The unwanted gas phase

reaction between TMA and H2O is called parasitic CVD growth, and can cause formation of Al2O3

nanoparticles above the sample surface. The nanoparticles will fall onto the surface, namely

“dusting”, and become local nucleation centers for ALD growth, so consequently the deposition

becomes non-uniform and ends up with a rough surface. A sufficient purging step in ALD growth

is of great importance to obtain high quality thin films. After completely purging TMA, the H2O

molecules are dosed to oxidize the surface DMA. H2O is highly reactive to DMA, by exchanging

its –OH with –CH3* groups in DMA and forming volatile CH4 as follows:

-O-Al(CH3)2* + 2H2O -O-Al(OH)2

* + 2CH4. Eq. 2.2

The purging step for H2O is usually longer than that for TMA, due to the stronger H2O

physisorption onto a –OH* terminated surface. Also, to insure a complete purge, an elevated

temperature is preferred (≥110 °C) to increase the desorption rate of surface H2O. From Eq. 2.1.2,

after purging H2O, the surface ends up with –OH* termination again, and is ready for the next ALD

cycle. By combining Eq. 2.1.1 and 2.1.2, the overall reaction for Al2O3 ALD is:

2Al(CH3)3 +3H2O Al2O3 + 3CH4, H= -376 kcal. [31] Eq. 2.3

37

In the real Al2O3 ALD experiments, the growth rate, defined by the growth per cycle (GPC), is

usually less than one nominal monolayer (m.l.). The nominal monolayer thickness of Al2O3 can be

estimated as follows: [31]

1/3

. .m l Ald . Eq. 2.4

Al is the atomic density of Al in the ALD grown Al2O3. In this calculation, Al atoms are assumed

to uniformly distribute in the material without any specific structures, namely completely

amorphous. Using a density of 3.0 g/cm3 reported for ALD grown Al2O3, [89] the atomic density

of Al is 3.54e22 cm-3, so the monolayer thickness is 3.0 Å. This calculation is slightly different

from Ref. [31], wherein the calculation is based on the atomic density of “Al2O3” unit rather than

“AlO1.5”. Using whichever as the monolayer thickness, each ALD cycle yields a much less

thickness than one monolayer, ranging from 0.86-1.2 Å/cycle. [31,35] The disagreement between

the monolayer thickness and GPC is usually attributed to steric hindrance of the ligands, that is, the

relatively bulky admolecule size of metal organic (MO) precursor (e.g. TMA) blocks part of the

surface sites from being adsorbed by MO molecules. [90]

A binary reaction using a combination of corresponding MO precursors and H2O is

commonly used and usually defined as a thermal ALD, but to obtain high quality materials, an

elevated temperature is required. This temperature requirement may become an issue when

performing ALD on an organic sample surface like polymer, [91] which is vulnerable to high

temperature. Also, a growth of elemental thin films (e.g. Si) is sometimes not accessible by a

thermal ALD process. [31] To address these issues, people have successfully developed a plasma

enhanced ALD process, namely PEALD, wherein the highly reactive radicals generated by the

plasma source can make possible those reactions that cannot be driven by just thermal energy. [31]

On the other hand, an application of PEALD using O2 plasma to grow dielectric oxides on

semiconductor surfaces is not very practical, because the aggressive O radicals quickly oxidizes

38

the underlying substrates to form thick native oxide, which is not preferable in the dimension

scaling for the semiconductor devices.

Figure 2.1. A schematic for the sequential process for growing Al2O3 by ALD using TMA and H2O as the

precursors.

2.1.2 Hardware Information for High-k ALD

In this thesis, the high-k dielectric ALD depositions on Ge were performed using a

commercial ALD system (ALD-150LX, Kurt J. Lesker Co.) equipped with an in-situ plasma source

and spectroscopic ellipsometer (see Fig. 2.2). The system is able to perform various plasma

treatments (Ar, H2, N2, and O2) and ALD/PEALD deposition of Al2O3 and HfO2. The integration

of showerhead structure allows to uniformly disperse the precursor/radicals across the sample. The

in-situ ellipsometry real-time monitors of all processes inside the ALD reactor chamber. The view

windows for the incident and reflected light are made of fused silica, which shows a high

transmission (>90%) within the ellipsometry spectrum range, and are installed onto the ALD

reactor chamber with annealed copper gaskets to avoid strain induced photo-elastic effect, which

creates a non-uniform phase shift to the light beam and therefore a large depolarization in the signal.

39

The process parameters used in this thesis are as follows. The substrate temperature ranges

from 110-330 °C. The process background pressure is typically ~1.2 Torr Ar gas. A carrier gas of

Ar is used to deliver all precursors and the plasma radicals. The detailed parameters for the thermal

ALD of Al2O3 and HfO2 and the plasma treatments are shown in Tab. 2.1 and 2.2, respectively.

Table 2.1. Parameters for the precursors used in thermal ALD.

Precursor Source

Temp. (°C)

Carrier Gas

Flow (sccm)

Dose

Time (sec)

Purge

Time (sec)

TMA 20 20 0.04 10

TDMAH 85 20 0.2 10

H2O 20 20 0.3 20

Table 2.2. Parameters for the plasma treatments.

Plasma Type Reactive Gas

Flow (sccm)

Carrier Gas

Flow (sccm)

Power

(Watt)

Dose

Time (sec)

O2 (pulse) 3 117 125 1.75

O2 (continuum) 3 117 125 ≥5

H2 (low power) 3 117 125 ≥5

H2 (high power) 8 112 300 ≥5

40

Figure 2.2. A schematic for the ALD system of ALD-150LX. The figure was originally plotted by Jason

Lapano from Roman Engel-Herbert group at MatSE, Penn State University.

2.1.3 Hardware Information for TMD ALD

A home-made ALD system was built to synthesize transition metal dichalcogenides. The

system was originally designed and built by Thomas Larrabee for ALD deposition of Al2O3

(TMA+H2O). [92] In order to make possible the deposition of transition metal dichalcogenides,

significant changes have been made to accommodate the low vapor pressure precursor of TaCl5,

chemically corrosive precursor of H2S, and high temperature heating capability.

The system is kept inside a ventilation enclosure to limit any possible exposure to the ALD

precursors especially toxic H2S (see Fig. 2.3). The ALD reactor is made of a 4.5” ConFlat (CF) 6-

way cube, with a lateral gas flow geometry. A showerhead with pinhole arrays of a hole diameter

41

Ø=0.06” and spacing d=0.15” was machined out of an Al metal disk (diameter Ø=2.5” and

thickness t=0.5”), mounted into a double-faced CF 4.5” flange (see Fig. 2.4), and used to uniformly

disperse the gas flow inside the ALD reactor. The heater of the ALD system was also home-made

(see Fig. 2.5). In order to minimize the heat sink effect due to the air exposure of the backside of

the bottom flange, the heater stage (0.9”×0.9”) is not in direct contact with the bottom flange but

suspended by long studs. A heater cartridge (Zoro, Ø=0.5”, 191 Watts/inch2) is used to heat the

stage. A K-type thermocouple is mounted near the heater cartridge to measure the temperature of

the sample stage. An in-situ self-heated quartz crystal microbalance (QCM) is installed to monitor

the ALD growth. An oscillator is used to find the resonant frequency which is then measured by

the frequency counter. Details about the QCM setup can be found in Ref. [92]. Note that the

pressure gauge is installed at the downstream of gas flow in order to avoid the degassing

contamination from the gauge to the ALD reactor chamber. The carrier gas flows are both 40 sccm

Ar for both dosing lines, and yield a background pressure of ~1.03 Torr.

H2S source is kept at in a lecture bottle and its flow is limited using a high-pressure mass

flow controller (MFC, 100 sccm max), due to the high vapor pressure of H2S (~252 PSI at 20 °C).

To avoid the corrosion from H2S exposure to the vacuum sealing, silver-plated gaskets are used for

CF flange sealing, and stainless steel VCR gaskets are used for VCR connections. Also, to protect

the inner wall of the ALD reactor from H2S, a thick Al2O3 coating layer (~ 1 m) has been deposited

by ALD (TMA+H2O) in the reactor chamber. Pneumatic ALD valves are used to control the dose

time of precursors with a time resolution of 10 ms. In the real ALD growth, a H2S flow rate of 1

sccm and a dose time of 15 ms are used to generate a dose pressure of ~500 mTorr.

A by-pass loop structure is used for TaCl5, so as to introduce a through flow to purge the

ALD valves to avoid clogging that may be caused by cold condensation. The TaCl5 source cylinder

is kept at 120 °C to generate a sufficient vapor pressure. The vapor pressure of TaCl5 is given

by: [93]

42

3 1ln /atm 12.710 6.4478 10 0.005P T . Eq. 2.5

At 120 °C, the vapor pressure is calculated to be 18.85±0.09 Torr, which is comparable to 17.5

Torr for H2O and 9.0 Torr for TMA at 20 °C. To avoid any other cold condensation, the dosing

lines, ALD valves, and the reactor chamber are heated to 150 °C. Note that the max allowed

temperature is 200 °C for an ALD valve body, and 148 °C for the manual valve of the TaCl5

cylinder. The dose time of TaCl5 is 0.3 sec controlled by an ALD valve and yields a dose pressure

of ~20 mTorr.

In order to control the ALD process automatically, a LabVIEW program was developed.

The program includes controls of all the ALD pneumatic valves, read-out of chamber pressure,

QCM signals, edition of deposition recipes, and growth data acquisition/saving.

Figure 2.3. A schematic of the home-made ALD system for synthesizing transition metal dichalcogenides.

43

Figure 2.4. The structure of the showerhead used in the home-made ALD system.

Figure 2.5. The structure of the heater used in the home-made ALD system.

44

2.2 Spectroscopic Ellipsometry

Spectroscopic Ellipsometry (SE) is an optical measurement of the near-surface dielectric

properties of materials. SE has been widely used in both manufacturing [94] and academic

researches. [95] SE measurement is a non-invasive, low cost, and fast characterization technique.

These advantageous features allows its application for both ex-situ and in-situ measurement in

surface-related processes. In particular, in-situ SE has been a powerful tool for real time monitoring

the deposition and etching, because of its high sensitivity to surface changes, especially film

thickness (sub-Angstrom resolution).

Early application of ellipsometry was not spectroscopic, namely monochromatic or using

only a few wavelengths in the light source to determine the optical constants of the materials. This

can result in multiple solutions when numerically de-convoluting for the material optical constants

and thin film thickness, especially for a non-standard material with unknown optical constants or

modified materials somewhat deviating from the standard properties. For example, the refractive

index n of Al2O3 grown by ALD is temperature dependent (e.g. n=1.77 at 600 nm for 270 °C, and

n=1.52 for 110 °C), while the growth rate is also sensitive to the temperature, so it become difficult

to solve for both refractive indexes and growth rates at the same time. In real applications, SE is

more advanced wherein a broad range light source is utilized to obtain a continuous spectrum, and

an optical model with well-defined dispersion relations (corresponding to the optical transitions in

the materials) is adopted to fit the spectrum. Thus, the material parameters given by SE are de-

convoluted with more constraints and more physically meaningful.

45

2.2.1 Principles of Ellipsometry

The working principles of ellipsometry is as shown in Fig. 2.3. A linearly polarized light

beam is incident onto the sample surface with an angle of ~70 °. The electric field (Ei) of the linear

polarization can be decomposed into two parts with respect to the incident plane, namely, Ep

parallel and Es perpendicular to the incident plane, respectively; the i subscription stands for

“incident”, p for “parallel”, and s stands for “senkrecht” (meaning “perpendicular” in German). A

mathematical description of the incident light can be expressed by a Jones vector as:

ip

i

is

EE

E

. Eq. 2.6

After the interaction between the incident light and the sample surface, the polarization state of the

reflected light is modified and can be elliptical, spherical, or even still linear, depending on the

optical properties of the sample surface. Using a Jones vector, the reflected light can be expressed

as:

rp

r

rs

EE

E

. Eq. 2.7

Then the reflectivity of the sample surface can be described by using Fresnel coefficients as

follows:

and rp rs

p s

ip is

E Er r

E E . Eq. 2.8

In the SE measurement, a relative reflection ratio defined by normalizing rp with respect to rs is

measured:

tan expp

s

ri

r . Eq. 2.9

Note that is a complex number, because the interaction between the polarized light and the

material surface can result in both amplitude change and phase shift. The amplitude ratio is

46

represented by tan, while the phase by exp(i). and are the two primary parameters of SE

measurement, and defined as ellipsometric angles. Since these two angles essentially describes the

interaction between the light and sample, we can use a Jones matrix to represent the sample:

tan exp 0 sin exp 0 or

0 1 0 cos

i iS

. Eq. 2.10

Fig. 2.4 shows the optics for measuring a simple homostructure sample with a flat surface

and sufficient thickness (so that no backside reflection). Usually, the medium (0) in the SE

measurements is either air or vacuum, so we can simply the measurement by approximating the

refractive index for medium (0) as n0=1, k0=0. Then from Fresnel equations, the relation between

the ellipsometric angles / and the refractive index n/k of the homostructure sample can be

expressed as follows: [96]

2 2

0 0

tan 1 4sin tan exp 2 tan exp tan exp

1 tan exp

i i i i in ik n ik

i

2 2tan 1 4sin tan exp 2 tan exp tan exp

1 tan exp

i i i i i

i

. Eq. 2.11

So we can directly determine the optical constants of the materials in homostructure samples. This

strategy works well for measuring absorbing substrate samples like Si and Ge, where the surface is

clean, atomically flat and thickness (≥200 m) is sufficient to avoid the transmission to the sample

backside and cause coherent backside reflection that will result in the depolarization of the

measurement signal (see Fig. 2.5a). For non-absorbing substrates like sapphire and SrTiO3, it is

necessary to roughen the backside to randomly scatter the transmitted light so as to eliminate the

backside reflection, e.g. using single-side polished substrates. Particularly, note in that in an

infrared (IR) light range, even substrates like Si are transparent, so the backside reflection effect is

still non-trivial (see Fig. 2.5b).

Fig. 2.6 shows the optics for measuring a heterostructure sample with a thin film overlayer

on a thick substrate. In this measurement, two interfaces are present, one between the film and

47

air/vacuum (S01), and other between the film and substrate (S12). A multi-reflection occurs between

the two surfaces, which is inside the thin film overlayer. Therefore, the overall reflection intensity

from the surface is a coherent sum of the direct reflection from the interface S01 and all the

transmitted lights from the multi-reflection at the interface S12. The effective Fresnel reflection

coefficient (both p- and s-polarizations) for the sample can be expressed as (see Ref. [97] for more

details about the equation derivation):

2

2 2 2 01 1201 01 12 10 01 12 10 10 12 2

01 121

ii i i

i

r r er r t r t e t r t e r r e

r r e

. Eq. 2.12

In this equation, r01, r12 represent the Fresnel reflection coefficients at the interfaces S01 and S12,

respectively; is the phase factor induced by one back-and-forth reflection inside the film, and

described by 2 2 2cosn ik dc

, where d is the thickness of the film overlayer. The

complexity is mainly introduced by the phase term of . For a non-absorbing (k1=0) thin film like

SiO2 at =632.8 nm, an increase of the film thickness only results a pure phase shift for the

exponential term in Eq. 2.12. As a consequence, the measured ellipsometric angles become periodic

functions of the film thickness (see Fig. 2.7a), namely forming a close loop trajectory in the /

space map (see Fig. 2.7b). But for an absorbing film, the exponential term in Eq. 2.12 will result a

non-reversible modification of the reflection coefficients, e.g. a-Si:H deposition on a single crystal

Si substrate (see Fig. 2.8). [98]

48

Figure 2.6. A schematic for the setup of the ellipsometry measurement.

Figure 2.7. Reflection and transmission of an incident light at the boundary between a homostructure sample

and air/vacuum in the SE measurement.

Figure 2.8. (a) A schematic for the backside reflection of the sample and resultant interfering signal in the

SE measurement of non-absorbing substrate. (b) Depolarization of the SE signals for single- and double-side

polished Si substrates.

49

2.2.2 Data Analysis

While analyzing the SE results for a homostructure sample can be done simply using Eq.

2.11, a direct analysis for multilayer-film samples is not straightforward because too many fitting

parameters in the structural model may results in completely random convergence into local

minimum with false outcomes. To obtain a reliable global minimum, a practical strategy is “bottom

up”, namely measure and fit layer-by-layer (see Fig. 2.9). Firstly, prepare a pristine substrate with

atomic flatness and contaminant-free. During the measurement, we need to assure that the surface

will be not contaminated again, so in-situ cleans and in-situ measurements are preferred. The

optical constants of the substrate (n0/k0) can be extracted using Eq. 2.11. Then, deposit the first film

overlayer (1), perform the SE measurement. With the optically defined substrate, we can now fit

for the optical constants (n1/k1) and thickness (d1) for the film (1). Following the same steps, we

can layer-by-layer extract the information of the film (2), (3) and so forth. The advantage of this

bottom-up strategy is that in each fitting process, there is only one unknown layer, so with fewer

parameters, the randomness of fitting can be significantly suppressed.

Various optical models are used to describe the energy dispersion of the dielectric function

(see Fig. 2.10). In general, a physical requirement for these models is the satisfactory of Kramers-

Kronig relation between the e1 and e2 in the modeling. [97] A classical description of the optical

transition is Lorentz model, wherein the electron interacts with the electric field of the light like an

oscillator around the nucleus. [97] The contribution of a single Lorentz oscillator to the dielectric

function is as follows (also see Fig. 2.11a):

01 2 2 2

0

Ei A

E E i E

. Eq. 2.13

In this equation, E is the photon energy, E0 is the resonant energy or the transition peak position,

is the peak broadening caused by damping, and A is the dimensionless amplitude. For multi Lorentz

50

oscillators, the overall dielectric function of the material can be expressed as a simple sum of all

oscillators (j=1, 2, 3 …) and the vacuum (1=1, 2=0), as follows:

1 21 j j

j

i . Eq. 2.14

For transparent materials (k, 2~0), namely the spectrum range is way lower in energy than

the transition peak positions (<<E0-E), the measured SE spectrum is the low energy tail of the

optical transitions, and can be described by a series expansion of Lorentz model as a function of

wavelength. This simplified model is called Sellmeier for describing 1 or Cauchy for n. In this

thesis, the Cauchy model is used for describing all the oxides grown by ALD, as follows:

2 4

and 0B C

n A k

. Eq. 2.15

is the wavelength number in nm. The Cauchy model is purely a mathematical approximation of

the real refractive index. Usually, a good evaluation can be achieved using up to the second order

approximation, namely only A, B, and C. Fig. 2.11b shows the Cauchy dispersion for Al2O3.

For asymmetric optical transitions caused by the cut-off effect of the band gap (Eg) of

materials, a modified Lorentz model, Tauc-Lorentz (TL), [99] is used as follows (also see Fig.

2.11c):

2

0

22 2 2 2

2

for

0 for

g

g

g

g

E E EAE E

EE E E E

E E

. Eq. 2.16

Note that the amplitude A is of an energy dimension, namely in eV. The contribution of the TL

oscillator to real part of dielectric function (1) is obtained by the Kramers-Kronig integration of

2, as follows:

2

1 2 2

2

gEE P d

E

. Eq. 2.17

P stands for the principal of the integral. The use of TL oscillators is necessary for describing the

major peaks in the dielectric functions, because the position of the band gap produces major

51

modification to the peak profile. But for minor peaks, a TL model does not significantly improve

the fitting quality compared to using a simple Lorentz model in the ultimate fitting results, because

of their low spectral weight, but just increase the complexity of fitting by increasing the number of

variables.

Other commonly used models include Drude model for describing the intraband transitions

in metallic/conducting materials at low energies, and effective medium approximation (EMA) to

describe the interface/surface roughness layer by mixing the dielectric functions of two or more

materials to form an intermediate material. These two models are not used in this thesis, so not

discussed in details. More information about them can be found in Ref. [96,97]. Also, all the

materials characterized by ellipsometry in this thesis are isotropic, so were measured by a standard

mode of ellipsometry. More details about measuring anisotropic materials can be found in

Ref. [96,97].

General steps for fitting absorbing film materials with generalized oscillators (namely

using Lorentz, TL, Drude, etc) are as follows: (1) Prepare a target film sample with a sufficient

thickness (≥ 5nm; the thicker, the better). (2) Precisely measure the film thickness with other

characterization methods, like X-ray diffraction (XRD), tunneling electron microscopy (TEM), or

atomic force microscopy (AFM), with an Angstrom-level thickness resolution. (3) Start fitting with

a B-spline model (choose a similar material as the starting n/k, and an energy step of ≤0.1 eV) to

obtain a raw dielectric function of the material; B-spline model is essentially a point-by-point

calculation of dielectric function or refractive indexes using Fresnel equations; in this calculation,

various combination of n/k and thickness can result in the same /, so a precise measurement of

the thickness becomes important in uniquely determining n/k or the dielectric function. (4)

Parameterize the imaginary part of raw dielectric function e2 with the generalized oscillators. (5)

Parameterize the real part of raw dielectric function e1; if the fitting of e2 is done properly, the mis-

fitting for e1 is usually an overall offset which can be simply corrected by fitting the high energy

52

dielectric contribution inf. (6) Globally fit for the whole sample to minimize the fitting mean square

error (MSE); an acceptable MSE range is ≤10 (corresponding to an error function of 1e-2).

Figure 2.9. Multi-reflection inside the thin film for measuring a film/substrate heterostructure by SE.

Figure 2.10. The evolution of ellipsometric angles / with respect to an increasing thickness of SiO2 thin

film (d=0~284.2 nm, n1=1.457, k1=0) on a Si substrate (n2=3.8812, k2=0.0195). The incident light is HeNe

laser with a wavelength =632.8 nm at an incident angle of 70 °. (a) /evolution as functions of the SiO2

thickness. (b) /evolution trajectory as the SiO2 thickness increases from 0 to 284.2 nm.

53

Figure 2.11. / evolution trajectory for growing of a-Si:H on a single crystal Si substrate. The refractive

index of a-Si:H is n=5-i0.85 at 500 nm. This figure is reproduces from Ref. [97].

Figure 2.12. Schematic for the bottom-up strategy of characterizing multi-layer structure by SE.

Figure 2.13. Dielectric function models used in ellipsometry data analysis. Figure is copied from Ref. [97].

HOA stands for harmonic oscillator approximation, and MDF for model dielectric function.

54

Figure 2.14. Examples for the dispersion of (a) Lorentz, (b) Cauchy, and (c) Tauc-Lorentz models used in

the ellipsometry data analysis.

2.2.3 Material Parameterization

As will be discussed later, T=270 °C is used as the primary process temperature of high-k

ALD on Ge surface, so the ellipsometry characterization is performed mainly for materials at this

temperature. Tab. 2.3 shows the numerical deconvolution of a clean Ge substrate at T=270 °C using

generalized oscillators (GenOsc). The fitted results are shown in Fig. 2.16, as well as the extracted

dielectric function of Ge. All the amorphous oxide films (GeOx/GeO2, Al2O3, and HfO2) grown in

our processes have high band gaps beyond the spectral range (1.24~5.17 eV), so a Cauchy

dispersion was used to describe their optical constants, as shown in Table 2.4. Note that all the

fitting models used for the oxide films are extracted from relatively thick films (>5 nm) so as to

minimize the error of fitting.

55

Table 2.3. The dielectric function of Ge substrate at T=270 °C is fitted with generalized oscillators (Gen-

Osc). inf is the contribution from the optical transitions at higher energies, is the peak broadening, E0 is the

peak transition energy, A is the transition amplitude, and Eg is the band gap of the corresponding optical

transition.

Ge (Gen-Osc) inf= 1.229±0.065 Unweighted error function= 3.033×10-3

Oscillator E0 (eV) (eV) A Eg (eV)

Tauc-Lorentz 1.997±0.016 0.667±0.067 295.322±15.587 eV 1.726±0.011

Lorentz 2.029±0.074 0.653±0.024 6.697±1.895 -

Lorentz 3.659±0.022 1.525±0.061 14.194±0.518 -

Lorentz 4.146±0.002 0.573±0.017 14.506±0.498 -

Lorentz 5.623±0.055 1.647±0.198 3.582±0.099 -

Table 2.4. The refractive index (n) of GeOx, Al2O3 and HfO2 deposited by ALD at T=270 °C are fitted with

Cauchy model. Cn term has trivial contribution, so is not included in the fitting. Since the band gaps of the

three oxides are beyond the spectrum range (1.24~5.18 eV), the oxides are considered as transparent with

extinction coefficient k=0.

Oxide (Cauchy)

Material An Bn Eg (eV)

GeOx/GeO2 1.642 0.01408 5.6

Al2O3 1.606 0.00746 6.9

HfO2 1.907 0.01993 5.7

2.2.4 Hardware Information

In this thesis, a model of M-2000U, J.A.Woollam spectroscopic ellipsometer is used for

both in-situ and ex-situ measurements. The spectrum range is 240-1000 nm or 1.24-5.17 eV. The

incident angle is ~70 °, with an elliptical light spot of ~0.3 cm × 1.0 cm on the sample surface. The

time resolution for in-situ application is t~1.68 sec. More details about the installation of in-situ

SE on ALD can be found in Section 2.1.

The optic configuration of the ellipsometer is a PCRSA type, where P stands for the

polarizer, CR for the rotating compensator, S for the sample, and A for the analyzer (see Fig. 2.12).

56

The purpose of rotating the compensator is to introduce frequency dependence into the polarization

state as well as the reflection intensity. A Fourier analysis of the measured intensity enables a

precise deconvolution of parameters by effectively excluding the perturbation from noise. More

mathematical details about this configuration can be found in Ref. [97].

To minimize the system error, a system check is required before real measurements. The

system check is measuring a calibration Si wafer with high quality thermal SiO2 (~25 nm). In

principle, any other absorbing semiconductor wafer (e.g. Ge, III-V) with a high quality and thick

enough native oxide can be the calibration sample. In the system check, the rotation angles of all

optical elements (P, CR, A) are calibrated. Also, a “DC offset” check is necessary before every

measurement to calibrate the dark signal, which contributes to the depolarization of the signal; this

check is done by shuttering the detector and measuring the background spectrum. Another

important note for the in-situ application is to correct the window effect, which is the phase shift

caused by the view windows for the incident/reflected light. When applied in-situ SE in a deposition

system, a protecting gas flow is necessary to screen the deposition onto the view windows. More

details about the system corrections can be found in Ref. [97].

Figure 2.15. The optic configuration for M-2000U J.A.Woollam spectroscopic ellipsometer.

57

2.3 Admittance Characterization

2.3.1 Principles of MOSCap Admittance

In this thesis, metal-oxide-semiconductor capacitor (MOSCap) devices were fabricated to

electrically characterize the dielectric/Ge interface properties. The structure of a MOSCap is shown

in Fig. 2.16a, which is a multilayer stacking on a semiconductor substrate. In this structure, the top

metal layer is used as the gate electrode on which a gate voltage (VG) is applied to tune the electrical

properties of the device. The oxide layer is the gate dielectric to isolate the gate metal from the

semiconductor channel but still allow the electric field to penetrate through; the gate dielectric can

also be other insulators like nitrides and oxynitrides. Compared to the above-mentioned MOSFET

structure, MOSCap is simpler without the source and drain (S/D) terminals. The equivalent circuit

of a MOSCap can be represented by conductor (G) and capacitor (C) in parallel (see Fig. 2.17a).

The admittance (Y) (or impedance Z=1/Y) of MOSCap are measured using a combination of DC

bias voltage and a small AC voltage (typically VRMS=10-50 mV). The admittance of the MOSCap

can be expressed as follows:

Y G j C . Eq. 2.18

is the angular frequency of AC voltage. The capacitance C is the dielectric response to the AC

voltage at certain DC bias, so C=dQ/dV. The overall capacitance of a MOSCap device is comprised

of two parts in series, oxide capacitance (Cox) and semiconductor capacitance (Cs), as shown in Fig.

2.17b. The oxide capacitance Cox is constant and expressed as:

oxox

ox

Ct

. Eq. 2.19

ox is the dielectric constant of the gate oxide, while tox is the thickness. In quantifying the oxide

capacitance Cox made of high-k dielectrics like HfO2, a more convenient description is to normalize

58

the oxide thickness with respect to SiO2, and obtain an equivalent oxide thickness (EOT) as

follows: [100]

2

2

0 20 3.4515F/cm

EOT/nm

SiOox oxox

SiOox ox

ox

ox

kkC

kt tt

k

. Eq. 2.20

kox and kSiO2 are the relative dielectric constant of the gate oxide and SiO2, and kSiO2=3.9. The

semiconductor capacitance Cs originates from the fact that electric field can exist and therefore

electric charge can be spatially stored in semiconductors. The semiconductor capacitance Cs varies

with the gate voltage due to the electric gating effect on the semiconductor, as will be discussed in

details later. The measured overall capacitance of the device (Cstack) can be expressed as follows:

1 1 1

stack ox sC C C . Eq. 2.21

Fig. 2.18 shows the simulations for an ideal p-type MOSCap device at various gate

voltages. In Fig. 2.18a, with a sufficient negative DC gate voltage VG and grounding the

semiconductor bulk, the potential difference between the two sides of the gate oxide induces an

electric field in the oxide. The electric field penetrates into the near interface semiconductor, and

consequently induces band up-bending. The relative position of the Fermi level EF with respect to

the bended band results in an accumulation of hole, namely a high density of holes is stored in the

near-interface semiconductor, so the semiconductor capacitance Cs becomes significantly larger

than the oxide capacitance Cox, that is, Cs >> Cox. From Eq. 2.21, the measured overall capacitance

becomes Cstack Cox= Cmax. So by checking the maximum capacitance Cmax from the measured

capacitance vs voltage (C-V) curve, the oxide capacitance Cox can be obtained, from which EOT of

the gate oxide can be derived using Eq. 2.20.

In Fig. 2.18b, a less negative voltage VG is applied on the gate metal, and is matching the

difference between the work function of the gate metal and the Fermi level EF of the semiconductor.

As a result, there is no bend banding or induced charge in either the gate metal or the semiconductor.

59

This stage is called a flat-band (FB) condition. There is a smaller semiconductor capacitance Cs

given by:

2

, where Debye length s Sis D

D a

kTC L

L q N

. Eq. 2.22

So if the doping concentration Na is known, the stack capacitance at the flat-band condition (CFB)

can be obtained by combining Eq. 2.21 and 2.22. Then by checking the measured C-V curve, the

flat-band voltage (VFB) can be obtained. Note for the Ge substrates used in this thesis (Na=0.66-

3.3e15 cm-3), the Debye length at room temperature is estimated to be LD=83.5-186.7 nm.

As the gate voltage becomes even less negative beyond the flat-band condition, down-

bending of the semiconductor occurs (see Fig. 2.18c). As a consequence, the doping carrier or

majority carrier is depleted and a space charge region is created near the interface, which is called

“depletion region” with a width of xd. The maximum depletion width is determined by the doping

concentration Na as follows:

2

4ln /s

dm a i

a

kTx N n

q N

. Eq. 2.23

For example, in a moderately doped p-Si with Na= 3.3e17 cm-3, xdm ~59.4 nm. The resultant

semiconductor capacitance Cs is given by:

2 s

s

dm

Cx

. Eq. 2.24

From this equation, the relatively large depletion width xdm results in a small capacitance density

Cs. From Eq. 2.21, the stack capacitance Cstack is minimized, i.e. Cstack= Cmin. So a measurement of

Cmin, the maximum depletion width xdm can be derived by combining Eq. 2.21 and 2.24 with a

known oxide capacitance Cox, and therefore the doping concentration Na can be obtained.

When a sufficient positive gate voltage VG is applied to a p-MOSCap (see Fig. 2.18d),

further down-bending of the semiconductor band results in an inversion from a majority-carrier

(hole) dominant type into minority-carrier (electron) dominance. The threshold condition of this

60

inversion condition is when the induced electron concentration is comparable to the doping

concentration with a semiconductor band bending (s) as follows:

2

ln as

i

NkTthreshold

q n

. Eq. 2.25

This is also when the depletion width xd is maximized (see Eq. 2.23). Beyond the threshold voltage

(Vth), a high density of electrons is induced and stored in the near-interface semiconductor region,

namely the inversion region. As a result, the semiconductor capacitance Cs becomes significantly

large again, i.e. Cs >> Cox, and therefore the stack capacitance Cstack Cox. But this requires a

sufficient delay time for the device to respond to the gate voltage to generate enough minority

carriers (electrons in p-type semiconductors):

0 02 / , where is minority-carrier lifetime.a it N n Eq. 2.26

Typically, the minority response time is 0.1-10 sec. [1] The response time for the Ge substrates

used in this thesis (Na ~ 2e15 cm-3, 0 ~ 0.001 sec) is ~0.2 sec. So in the inversion region, the

minority-carriers (electron in p-type semiconductors) cannot respond to AC signal with frequencies

>100 Hz, while only majority-carriers (depletion charge) contribute to the semiconductor

capacitance, similar to Eq. 2.24. So a high frequency C-V measurement will detect a minimized

stack capacitance, i.e. Cstack = Cmin, as indicated by the dash line in the C-V simulations in Fig.

2.18d.

Figure 2.16. (a) Equivalent circuit for the admittance measurement of a MOSCap device. (b) The

contributions to the capacitance from the gate oxide and semiconductor.

61

Figure 2.17. Simulated band diagrams, stack capacitance Cstack, and semiconductor capacitance Cs for (a)

accumulation, (b) flat band, (c) depletion, and (d) inversion regimes of an ideal MOSCap made of

Cu(3nm)/HfO2(3nm)/p-Si (Na=3.3e17 cm-3). The work function of Cu is m=4.5 eV, dielectric constant of

HfO2 is k=25, and dielectric constant of Si is k=11.7. The calculated Debye length is LD 7.1 nm. The

threshold voltage is Vth= 0.368 V.

62

2.3.2 High Frequency Admittance Measurement

In this thesis, the dielectric/Ge interface properties were characterized by measuring the

device admittance of high frequencies (f= 75 kHz-1.5 MHz). There are two major types of interface

trap states, (i) trap states inside the band gap of the semiconductor which are related to defects from

the near-interface semiconductor region, and (ii) border trap states which locates in the near-

interface dielectric region. Their influence on the admittance of MOSCap devices will be discussed

respectively later.

The density of the trap states inside the band gap is quantified as Dit, which is a density of

state (DOS) for the interface trap states in a unit of cm-2∙eV-1. A major source of the interface traps

is the dangling bonds on the semiconductor surface, with energies usually locating inside the band

gap of the semiconductor. [101] For example, for a clean Si(100) substrate without any overlayer,

each surface Si atom has two dangling bonds, corresponding to a dangling bond density of 2×Si

areal density2.72×1015 cm-2, which corresponds to 2.38×1015 cm-2∙eV-1 after normalizing by the

band gap of Si (EG=1.14 eV at room temperature). This high density of trap states is not acceptable

for the industrial manufacturing of high quality and reliable devices, which require ~×1010 cm-2∙eV-

1 and below. To address this issue, surface passivation is necessary, namely using chemical/physical

approaches to satisfy the dangling bonds so that the trap states are removed or become electrically

inactive. More details about surface passivation will be discussed later.

In admittance measurements, those interface trap states Dit locating inside the band gap of

the semiconductor electrically contribute to the signal. One effect from these interface trap states

(Dit) on MOSCap devices is to stretch out the C-V characteristics (see Fig. 2.19). As mentioned

above, the electrical characteristics of the semiconductor is gated by band bending of the near-

interface semiconductor region and therefore shifting the relative position of the Fermi level EF

inside the band gap. When the Fermi level EF comes across an interface trap state (see Fig. 2.19a),

63

the state becomes electrically active and its charge occupancy is tuned by external voltage, that is,

extra charge needs gating in order to shift the Fermi level EF compared to a clean interface (no Dit).

In other words, there is another capacitor (Cit) formed by the interface trap states involved in the

electric gating. So the gating efficiency on the semiconductor becomes less, and consequently the

response of the stack capacitance Cstack is delayed, namely a stretched C-V curve (see Fig. 2.19b).

The electrical consequence of the stretch-out is the degradation of device speed. The gating time

constant can be estimated as ~ G oxR C , where RG is the gate resistance. The stretch-out results

in an increase of voltage variation VG required to turn on/off the device, therefore increasing the

gating time, i.e. slowing down the device speed.

Another consequence of the interface trap states Dit inside the band gap is the frequency

dispersion in the admittance characteristics. As shown in Fig. 2.20a, when the Fermi level EF

crosses an interface trap state (Eit), the change of charge occupancy is realized by the

communication between the trap state and the majority-carrier band (valence band for a p-type

semiconductor). This exchange process requires a response time (it), given by Shockley-Read-Hall

statistics of capture and emission rates: [102]

exp /

it

t eff

E kT

v N

. Eq. 2.27

In this equation, E=Eit - EV is the energy difference between the trap state and majority-carrier

band edge, is the capture cross section of the trap, vt is the average thermal velocity of the

majority-carriers, Neff is the effective density of states of the majority-carrier band. As a result of

the response time, the interface trap capacitance Cit and conductance Git become frequency

dependent as follows: [103]

64

2

2

2

2

1

1

1

it it

it

it itit

it

C q D

Gq D

. Eq. 2.28

When sweeping in the frequency domain, an admittance measurement (C-V + G-V) allows to map

out the different trap states with various response time it, and therefore figure out the corresponding

energy distribution of the trap states inside the band gap of the semiconductor. From the equivalent

circuit shown in Fig. 2.19a, the frequency dispersion in C-V is more obvious when the MOSCap is

at a depletion status, because the semiconductor capacitance Cs is relatively small. On the other

hand, in accumulation region, the semiconductor capacitance Cs is significantly large and therefore

dominates, so the frequency dispersion caused by the interface trap states diminishes.

Other than compromising the admittance characteristics, the interface trap states reduce the

effective conduction current by trapping the carriers in the MOSFET channel, and also act as

charged scattering centers for the channel carriers near the interface, lowering their effective

mobility. [1]

The second major type of interface trap states are so-called “border traps”. [104,105] These

trap states locate in the band gap of the dielectric and physically distribute in the near-interface

dielectric region. Since the border trap states may be across the whole dielectric thickness, the

density of border trap states (Nbt) is a bulk density in a unit of cm-3∙eV-1. When the Fermi level EF

in the semiconductor crosses a border trap state Nbt, it becomes electrically active and exchange

charge with the accumulation charge (holes in a p-type semiconductor) through a quantum

tunneling mechanism (see Fig. 2.21a). Similar to Eq. 2.27, the response time of the tunneling

process can be written as:

2

0

exp 2e x

bt

t s

x

v N

. Eq. 2.29

65

In this equation, is the capture cross section of the trap, vt is the average thermal velocity of the

accumulation charge, Ns is the carrier density of the surface semiconductor at accumulation, is

the attenuation factor caused by the tunneling effect, as given by: [104]

*2 ox

Vm E E

. Eq. 2.30

ox

VE is the valence band edge of the gate oxide, E is the energy of the accumulation charge carrier,

andox

VE E represents the barrier height of tunneling from the semiconductor to the border trap

state in the dielectric. An equivalent circuit for a MOSCap with border traps is shown in Fig. 2.21b.

Since the border traps spatially locate across the whole gate oxide thickness, the exponential factor

in Eq. 2.29 can cause wide range of the response time, and therefore the admittance of accumulation

region becomes frequency dependent (see Fig. 2.21c). The calculation of the border trap state

density Nbt will be discussed later.

Figure 2.18. (a) The capacitance contributions for a MOSCap device with interface trap states (Dit). Cit

represents the capacitance from the interface trap states Dit. (b) Stretch-out effect in C-V characteristics

caused by the interface trap states Dit.

66

Figure 2.19. (a) Charge exchange between an interface trap state (Eit) and the majority-carrier band (valence

band for p-type semiconductor). (b) Frequency dispersion in C-V characteristics caused by the interface trap

states Dit inside the band gap.

Figure 2.20. (a) Charge exchange between a border trap state (Nbt) and the accumulation charge carrier (holes

for a p-type semiconductor) at the semiconductor surface near the Fermi level (EF). (b) The equivalent circuit

for a MOSCap device with border trap states; this schematic is reproduced from Ref. [104].

2.3.3 Other Electrical Characteristics

Fig. 2.22a shows an example of hysteresis in bi-directional C-V scans. The hysteresis is

quantified using the flat band voltage shift (VFB) between the forward and backward C-V scans.

The origin of hysteresis is mainly caused by the mobile charges in the gate dielectric. These charges

67

are usually ionic contaminants introduced during the device fabrication processes, like Li+, Na+, K+

and even H+. [1] The mobility of these cations inside the gate dielectric is mainly through a

diffusion process, which is related to the microstructure of the dielectric materials. A dielectric with

a high density of defects, like grain boundary, point defects, and line defects, can provide sufficient

diffusion channels for the cations, which in consequence become mobile enough to respond to the

DC gate voltage during the admittance measurement (see Fig. 2.22b). This spatial charge transfer

modifies the electric field inside the gate dielectric, and therefore results in a shift of the flat band

voltage. Since the response of mobile oxide charges to a positive gate voltage is different from that

to negative (see Fig. 2.22b), a deviation occurs between forward and backward C-V scans (see Fig.

2.22a). The density of the mobile oxide charge (Not) can be quantified as follows: [106]

/ot acc FBN C V q . Eq. 2.31

Cacc is the accumulation capacitance, so Cacc = Cox = Cmax. This quantification is an important

characterization of the dielectric material quality. Other qualities of dielectrics include the dielectric

constants (related to EOT calculation) and leakage current JG, as will be discussed later.

As mentioned above, the EOT of MOSCap devices can be extracted from the max stack

capacitance Cmax or oxide capacitance Cox using Eq. 2.20. By measuring EOT of MOSCap devices

with a series of gate dielectric thickness, one can extract for the dielectric constant with a simple

linear extrapolation. For a multilayer gate dielectric, EOT can be calculated using a series

capacitance model:

,1 ,2

,1 ,2

3.9

EOT

ox ox

ox ox

k k

t t . Eq. 2.32

3.9 is the relative dielectric constant for SiO2.

Another important parameter of characterizing MOSCap device performance is the gate

leakage current JG. Suppressing gate leakage JG in MOSFET is important for scaling down the

device power consumption. There two types of leakage current in a MOSCap device: (i) quantum

68

tunneling effect due to the small thickness (<5 nm) of the gate dielectric layer as required by EOT

down scaling, and (ii) defects induced leakage current across the dielectrics.

To address the tunneling current issue, Intel successfully developed 45nm-node Si-CMOS

technology using a high-k dielectric in conjunction with metal gate. The high dielectric constant of

the gate dielectric allows to further scale down EOT or increase the gate capacitance but maintain

a thick enough dielectric to suppress the tunneling current (see Fig. 2.22a). A benchmark of

common high-k dielectrics is shown in Fig. 2.23b. For better suppressing the gate leakage, a high

band gap EG (>5.5 eV) of the dielectric is necessary because of a higher intrinsic resistivity. Among

all dielectrics, HfO2 and La2O3 show outstanding performance in both band gap EG and dielectric

constant k, but La2O3 is reactive to moisture exposure and therefore limited in the industrial

manufacturing.

The defects inside the dielectric can cause leakage current by diffusion, filament

conduction, grain boundary, etc. This issue can be well addressed by the ALD technology that is

nowadays widely used in industry to prepare the gate dielectrics. The low temperature ALD process

has resulted in high conformal, pin-hole free, uniform, grain-boundary free and amorphous

dielectrics. All these characteristics are favored in suppressing the gate leakage current.

JG-V measurements are used to characterize the gate leakage current. Fig. 2.24a shows an

all-through configuration, wherein the DC bias is applied on the gate metal and the semiconductor

bulk is used as the bottom electrode. While the gate electrode is patterned into finite size (≤100

m), the semiconductor bulk is wafer scale, so the semiconductor bulk resistance (Rbulk) is

negligible. For example, the Ge substrates used in this thesis has a resistivity of 1-5 ∙cm,

thickness=300 m, size= 1 cm ×1 cm, so Rbulk ~ ×10-2 . On the other hand, the contact resistance

(Rc) or series resistance on the gate metal can be as high as ~×103 due to the small contact area

between the probe and gate electrode. Also, the contact resistance Rc can contribute to the

69

admittance measurement by generating frequency dispersion, due to a similar configuration to a

series RC circuit. [107] To minimize the frequency dispersion, it requires a small contact resistance

RC as follows:

3

22

1

1 1~ ~ 10

2 1 MHz 2 μF/cm 100 μm

C stack

C

stack

R C

RC

. Eq. 2.33

In the gate dielectric, three leakage currents contribute to the transport, the oxide conductance from

the intrinsic properties (Gox), quantum tunneling (Gqt), and defects (Gdf). As mentioned above, Gox

is mainly determined by the band gap EG of the dielectric and is negligible if EG >5.5 eV. The

quantum tunneling effect Gqt can be well addressed by using a thick high-k dielectric. The defect

conductance Gdf is one of the key parameters of the dielectric quality, and is primary target to check

in the JG-V measurement. In the semiconductor surface region, there are two contributions to the

transport, surface conduction GSS and tunneling current Gqt from the semiconductor to the gate

dielectric, while the latter can be ignored compared to the former. Since the electrical properties of

the semiconductor surface is tuned by the gate voltage VG, the surface conductance Gss is variable.

At a depletion or inversion state, either whole or part of the semiconductor surface region is

depleted with a minimized carrier density, so Gss is low and limits the overall conductance. On the

other hand, at an accumulation or inversion state, the semiconductor surface shows a high

conductance, so the JG-V measurement is more representative of the gate dielectrics. In this sense,

the gate leakage current JG at an overdriven voltage (VFB-1 V) is identified to characterize the

transport properties of the gate dielectric in MOSCap.

70

Figure 2.21. (a) Hysteresis (VFB) in a bi-directional C-V measurement of MOSCap. (b) The response of

mobile oxide charge to the gate voltage modifies the electric field inside the gate dielectric, and therefore

modifies the band bending (dash).

Figure 2.22. (a) MOSCap using metal/high-k/Si structure. The figure is released by Intel. [108] (b)

Benchmarking the band gap vs the dielectric constant for various dielectrics. The figure is copied from

Ref. [109].In particular, the band gap and dielectric constant for GeO2 are EG=6.1 eV [110] and k~4.5. [16]

Figure 2.23. (a) Schematic and (b) equivalent circuit for measuring JG-V measurement of MOSCap. (c)

Example of leakage current as a function of gate voltage.

71

2.3.4 Data Analysis

This section discusses in details about the data analysis for border trap states. As shown in

the equivalent circuit of Fig. 2.20b, the evolution of the admittance Y for a MOSCap with border

trap states satisfies: [104]

22

2

01

bt

x

ox

j q NdY Y

dx j j e

. Eq. 2.34

x is the depth of the border traps into the gate dielectric. The boundary condition is

0 sY x j C . The first term on the right side of Eq. 2.34 originates from the gate dielectric

capacitance, while the second from the border trap states. Analytical solution for Eq. 2.34 is

extremely complicated, and therefore not applicable in the real data analysis. Instead, numerical

simulations with parameters to be fitted are more straightforward and give relatively reliable

estimation of the border trap state density Nbt.

The ratio between these two terms can be estimated as follows when x = 0+:

222

2 3 2

0 02 2

0

: 1 3 10 1 11

s x xbt

x

ox bt ox

Cj q NYj e j e

j j e q N

. Eq. 2.35

This approximation is using Y(x=0+) jCs, Cs≥100 F/cm2 at accumulation, Nbt~ 4.5e19 cm-3∙eV-

1, and ox=GeO2=60. So the border trap term is a perturbation to the overall admittance. By

substitutingY

j

, Eq. 2.34 can be simplified as follows:

22

2

01

bt

x

ox

q Nd

dx j e

. Eq. 2.36

The boundary condition becomes 0 sx C . At high frequencies, the border trap states are

unable to follow up the AC voltage, namely the border trap term can be neglected, and therefore

Eq. 2.36 can be further simplified as:

72

2

(high frequencies)ox

d

dx

. Eq. 2.37

This is the zeroth order approximation solution of Eq. 2.36 with the following solution:

0

1 1

s oxx

x

C . Eq. 2.38

When x=tox, 0 oxC , which is a simple case of a non-defective MOSCap at accumulation region.

At low frequencies, the response time increases exponentially with the x-position of the border

traps, so those traps far away from the interface are unable to respond to the AC voltage. The

furthest distance of responding traps can be estimated as follows:

02

0 max

ln~ 1 ~ 0.56 nm

2

xe x

. Eq. 2.39

In this approximation, a relatively low AC frequency of 10 kHz and 0~×10-10 sec are used. So an

xmax=1 nm oxide thickness is sufficient for the numerical calculation border trap states, while for

x>1 nm, the evolution of admittance can be described by Eq. 2.37, so

22

2

0

2

0 1 nm, 1

1 nm ,

bt

x

ox

ox

ox

q Ndx

dx j e

dx t

dx

. Eq. 2.40

2.3.5 Hardware Information

In this thesis, electrical measurements were carried out on a Cascade probe station (Summit

11000). The Cascade station has an in-situ camera with a calibrated pixel: distance relation, which

allows to take images of the sample electrodes. The electrode sizes were quantified by processing

the images using ImageJ software. A LCR meter (Hewlett Packard 4285A) was used for measuring

capacitance-voltage (C-V) characteristics; a combination of DC and AC voltage was used, with an

73

AC amplitude of 50 mV and a frequency f=75 kHz-1.5 MHz. A parameter analyzer (Keithley

4200-SCS) was used for measuring leakage-voltage (JG-V) characteristics.

74

2.4 X-ray Photoelectron Spectroscopy

2.4.1 Principles of X-ray Photoelectron Spectroscopy

X-ray photoelectron spectroscopy (XPS) is a surface-sensitive characterization that

quantifies the elemental composition, and chemical states of the elements in materials. An ultrahigh

vacuum (<10-8 Torr) is required for the measurement, so in principle this technology can be applied

to only all solid state materials, but liquids are also possible with a special design of differential

pumping mechanism. [111] The sampling depth by XPS is limited to <10 nm. A deeper

measurement, namely depth profiling, can be realized by combining an in-situ sputtering. A typical

resolution of the elemental analysis (3Li-92U) by XPS is ±5%, while H and He are indistinguishable.

XPS is sensitive to small changes of the chemical states in materials, and therefore provides high

resolution spectra particularly in recognizing the oxidation states in semiconductor devices. A

spatial XPS mapping is possible but with a relatively low resolution (>3 m).

The basic principle of XPS is the photoelectron effect discovered by Heinrich Rudolf Hertz

and later explained by Albert Einstein (see Fig. 2.25a). The incident X-ray photon interacts with

the inner core electrons (K, L, etc.), and the photon energy (h) is transfer to an electron so that the

electron overcomes the binding energy (BE) from the nucleus and becomes free with a kinetic

energy satisfies Einstein’s theory:

-photo-eKE h BE . Eq. 2.41

By detecting the kinetic energy (KE) of the photoelectron and knowing the photon energy,

we can find the binding energy of the electrons of inner cores. Fig. 2.25b shows an example of XPS

survey scan for a GeO2(10 nm)/Ge sample. Multiple peaks for Ge (2p, 3s, 3d, etc.) are identified,

as well as strong signals for O peaks. Note that a small signal for C1s is also observed, which is

originating from the surface organic contamination. Since this carbon contamination is inevitable

75

and its peak position is rather constant with a binding energy of BE=284.8 eV, C1s peak has been

generally used as a reference for correcting the spectrum offset. This offset can be from the XPS

system or the electrostatic potential caused by the surface charging effect on insulating samples.

Figure 2.24. (a) Principles of XPS measurements: interaction between the X-ray photon and inner core

electrons. (b) An XPS survey scan for a GeO2(10 nm)/Ge sample.

2.4.2 Data Analysis

The generated photoelectron intensity from a homogenous material with certain thickness

(z) is expressed as follows (also see Fig. 2.26a):

I n SF z . Eq. 2.42

n is the atomic density of the target element in the material, and SF is the sensitivity factor for the

target element, which is determined by the interaction cross section between the target atom and

photon as well as tooling factors. The sensitivity factor SF is independent of the atomic density n

or the chemical states of the target element. If this thin layer is buried inside the material with

certain depth (z), the emitted photoelectron will be in-elastically scattered by the material when

passing through the z-thick overlayer and become attenuated as follows:

76

expcos

zI

. Eq. 2.43

The exponential term is the attenuation, while is the take-off angle or the measurement angle with

respect to the sample surface normal, and is the inelastic mean free path (IMFP) of the

photoelectron in the material. IMFP is the average distance that photoelectrons of certain kinetic

energy can travel before experiencing an inelastic scattering inside the material, so it is a function

of both the photoelectron energy and material. By integrating all the signals from different depths,

the total intensity from a homogenous sample with finite thickness z0 is as follows:

0

0 / cos

00

exp cos 1cos

zzz

I n SF z n SF e

. Eq. 2.44

For an infinitely thick sample, this equation simply becomes:

0 cosI n SF . Eq. 2.45

The mathematics for a heterostructure is similar but using different parameters. For example, in a

GeO2(10 nm)/Ge sample, the intensity from the substrate is similar to Eq. 2.45, but attenuated by

the GeO2 overlayer is as follows (also see Fig. 2.26b):

2

0 02

cos expcos

GeOGe sub Ge sub Ge sub

Ge Ge GeOGe GeGe

tI n SF

. Eq. 2.46

The quantities with both super- and subscriptions are sensitive to the material, while the

superscription represents the material, and the subscription the atom type (with a chemical state).

For Ge atoms of different chemical states, the energy shifts are relatively small (~ eV) compared

to the large kinetic energy (~×103 eV), so the inelastic mean free path can be approximated as

independent of the chemical state, but still strongly depends on the scattering material. The

sensitivity factor SF is generally the same for Ge atoms in the Ge substrate and GeO2. The signal

from the 10nm-thick GeO2 overlayer is similar to Eq. 2.44 as follows:

22 2 2

4 42

cos 1 expcos

GeOGeO GeO GeO

Ge Ge GeOGe GeGe

tI n SF

. Eq. 2.47

77

So the ratio between the measured signals of Ge4+ and Ge0 can be expressed as:

2 2

224 0

2

: exp 1cos

GeO GeOGeOGeO Ge sub Ge Ge

GeOGe sub Ge subGe GeGe Ge Ge

tnI I

n

. Eq. 2.48

If knowing the take-off angle , density of Ge atoms in Ge substrate and GeO2 ( 0

Ge sub

Gen

, 24

GeO

Gen ) and

the corresponding inelastic mean free paths (Ge sub

Ge , 2GeO

Gen ), we can calculate for the thickness of

GeO2 (2GeOt ). The values for these parameters used in this thesis can found in Tab. 3 and 4.

Fig. 2.27 shows the high resolution XPS spectrum of Ge 3d orbital for the GeO2(10 nm)/Ge

sample with a take-off angle of = 90 °. To quantify the spectrum, a numerical fitting is necessary.

In this thesis, the fitting method is based on Ref. [112–114]. The background of XPS spectra for

Ge 3d is fitting with a Tougaard profile. Each Ge 3d peak is comprised of two sub-peaks,

corresponding to Ge 3d5/2 and Ge 3d3/2, respectively. The sub-peak shapes are described with a

Voigt function with a branch ratio of 0.667. The Ge0 3d5/2 peak is referenced to 29.3 eV. Binding

energy shift for Ge 3d3/2 with respect to Ge 3d5/2 is 0.58eV. The areal ratio between Ge 3d3/2 and

Ge 3d5/2 is fixed at 0.67 which is the ratio of the orbital degeneracy. The core level shifts for +1,

+2, +3, and +4 are 0.8, 1.8, 2.7, 3.3 eV, respectively. A CasaXPS software is used to perform the

numerical fitting.

The numerical fitting of Fig. 2.27 shows that the Ge4+ state dominates in the signal. This is

because the thick GeO2 overlayer strongly attenuates the signal from the underlying Ge substrate.

Plugging the fitted areal ratio between Ge4+ and Ge0 (from the Ge substrate) and the other

parameters from Tab. 3 into Eq. 2.46, the thickness of GeO2 is calculated to be 9.84 nm, which

agrees well with the result of 10.1±0.3 nm given by spectroscopic ellipsometry measurement. So

XPS is a relatively precise method of determining film thickness, especially the native thickness

on semiconductors. When characterizing ultrathin films (tox≤ 3 nm), the signal from the oxide

78

becomes small compared to that of the semiconductor, and therefore may result in large uncertainty

in determining the oxide thickness. To further improve the precision, an angel-resolved XPS

measurement is necessary. From Eq. 2.46, the intensity ratio is sensitive the take-off angle . By

varying the take-off angle from 90 ° to , the effective sampling depth changes by a factor of cos.

By a linear extrapolation, we can obtain a precise thickness of native oxide (see Fig. 2.28). As for

multilayer structures, the principles of data analysis is the same, but much more complicated in

mathematics.

Table 2.5. XPS parameters for Ge3d, O1s and Al2p orbitals used in this work. The calculation of the inelastic

mean free path is performed based on Ref. [115]. The XPS light source is Al K (h= 1486.70 eV).

Orbital RSF BE (eV) KE (eV) IMFP in Ge (Å) IMFP in GeO2 (Å) IMFP in Al2O3 (Å)

Ge3d 0.380 29.3 1457.41 29.92 26.46 33.60

O1s 0.660 530.0 956.71 21.61 19.14 24.32

Al2p 0.185 74.6 1412.11 29.19 25.82 32.78

Table 2.6. Some basic properties of materials used in this work.

Material Density (g/cm3)

Formula density (cm-3)

Monolayer thickness

(Å) Dielectric constant

GeO2 4.25 2.44E+22 3.45 4.5 [16]

AlO1.5 3.00 3.54E+22 3.04 8

HfO2 9.68 2.77E+22 3.31 22

Ge 5.32 4.60E+22 2.79 16

Figure 2.25. (a) XPS signal from a homogenous material. (b) XPS signal from a GeO2(10 nm)/Ge sample.

79

Figure 2.26. A high resolution XPS measurement of Ge3d for a GeO2(10 nm)/Ge sample. The take-off angle

of the measurement is 90 °.

Figure 2.27. Calibrating SiO2 thickness on Si substrate using angle resolved XPS. This figure is reproduced

from Ref. [116].

2.4.3 Hardware Information

Fig. 2.29 shows the typical hardware configuration for a monochromatic XPS system. The

model of XPS system used in this thesis is Kratos Analytical Axis Ultra (Al K source). An e-beam

is shone on an Al anode to generate the characteristic Al Klines. Based on the Bragg’s

diffraction law, a quartz crystal is used to filter the unwanted K line that causes the line

1.0 1.2 1.4 1.6 1.8 2.00

1

2

3

4

5

1.9 nm

2.3 nm3.6 nm

4.3 nm

6.4 nm

ln(1

+I S

iO2/I

Si)

1/cos

9.0 nm

80

broadening. The energy resolution of XPS is mainly determined by the quality of the quartz crystal

mono-chromator. Then the X-ray beam is focused and shone on the surface. The spatially resolution

of XPS mapping is mainly determined by the beam spot size. The generated photoelectron signal

is then energetically analyzed in a hemispherical shape analyzer by dispersing the photoelectrons

of different energies into different channels which are simultaneously collected to form a complete

spectrum.

Figure 2.28. The hardware setup for an XPS measurement. The figure is take from Ref. [117].

Chapter 3

Atomic Layer Deposition of High-k Dielectrics on Germanium

3.1 Introduction

Integrating high-permittivity (high- or high-k) dielectrics on Ge is necessary for utilizing

Ge as the channel material in the CMOS technology, which requires highly downscaled device

dimensions and high carrier mobility. As mentioned above in Chapter 1, the deposition of high-k

dielectrics on Ge, usually performed by atomic layer deposition (ALD) method, is not

straightforward. [10,15] In qualifying the electrical performance of Ge devices, a few factors need

to be under consideration: (i) a high capacitance density from the gate dielectric, which requires an

small thickness and large dielectric constant of the dielectric layer. (ii) Superior qualities of the

dielectric/Ge interface, including a low interface roughness, a low density of electrically active

defect trap states locating in the near-interface semiconductor and the gate dielectrics. (iii) A low

leakage current through the gate dielectrics, which is primarily a characteristic of the quality of the

dielectric itself; defect states in the dielectrics related to the leakage current are created during the

fabrication processes, e.g. unideal initial nucleation and interfacial reaction, etc.

To take into account all the three important factors, a process control metrology is

inevitable. Meanwhile the conjunction with an atomic scale understanding of the mechanism is

beneficial to facilitate the process optimization. This chapter will bring onto the table about how to

use in-situ and ex-situ metrologies to control the fabrication process of Ge MOSCap devices,

entailing in-situ preparing atomically flat and oxygen free Ge surface in Section 3.2, exploring the

surface chemistries of ALD nucleation behaviors on Ge substrates with different surface states in

Section 3.3, and optimizing a gate dielectric stack on Ge with superior electrical characteristics in

82

Section 3.4. The experimental discoveries are compared to ReaxFF simulations from the

collaborators (Prof. Adri van Duin group at Penn State University) for more mechanism details at

an atomic scale. This combinatory approach of experiments and simulations may offer a more

straightforward and rational route in developing Ge-based CMOS technologies in the future

semiconductor manufacturing.

Note that the simulation results were performed by the collaborators from Prof. Adri van

Duin group at Penn State University. In order to give complete explanation of the ALD mechanism,

some of the simulation results are also included in this thesis. Also, note that part of the results in

this chapter have been published in our previous papers, so similar figures, data analysis, and

discussions may be found somewhere else. [35,118]

3.2 Preparing Pristine Ge Surfaces

To ensure the performance of Ge-based transistors, previous researches have pointed out

the key role of the interface between Ge and the gate insulator, which affects the gate leakage

current, capacitance density, sub-threshold characteristics and the effective carrier mobility in the

channel. [119] So it becomes important in the Ge device manufacturing to prepare a pristine Ge

surface with atomic flatness and free of organic contaminants, native oxide, and metal

contaminants, so as to minimize the potential origin of interface defects in the subsequent

semiconductor processing.

Various surface preparation approaches have been successfully developed to remove the

native oxide and organics on Ge, including wet chemical clean, argon ion bombardment and H2

plasma or atomic hydrogen (H*) clean. [119] Compared to other cleaning processes, the H2

plasma [120] or atomic hydrogen (H*) [121] clean appears attractive because it can simultaneously

remove the native oxide and organic contaminations on Ge. Also, it can be performed in-situ at a

83

relatively low temperature (≥180 °C), [122] and does not require an ultra-high-vacuum (UHV)

annealing at high temperatures (>390 °C) to achieve a pristine Ge surface. [119] Because the

relatively low dielectric constants of GeO2 (k~4.5) limits the device scaling, a removal of native

oxide on Ge by H2 plasma is particularly important for developing high-k/Ge transistors. Previous

studies on this issue mainly focused on the effectiveness of removing oxygen from the Ge

surface, [121–125] but rarely discussed the reaction mechanism between the H* radicals and GeO2,

which is essentially important to understand the origin of electrically active defects present at the

dielectric/Ge interface of the transistor devices. As will be shown in this section, the H2 plasma

clean of GeO2 undergoes a distinct reaction scenario at low temperatures compared to that at high

temperatures. The reaction details were examined through the analysis of atomic force microscopy

(AFM) and Raman spectroscopy, and a transition temperature was mapped out between the two

mechanisms.

3.2.1 Experimental details

The installation of an in-situ plasma source on ALD system can be found in Section 2.1.2.

Inside the plasma source, H2 gas mixed with Ar gas is dissociated or ionized by the RF

electromagnetic field inside the induction coil. The majority of the plasma particles are the same

as the initial input flow, with neutral Ar and H2, while a small portion (<10 %) is dissociated or

ionized, forming Ar+, H2+, H+, H-, atomic hydrogen (or H* radicals) and free electrons (see Fig.

3.1). Among all these particles, the H* radicals are believed to be the most reactive in removing the

native oxide on Ge. [125]

GeO2(10 nm)/p-Ge(100) samples were used in this work for studying the H2 plasma clean.

P-Ge(100) substrates (Ga-doping, =1.0-5.0 cm, supplied by Umicore Electro-Optic Materials)

were degreased by acetone/isopropyl-alcohol(IPA), and then rinsed by DI-H2O to remove the

84

surface native oxide. The pre-cleaned Ge substrates were annealed in a muffle furnace in air at 490

°C/2 hrs to grow a GeO2 overlayer, with a thickness of 10.1±0.3 nm determined by spectroscopic

ellipsometry (M-2000U, = 240~1000 nm, J.A. Woollam Co.). Then the GeO2/Ge samples were

loaded into an ALD chamber (ALD-150LX, Kurt J. Lesker Co.) for in-situ H2 plasma treatments

at various temperatures (substrate temperature Tsub= 110-330 °C). The H2 plasma was generated by

a remote radio frequency (RF) plasma source (H2: Ar= 8: 112 sccm, 300 Watts/30 min) with a

background Ar pressure of 1.2 Torr. After the H2 plasma clean, a capping layer of 2 nm Al2O3 was

grown (using tri-methyl-aluminum and H2O as the precursors), to protect the Ge surface from the

air contamination during the sample transfer for further measurements. The chemical states of Ge

were characterized by X-ray photoelectron spectroscopy (XPS, Kratos Analytical Axis Ultra, Al

Ksource). The surface morphology of all samples was measured by atomic force microscopy

(AFM, Bruker Dimension Icon). The vibrational modes of surface Ge was examined by Raman

spectroscopy (=488 nm, 5.0 mWatts, Horiba LabRam).

85

Figure 3.1. A schematic for the remote inductively coupled H2 plasma source. An induction coil is used to

generate AC electromagnetic field inside quartz tube. The flowing gas molecules are ionized or dissociated

and then react with the sample surface.

3.2.2 XPS Studies

The effectiveness of cleaning GeO2 by the H2 plasma was firstly examined by high

resolution XPS measurements. In Fig. 3.2a, the as-prepared GeO2(10 nm)/Ge sample before the H2

plasma shows a strong signal (33.4 and 32.8 eV for Ge 3d3/2, 3d5/2) for Ge+4 state originating from

the 10 nm GeO2 overlayer, while the underlying Ge substrate shows a weak Ge0 signal (29.9 and

29.3 eV for Ge 3d3/2, 3d5/2), due to the attenuation of the photoelectron intensity by the thick GeO2

overlayer. After the H2 plasma (T= 110 and 330 °C in Figs. 3.2b, c), the samples showed no

86

observable Ge+4 signal but a strong peak doublet for Ge0, so GeO2 has been effectively removed by

the H2 plasma. [121,124] The weak signal for Ge+1 is attributed to the interfacial transition between

the Al2O3 capping layer and Ge substrate. [35,126]

The driving force of cleaning GeO2 is mainly attributed to the highly reactive atomic

hydrogen (or H* radicals) generated by the plasma source. [120–122,124] As a control experiment,

a simple exposure of GeO2/Ge to a H2 flow with zero plasma power did not result in obvious change

of the GeO2 thickness or the chemical states of Ge (data no shown here). Three reaction routes are

proposed, with the free energy (G) calculations based on Reference [33]:

(I) GeO2(s) + 8H*(g) GeH4(g) + 2H2O(g),

G= -1410 kJ/mol at 400 K and -1328 kJ/mol at 600 K;

(II) GeO2(s) + 4H*(g) Ge(s) + 2H2O(g),

G= -739 kJ/mol at 400 K and -716 kJ/mol at 600 K;

(III) GeO2(s) + 2H*(g) GeO(g) + H2O(g),

G= 120 kJ/mol at 400 K and 86 kJ/mol at 600 K.

While the route (III) is thermodynamically prohibited, (I) and (II) are energetically favored

candidates for the H2 plasma cleaning mechanisms. Comparing (I) and (II), both reactions result in

a removal of O species in the GeO2 layer by releasing volatile H2O and leave no oxidized Ge states

in the samples, so they cannot be distinguished by using only XPS measurements. In the route (I),

the whole GeO2 layer is removed and the underlying Ge bulk is directly exposed, while an overlayer

of elemental Ge is created on the Ge bulk in (II), so the overlayer thickness after the H2 plasma

treatment is an effective indicator to identify the reaction mechanism.

87

Figure 3.2. High resolution XPS measurements of Ge 3d at a 90 ° take-off angle for GeO2(10 nm)/Ge samples

(a) before H2 plasma, (b) after H2 plasma at T=110 °C, and (c) after H2 plasma at T=330 °C. The numerical

deconvolution is based on References [112–114] with a Tougaard background profile. A Voigt profile (0.667

branching ratio) is used to describe all the peak shapes. The Ge0 3d5/2 peak is referenced to 29.3 eV. Binding

energy shift for Ge 3d3/2 with respect to Ge 3d5/2 is 0.58eV, and the areal ratio between Ge 3d3/2 and Ge 3d5/2

is fixed at 0.67. The core level shifts for +1, +2, +3, and +4 are 0.8,1.8,2.7,3.3 eV, respectively.

3.2.3 AFM Studies

A shadow mask method was used to figure out the overlayer thickness after the H2 plasma.

During the plasma exposure, a piece of smooth sapphire substrate (surface roughness Rq=0.1 nm)

was used to partially mask the GeO2/Ge samples (Fig. 3.3a). As a result, a step was created across

the shadowed and un-shadowed regions. After the plasma, no A2lO3 capping was grown and the

samples were directly transferred to the AFM system to measure the step height. Fig. 3.3b shows

the step profile for the GeO2/Ge sample after the H2 plasma at 110 °C. The left side of the step was

shadowed by the sapphire mask, namely made of GeO2, while the right side was cleaned by the H2

plasma. The step height is ~8.48 nm, slightly smaller than the expected height of 10 nm if the whole

GeO2 is completely removed by following the route (I). But this small deviation in the step height

88

can be understood as a result of surface oxidation on the right side of the step by air exposure. A

typical thickness of native GeOx formed by air oxidation on Ge is ~3.06±0.04 nm as measured by

spectroscopic ellipsometry, corresponding to a transformation of ~1.70 nm thick surface Ge layer

into native GeOx. Consequently, the formation of GeOx overlayer on the cleaned Ge side lessens

the overall step height (see the structure schematic in Fig. 3.3b). Regarding this matter, the

theoretical step height was calculated to be 8.74 nm, which is comparable to the experimental result

(Tab. 1). So the route (I) is identified as the reaction mechanism for the H2 plasma clean at 110 °C.

As for the high temperature plasma at 330 °C, however, the measured step height was only ~3.25

nm (Fig. 3.3c), indicating a distinct mechanism. Considering the route (II) as the mechanism and

taking into account the surface oxidation effect (see the structure schematic in Fig. 3.3c), the

theoretical step height becomes 3.14 nm and is comparable with the high temperature result (Tab.

1). So we conclude that at 330 °C, the H2 plasma only reduces GeO2 into elemental Ge by following

the route (II). The absence of GeH4 formation in the route (II) at high temperatures agrees with the

residual gas analysis (RGA) at 300 °C by Schneider et al. [120] Unfortunately, they did not show

the results for lower temperatures.

Another interesting evidence of the temperature dependence of the cleaning mechanism is

the Ge surface roughening induced by the H2 plasma. In-situ spectroscopic ellipsometry (SE) has

been used to real-time monitor the H2 plasma clean on GeO2(~1 nm)/Ge surfaces at various

temperatures (see Fig. 3.4a). At 110 C, H2 plasma can effectively remove the native GeO2, but an

over-exposure resulted in an irreversible increase in the “GeOx” thickness. This increase indicates

that the surface is deviating from pure Ge during the over-exposure of H2 plasma. Ex-situ atomic

force microscopy (AFM) measurement (see Fig. 3.4b) showed that the surface roughness was

significantly increased (root mean square of roughness RMS=0.472 nm), compared to that before

the plasma clean (RMS=0.295 nm). So the ‘GeOx’ thickness increase observed by in-situ SE is the

surface roughness layer, whose optical properties are different from Ge bulk. [127] A similar effect

89

was also observed in the H2 plasma exposure on Si surface. [128] In contrast, extra H* dose at high

temperatures (≥270 C) maintained a smooth surface (RMS=0.294nm). The reversible increase of

the nominal “GeOx” thickness in SE (Fig. 3.4a) can be attributed to a surface heating effect by the

plasma exposure. The critical temperature to avoid the surface roughening by the H2 plasma was

found to be at TC≈ 270 C. As will be shown later, this temperature is used as the process

temperature of high-k ALD growth on Ge.

Table 3.1. Comparison between the calculated and measured step heights created by the H2 plasma clean at

110 and 330 °C. The starting sample before the H2 plasma was using a GeO2(10.1 nm)/Ge structure. The

calculation was assuming (i) reaction mechanisms of routes (I) and (II) for 110 and 330 °C, respectively, and

(ii) 3.06 nm GeOx formation by air oxidation on a plasma cleaned Ge surface. A density of 5.32 g/cm3 was

used for both the Ge bulk and Ge overlayer created by the H2 plasma, while 4.25 g/cm3 for GeO2 and native

GeOx formed by air exposure.

Temperature (°C) Calculated Height (nm) Measured Height (nm)

110 8.74 8.48±0.34

330 3.14 3.25±0.28

90

Figure 3.3. (a) A schematic for using a shadow mask to create a step during the H2 plasma clean. AFM line

scans across the height steps and the corresponding schematics for the resultant structures created in the H2

plasma at (b) 110 °C and (c) 330 °C, with the left side of the steps covered by the shadow mask.

91

Figure 3.4. (a) In-situ SE monitoring of H2 plasma clean of GeO2(~1 nm)/Ge(100) at 110 and 270 °C. (b)

The surface morphology of Ge(100) substrate after H2 plasma clean. (c) The surface roughening as a function

of temperature.

92

3.2.4 Raman Studies

The reaction mechanisms were further examined by Raman spectroscopy. The Raman

spectrum for the as-prepared GeO2(10 nm)/Ge sample (Fig. 3.5a) is an almost identical to that a Ge

substrate right after the acetone/IPA/DI-H2O clean (data not shown here), so the Raman peak (300.9

cm-1, FWHM 2.9 cm-1) is identified as the Ge bulk. [129] The small asymmetry at lower

wavenumbers (~290.9 cm-1) is ascribed to the electronic Raman scattering induced by the free

carriers in the Ge bulk, [130,131] and fitted with a secondary peak labeled as ‘doping’. No Fano

function was used to fit the Ge bulk spectrum because of the relatively low doping level (0.7~3e15

cm-3). After a 100 °C H2 plasma and subsequently capping with an Al2O3 protecting layer, the

Raman spectrum (Fig. 3.5b) is similar to that of the Ge bulk, but with another small and broad peak

(red curve, 296.8 cm-1, FWHM 6.8 cm-1), which is attributed to the defects on the Ge surface

caused by the low-temperature H2 plasma damage [130] and is labeled as ‘poly-Ge’. On the

contrary, the high-temperature H2 plasma treatment (330 °C in Fig. 3.5c) resulted in a significant

intensity of poly-Ge peak. The rise of poly-Ge signal agrees with the above-mentioned reaction

route (II) for the high-temperature H2 plasma. In the route (II), GeO2 is reduced into elemental Ge

and forms an additional overlayer on the Ge bulk. Since an even higher temperature (≥350 °C) is

required to fully crystallize Ge, [132–134] one can expect that this Ge overlayer formed at 330 °C

is nano-polycrystalline and highly defective. This explains the red-shift and peak broadening of the

poly-Ge peak (296.5 cm-1, FWHM 5.9 cm-1) in Fig. 3.5c. [134] Note that the spectral weight of

the doping peak in Fig. 3.5c is also increased significantly, suggesting additional doping source or

charge injection. As reported by J. Cho and R. J. Nemanich, H-induced surface states were absent

after a 300 °C H2 plasma on Ge(100), but the dangling bond states were well identified by angle-

resolved ultraviolet photoemission spectroscopy (ARUPS). [122] Previous studies have showed

that the surface dangling bonds on Ge(100) serve as acceptors or hole injectors by effectively

93

pinning the Fermi energy (EF) close to the valence band maximum (EV). [121,135,136] So the

enhancement of the doping peak observed in Fig. 3.5c is attributed to the abundant dangling bonds

in the plasma-created Ge overlayer and at the overlayer/bulk interface. On the other hand, the still

weak signal for ‘doping’ after the 110 °C plasma indicates an effective passivation of the surface

dangling bonds by H* radicals, namely forming Ge: H termination, which can result in a EF-

unpinned Ge surface, [121] as also evidenced by previous ARUPS measurements. [122,125]

The transition temperature between the reaction routes (I) and (II) were double-checked by

mapping the signal intensity of poly-Ge in Raman spectra as a function of temperature. As shown

in Fig. 3.5d, the transition begins around 270 °C with a significant rise of the poly-Ge signal, while

a full onset of the route (II) occurs at T≥ 300 °C. The driving force for this transition at higher

temperatures is ascribed to the thermodynamically unstable nature of GeH4, [35] as shown in the

Ellingham diagram calculation in Fig. 3.6. Previous report of GeH4 decomposition confirmed a

critical temperature of 305 °C to overcome the activation barrier, [137] which agrees with our

Raman measurements. As a comparison, SiH4 is more stable (see black curve in Fig. 3.6) and

therefore requires a higher decomposition temperature (>375 °C). [138] This explains the presence

of SiH4 in RGA measurements during the H2 plasma on Si at 300 °C. [120]

94

Figure 3.5. Raman measurements for GeO2(10 nm)/Ge samples (a) before H2 plasma, (b) after H2 plasma at

T=110 °C, and (c) after H2 plasma at T=330 °C. The plasma treated samples were subsequently in-situ capped

with 2 nm Al2O3 by ALD in order protect the surface from air contamination during the sample transfer. (d)

Areal percentage of poly-Ge signal for H2-plasma treated GeO2(10 nm)/Ge samples as a function of the

process temperature. The poly-Ge percentage is normalized to the Ge bulk signal. A linear background profile

was used for the numerical deconvolution. A Voigt profile (0.98 branching ratio) is used to describe all the

peak shapes. The Ge bulk peak was referenced to 300.9 cm-1.

Figure 3.6. Ellingham diagram calculation for the decomposition of GeH4. The same calculation was also

done for SiH4 as a comparison. The calculation was based on the thermochemical data from Reference [33].

200 400 600 800 1000 1200

-200

-160

-120

-80

-40

SiH4 (g) →Si(s) + 2H

2 (g)GeH

4 (g) → Ge(s) + 2H2 (g)

G

(kJ/m

ol)

Temperature (K)

95

3.2.5 Discussions and summary

The unstable nature of GeH4 can explain the above-mentioned temperature dependence of

the H2 plasma roughening to Ge surfaces. The roughening mechanism at low temperatures is

proposed as: Ge(s) + 4H*(g) GeH4(g). Analogue to Si surface, [139] we believe that the H2

plasma etching of Ge should also show a preference to certain crystallographic orientation ({111}

for Si). This preference results in isotropic removal of surface Ge atoms and therefore a surface

roughening. At high temperatures (≥ 270 °C), the unstable nature of GeH4 inhibits the direct

reaction between H* and Ge, and therefore eliminates the surface roughening. In this sense, the

route (II) mechanism without the GeH4 formation becomes thermodynamically favored than (I).

Regarding the free energy change for both reactions, the route (I) appears more possible than (II),

but this is because of the different initial reaction condition, namely a different H* amount involved.

To equal the initial conditions, extra H* radicals are needed in the route (II) but do not participate

in the reducing reaction with GeO2, so will consequently recombine and form molecular H2 as

follows: 4H* 2H2. Then the route (II) can be rewritten as:

(IV) GeO2(s) + 8H*(g) Ge(s) + 2H2O(g) + 2H2,

G= -1532 kJ/mol at 400 K and -1467 kJ/mol at 600 K.

Overall, the route (IV) after balancing the initial condition is more thermodynamically favored than

(I) at temperatures ≥300 °C, while at low temperatures, the route (IV) is kinetically inhibited.

In summary, the H2 plasma can effectively remove GeO2 on Ge surfaces. However, the low

temperature H2 plasma results in surface roughening, while the high temperature plasma leaves a

defective nanocrystalline Ge overlayer. Neither of these results are favorable for improving Ge

device performances. To address this issue, a combination of wet-chemical clean with a high

temperature H2 plasma is used (see Fig. 3.7). The wet-chemical clean is able most of the organic

contaminants and native oxide, with an ultrathin layer of residual suboxide (<0.5 nm). A 270 °C

96

H2 plasma can remove this small amount of residual oxide without roughening the surface or

creating thick nanocrystalline Ge layer. As shown by a reflection high-energy electron diffraction

(RHEED) measurement in Fig. 3.7, the optimized cleaning process has yielded a surface with a

well-ordered structure of 2×1 surface reconstruction, which is an indication of pristine Ge surface.

Figure 3.7. The optimized cleaning process results in a pristine Ge surface free of oxygen, with minimized

formation of nanocrystalline Ge, atomically flat surface and a well-defined (2×1) surface reconstruction.

97

3.3 Al2O3 ALD Nucleation Mechanism on Ge

Atomic layer deposition (ALD) has become an important synthesis technique in various

fields of nanotechnology, e.g. semiconductor processing and microelectronics [30,81,140–143],

biomedical applications [144–146] and protective coatings [31,80,147–154]. Considering the large

application space in diverse technological areas it seems surprising that relatively little is known

quantitatively about the kinetic processes in ALD at an atomic level, such as adsorption/desorption,

diffusion, and chemical reactions of molecular species.

When replacing Si with Ge as the new channel materials in CMOS technology, the rapid

pace in the semiconductor industries demands for a better comprehension of ALD processes so as

to master the increasing complexity level and therefore to accelerate optimization cycles. Given

this scenario, the capability of in-situ metrology in conjunction with an atomic scale understanding

of ALD mechanism becomes inevitable. Here, ALD is used to integrate an ultrathin dielectric film

as the gate insulator. The resultant device characteristics are governed by the quality of the

dielectric, but more so by the electrical properties of the dielectric/semiconductor interface.

However, due to the lack of atomic scale understanding of ALD growth kinetics on the

semiconductor surface, the roadmap to obtain highly-scaled and electrically reliable gate stacks is

not straightforward. [30] In case of Ge, it has been found that a direct ALD of high permittivity

(high-k) dielectric on a Ge surface results in poor electrical characteristics dominated by the trap

response formed at the dielectric/semiconductor interface. [17,19,155–158] To address this

challenge, the introduction of an ultrathin interlayer that forms an electrically well behaved

interface with Ge has been proposed. This is layer is thin enough to still allow for competitive

capacitance density scaling and yet thick enough to avoid trap formation through chemical reaction

mechanisms between the high-k and Ge. [159–161] Various materials and deposition conditions

have been applied and resulted in noticeable improvements, but details of the dominant

98

mechanisms for the ALD processes at the atomic scale and how they can be utilized to improve

film nucleation and surface passivation still remain unclear.

In this chapter, a combined experimental and theoretical approach is used to gain insights

into the complex ALD process using in-situ real time spectroscopic ellipsometry (SE) in

conjunction with ReaxFF reactive force field simulations. [162] The ALD deposition kinetics of

Al2O3 on hydrogenated and oxidized Ge surfaces were investigated to quantify the influence of

initial surface chemical states on the reaction kinetics and to relate it to the trap formation during

ALD nucleation. Metal-oxide-semiconductor capacitor (MOSCap) devices were fabricated to

quantify the electrical characteristics of dielectric/Ge interfaces using impedance spectroscopy and

were correlated to the surface chemistry and nucleation mechanism.

3.3.1 Baseline for ALD growth

Before proceeding to the high-k ALD on Ge surface, the baseline of ALD growth needs to

be established. A growth baseline is a parameter set for a standard and optimized ALD process of

certain material, including the growth temperature, precursor dose/purge time, growth rate, optical

and electrical properties of the grown materials, etc. The baseline is particular important when

performing ALD process on a surface that has not been well studied before, because the parameters

from the baseline are good references for the growth of interest that may significantly differ from

the standard growth due to a different surface chemistry. This comparison serves as a good source

of clues for figuring out the nucleation behavior on the unfamiliar surface.

In this thesis, in-situ SE is generally used for real-time monitoring all the processes inside

the ALD chamber, including the ALD growth. The following baseline parameters can be extracted

from the standard ALD growth using in-situ SE:

99

(i) Growth rate or growth per cycle (GPC), which is typically obtained by performing a

long deposition so that the as-obtained thin films are thick enough (> 5nm) for precise thickness

measurement (typical thickness resolution for SE is sub-Angstrom).

(ii) Dielectric functions of the materials 1/2, which are also extracted from the long ALD

deposition; the parameterization of all the ALD grown oxides has been addressed in Section 2.2.3.

(iii) Dose amplitude of metal organic (MO) precursor or adsorption level, which is

represented by the nominal thickness increase measured by SE when MO precursor is dose on the

sample surface; as will be shown later, this quantity is particularly useful for analyzing the initial

nucleation behavior of ALD deposition.

(iii) The ratio between GPC and MO adsorption, which is namely mathematical division

between the two already measured quantities; this ratio is not trivial, because for a standard growth,

the reaction mechanism is well defined, namely the product (i.e. GPC) is proportional to the reactant

(MO dose amplitude); deviation of this ratio from the baseline indicates a new reaction chemistry

present during the ALD growth.

Since Al2O3 is the primary material of interest, detailed baseline has been established for

its ALD growth at T=270 °C (see Fig. 3.8). The baseline ALD of Al2O3 using TMA and H2O is

performed on a Al2O3(30 nm)/SiO2(25 nm)/Si substrate, whose optical properties were well

characterized before the deposition. The baseline growth shows a linear growth mode (see Fig.

3.8a), which is typically observed for a homo-deposition of oxide ALD. The average growth rate

GPC is determined to be 0.86 Å/cycle. Fig. 3.8b shows the details of the ALD growth. Each cycle

starts with a TMA dose, which results in a nominal thickness increase. The fully saturated

amplitude of this increase represents the TMA adsorption level. After purging the residual

unreacted TMA in the chamber with Ar, H2O dose oxidizes the adsorbed TMA into Al2O3, resulting

a smaller drop in thickness. Another Ar purge removes the residual unreacted gas phase H2O. At

the end of the cycle, a net thickness of Al2O3 is grown, which is GPC. The GPC/TMA adsorption

100

ratio is determined to be ~0.404 (see Fig. 3.8c), which on the other hand means the average TMA

adsorption of ~2.13 Å/cycle.

Another useful parameter one can further extract from the above-mentioned baseline

parameters is the density of bonding sites by calculating the areal density of MO adsorption. For

example, to evaluate the bonding site density available for TMA adsorption, we can do the

following calculation: assuming a density of 3.0 g/cm3 for ALD grown Al2O3, [31] then the density

of Al atom is = 3.54×1022 cm-3, areal density of Al atom is 2/3= 1.08×1015 cm-3, and the monolayer

thickness of AlO1.5 is -1/3= 3.04 Å. The baseline GPC of ~0.86 Å corresponds to 0.28 monolayer

of AlO1.5, so an Al areal density of 0.28×2/3=3.05 nm-2 is deposited in each ALD cycle. Since one

TMA molecule contributes one Al atom, the TMA adsorption density= 3.05 nm-2, which is also the

bonding site density available on the sample surface in each cycle.

Figure 3.8. Al2O3 ALD on Al2O3(30 nm)/SiO2(25 nm)/Si. (a) In-situ SE monitoring of real-time thickness.

(b) Enlarged section for Figure (a) with details of the sequential precursor doses, TMA adsorption and GPC.

(c) Extracted ratio of GPC to TMA adsorption as a function of growth cycle. Results published in Ref. [35].

3.3.2 Al2O3 ALD on H-terminated Ge surface

The H-terminated Ge surface (Ge:H) is prepared by exposing the Ge surface to H2 plasma,

as discussed in details in Section 3.1. Even though GeH4 is thermodynamically unstable at 270 °C,

101

a single bonded structure of Ge-Ge-H* can be still allowed to exist to a large extent on a pristine

Ge surface without the defective nanocrystalline Ge, as shown by the Raman spectrum in Fig. 3.7.

A direct Al2O3 ALD on Ge: H surface shows an obvious nucleation delay. The growth

curve measured by in-situ SE is nonlinear (Fig. 3.9a), and the resultant growth rate GPC is gradually

increasing with the ALD cycle number, with an initial growth rate significantly lower than the base

rate for Al2O3 ALD baseline (Fig. 3.9b). The TMA dose rise in initial ALD cycles is weak (red

circles in figure 1c), indicating a difficulty of TMA adsorption on Ge: H. For example, the TMA

adsorption on Ge:H in the very first cycle is only ~17% of the baseline. As mentioned above, the

baseline adsorption density is calculated to be 3.05 nm-2, so in the first cycle, there is only ~0.5/nm2

coverage. The adsorption level approaches the baseline around 25~27th cycle. By comparing the

GPC and the TMA adsorption, we can see that they are relatively proportional to each other (Fig.

3.9c). The large uncertain of the ratio in the initial few cycles is due to the low GPC and TMA

adsorption, which are comparable to the noise level. So we can conclude that the nucleation delay

is mainly caused by the low sticking coefficient of TMA precursor on Ge:H surface. The improved

GPC in later cycles is due to an increased Al2O3 surface coverage through lateral expansion from

initial nucleation sites. Until the surface is fully covered by Al2O3, the growth rate GPC and TMA

adsorption are lining up with the baseline parameters.

AFM measurements of the surface morphology at different ALD stages (Figs. 3.9d-f)

shows that shows that Ge surface is significantly roughened during the nucleation, meaning that

Al2O3 ALD on Ge: H surface is a non-conformal island growth. While the horizontal size of the

islands is hard to estimate due to the relatively low lateral resolution in AFM measurements, the

vertical height can be well distinguished, which is the RMS difference between the Al2O3 fully

covered surface (at 27th cycle) and the starting surface (Ge:H), namely 0.626- 0.294= 0.232 nm,

with a max peak-to-valley difference of ~2.3 nm. Due to this non-conformity, the Al2O3 thickness

102

detected by in-situ SE can be better interpreted as an average of Al2O3 islands and the in-between

voids.

The Al2O3 ALD nucleation delay on Ge: H surface can be supported by our ReaxFF-

nudged elastic band (ReaxFF-NEB) calculations. In Fig. 3.10a, the reaction profile indicates that a

fully hydrogenated Ge surface is chemically inert for TMA precursor due to the high reaction

barrier (Ea~1.41 eV). Also, the H2O adsorption on Ge: H surface is not energetically favorable to

activate the Ge: H surface, namely, the hydroxylation is inhibited (Fig. 3.10b). These two facts

together contribute to the initial nucleation difficulty observed in the experiments.

To figure out the mechanism for initializing TMA chemisorption, additional ReaxFF-NEB

calculations of reaction paths for H diffusion on Ge: H surface, and we found that the formation of

local dangling bonds can serve as the nucleation center. That is, though the ground state is fully

hydrogenated Ge surface, the local dangling bond can form by H-diffusion from the surface into

the interstitial sites in the Ge-sublayer by overcoming a moderate reaction barrier (Ea~0.68 eV; see

Ref. [35] for the reaction profile for H-diffusion into Ge-sublayer) which is feasible at the elevated

temperature in the real experiment (T=270 °C). To demonstrate this, we also performed MD

simulations of H diffusion into Ge-sublayer using H terminated Ge slab (60 Å × 60 Å × 10 Å) with

a vacuum environment (see Fig. 3.11). A density of 0.3 nm-2 for Ge dangling bonds was confirmed

at 800 K, corresponding to ~4.0 % of Ge atoms on the top layer of the H terminated Ge surface. If

assuming TMA will only bond to the Ge dangling bonds, the TMA adsorption site density becomes

0.3 nm-2, which is comparable to the experimental observation of 0.5 nm-2 as mentioned above.

Figs. 3.10c-e shows the ReaxFF-NEB results for the TMA+H2O nucleation on Ge-dangling

bond, which account for effects of the Ge-dangling bond on the TMA chemisorption. In figure 2c,

the TMA precursor energetically favors the local Ge-dangling bond (Ge-*) as following:

Ge-*+ Al(CH3)3 Ge-Al(CH3)3*

Ge-Al(CH3)3*+ Ge-H* Ge-Al(CH3)2*+ CH4+Ge-*

103

The asterisks denote surface species. It is interesting to note that the TMA chemisorption

simultaneously triggers the formation of an additional dangling bond at the neighboring site.

Subsequently, as expected, a H2O dose into the system only hydroxylates the TMA adsorbed site

(Fig. 3.10d, e). The adsorption preference during the above-mentioned TMA and H2O dose was

also confirmed by ReaxFF-MD simulations at 500K (see Supporting Information, TMA and H2O

adsorption on H-terminated Ge surface, for more details). At this stage, there are two available sites

where TMA precursors preferably chemisorb: (1) the newly formed Ge-dangling bond, and (2) the

hydroxylated site [Ge-Al(OH)2*], which originates from the initial local Ge-dangling bond

(Fig.3.10f). As a result, the subsequent Al2O3 ALD nucleates locally around the initial Ge-dangling

bond site, namely, island growth mode.

Two major consequences of the Al2O3 island growth mode on Ge can be expected. First,

the Al-Ge bonds formed in TMA adsorption contribute to the interface trap states in the Ge band

gap. [126] Second, one could expect a high density of boundaries in the Al2O3 layer; [30] these

boundaries act as a leakage channel in the dielectrics. [163] Both consequences imply that a direct

Al2O3 deposition on Ge: H surface yields inferior electrical characteristics of Ge-based devices, as

addressed in our previous paper. [118]

104

Figure 3.9. Al2O3 ALD on Ge: H surface. (a) In-situ SE monitoring of growth curve. (b) TMA dose rise and

extracted growth per cycle (GPC) as a function of ALD cycles; the blue dash line is the base growth rate for

baseline Al2O3 ALD (~0.86 Å/cycle). (c) GPC: TMA adsorption ratio as a function of ALD cycles; the blue

line is the baseline ratio of ~0.404. AFM images for different A2lO3 ALD stages: (d) Ge: H surface before

ALD, (e) 15 ALD cycles, and (f) 27 ALD cycles, respectively. Surface roughness is represented by root-

mean-square (RMS) of height. Results published in Ref. [35].

105

Figure 3.10. Reaction profiles obtained by the ReaxFF-nudged elastic band scheme for Al2O3 ALD on H-

terminated Ge(100). (a) TMA and (b) H2O adsorptions at H-terminated site. (c) TMA adsorption on a Ge-

dangling bond. (d) H2O removes one CH4 group from Ge-Al(CH3)2* site. (e) H2O removes CH4 group from

Ge-Al(CH3)(OH)* site. (f) TMA adsorption at Ge-Al(OH)2* site. The insets correspond to the atomic

structures for the stage at the solid points in each profile, while the neighboring number denotes the

corresponding energy. Results published in Ref. [35].

106

Figure 3.11. Results of MD simulations at 500 K for (a) 80 TMA molecules on a H-terminated Ge(100)

surface with a single Ge-dangling bond; (b) 100 H2O molecules on the H-terminated Ge(100) surface with a

Ge-Al(CH3)2* site. ReaxFF-MD results confirm that TMA and H2O molecules preferably adsorb on the Ge-

dangling and Ge-Al(CH3)2* sites, respectively, rather than H-terminated Ge sites. Results published in

Ref. [35].

107

3.3.3 Al2O3 ALD on oxidized Ge surface

Oxidized Ge surfaces (GeOx/Ge) were obtained by oxidizing Ge:H surfaces prepared by

H2 plasma at 270 °C. Three oxidation modes were evaluated: (i) O2 gas, (ii) continuous O2-plasma

(O*), and (iii) O2 plasma pulses, as shown in Fig. 3.12. Exposing the Ge surface to O2 resulted in a

slow and limited formation of sub-monolayer GeOx (~1 Å) (Fig. 3.12a), while the continuous O2

plasma (Fig. 3.12b) caused rapid Ge oxide formation (~0.7 Å/sec), not suitable for precise control

of the targeted GeOx thickness. A pulsed O2 plasma mode was used (Fig. 3.12c), allowing a precise

adjustment of GeOx thicknesses up to ~5 Å (~1.5 monolayer). The self-limiting behavior was

interpreted as a result of the mild oxidation conditions enabled by the short pulses, suggesting that

the top-most monolayer of GeOx acted as an oxygen protection layer. So the short O2 plasma pulse

mode was used as the optimal preparation process of GeOx/Ge surfaces.

Figure 3.13a shows the in-situ SE monitoring of Al2O3 ALD on GeOx(5 Å)/Ge(100)

surface. The ALD nucleation behavior (Figure 3.13a) is also nonlinear but with distinctively

different characteristics of TMA adsorption and GPC compared to Ge:H surfaces. The nucleation

can be distinguished into four regions. During the first ALD cycle (region I), TMA adsorption is

pronounced, ~45% higher than the baseline; this is attributed to the high density of TMA

chemisorption sites on GeOx. The resultant GPC/TMA-adsorption ratio is significantly higher than

the baseline (Fig. 3.13c) indicating that H2O is not the only oxidizing agent.

In the following five cycles (region II in Fig. 3.13b), the TMA adsorption level is moderate

and relatively constant (~60% of the baseline), but the GPC is close to zero, resulting in a large

deviation of GPC/TMA-adsorption ratio from the baseline (Fig. 3.13c). Even with a sizeable TMA

adsorption, a small Al2O3 overlayer is formed due to intermixing, whereby Al atoms penetrate into

GeOx densifying the oxide. In region III, TMA adsorption level and growth rate gradually approach

the baseline. After ALD cycle n=9 in region IV, all parameters remain relatively constant with

108

values similar to the baseline. At this point, the Al2O3 overlayer coalesces and conventional ALD

Al2O3 process becomes the dominant reaction mechanism

Ex-situ post-deposition characterizations were performed to figure out the mechanism of

Al2O3 ALD on GeOx/Ge. AFM measurement (inset in Fig. 3.13a) reveals that no roughening

occurred during Al2O3 ALD on GeOx/Ge, and therefore excludes the possibility of island growth

mode. The chemical status was examined by X-ray photoelectron spectroscopy (XPS)

measurements (Fig. 3.13d); the samples with Al2O3 ALD on Ge: H and GeOx(3 Å)/Ge(100) serve

as control groups. The deconvoluted peaks for different Ge valent states are shown in Table 3.2.

The first important feature of the XPS results is that the Ge+1 and Ge+2 intensities dominate and

stay relatively constant (<1% change) regardless to the starting surface for Al2O3 ALD. The

equivalent GeOx thickness with Ge+1 and Ge+2 is estimated to be ~4 Å. This sub-oxide layer could

be understood as a transition region from Ge bulk to the oxide overlayer. [126] Second, for Al2O3

ALD on GeOx/Ge surface, Ge+3 and Ge+4 peaks are almost absent, with a sub-angstrom equivalent

thickness. So Ge+3 and Ge+4 have been scavenged during Al2O3 ALD; we call this a TMA self-

clean effect of GeOx. This observation is similar to a previous result reported on Ge by Delabie et

al, [164] and has also been well recognized for TMA [165,166] and Tetrakis (Ethyl-Methyl-

Amino) Hafnium (TEMAH) [167] on III-V semiconductors. The self-clean effect is primarily

driven by thermo-dynamics, wherein Al-O bond is much stronger than Ge-O (see Ellingham

diagram in Fig. 3.14). The electrical consequence of the self-clean effect will be discussed later.

In order to further explore the nucleation mechanism, ReaxFF MD simulations were also

performed for Al2O3 ALD on GeOx/Ge(100), as shown in Fig. 3.15a. The adsorption in the first

TMA is obvious and non-local, creating a TMA overlayer directly bonded to O atoms in GeOx; this

is because the TMA adsorption on GeOx surface has no reaction barrier and is highly exothermic

(-3.02eV; see Fig. 3.15b). Note that part of –CH3 groups already desorb from the surface and form

methane (CH4), and ethane (H3C-CH3) (as highlighted in Fig. 3.15a), so the TMA is partially

109

oxidized by GeOx during adsorption, or in other words, the GeOx is partially reduced, as revealed

in figure 4c. This well explains the abnormal high growth/dose ratio observed in the 1st Al2O3 ALD

cycle on GeOx/Ge (region-I in Fig. 3.13b). Subsequently, the 2nd TMA dose results in little

adsorption and slight change in the chemical state of GeOx, meaning that the surface is somewhat

saturated. Note that the more Al atoms diffuse into GeOx layer due to a longer time frame. The

subsequent H2O dose hydroxylates the adsorbed TMA; during this stage, the Al diffusion into GeOx

layer continues, leading to an intermixture of [AlOx/GeOx]. The statistics of the simulation results

show that the O-atoms in AlOx mainly originate from GeOx (Fig. 3.15d). From this scenario, the

counter-intuitive observation of decent TMA dose but low growth in region-II of Fig. 3.13b can be

understood as a result of intermixing effect, wherein a strong TMA adsorption occurs but little

thickness is accumulated. The following region-III in figure 3b is when the intermixing saturates

and transits to the coalesced Al2O3 ALD in region-IV. Based on the results of ReaxFF MD

simulations, we confirm that GeOx surface is chemically active for TMA+H2O nucleation and

therefore leads to a conformal ALD, in accordance with the above-mentioned AFM study.

In order to evaluate the interface properties, more ReaxFF simulations were performed

about the [AlOx/GeOx]/Ge structure. Fig. 3.16 shows ReaxFF-MD simulations for the thermo-

stability of GeOx/Ge (100) and [Al2O3/GeOx]/Ge (100) interfaces. In the case of GeOx/Ge(100), a

large portion of oxygen atoms diffuses from GeOx layer into Ge subsurface, indicating the thermo-

dynamically unstable nature of GeOx/Ge interface. In contrast, the oxygen diffusion is significantly

suppressed by the existence of Al2O3 layer on GeOx/Ge surface, so the dielectric/Ge interface is

stabilized. Similar stabilization effect has been observed by incorporating other metal or nitrogen

atom into GeO2. [10] Since [Al2O3/GeOx] can form a stable and high quality interface with Ge, we

can use it as the interlayer between the high-k and Ge for device applications. More details about

the fabrication and electrical performance will be discussed later.

110

Table 3.2. De-convoluted peak areas of XPS measurements for samples after Al2O3 ALD on Ge: H, GeOx(3

Å)/Ge(100), and GeOx(5 Å)/Ge(100), respectively. The peak area is normalized by the area sum of Ge0 3d3/2

and Ge0 3d5/2. Results published in Ref. [35].

Starting Surface Ge1+ Ge2+ Ge3+ Ge4+

Ge: H 8.4% 5.1% 0.0% 0.0%

GeOx(3Å)/Ge 8.6% 4.6% 0.4% 0.8%

GeOx(5Å)/Ge 8.7% 4.2% 2.4% 1.2%

Figure 3.12. In-situ SE monitoring for three GeOx growth modes at 270 C: (a) molecular O2 (pO2≈33

mTorr/2min), (b) continuous O*-plasma (8 sec), and (c) sequence of O*-plasma pulses (1.75 sec/pulse). Note

the shorter time scale in (b). Results published in Ref. [118].

111

Figure 3.13. (a) In-situ SE monitoring for Al2O3 ALD on GeOx(5 Å)/Ge(100). Inset is the AFM image right

after Al2O3 ALD; RMS=0.285 nm. (b) TMA dose rise and extracted GPC as a function of ALD cycles; the

ALD base GPC (blue dashi line) is ~0.86 Å/cycle. (c) Al2O3-growth/TMA-dose ratio extracted from

ellipsometry; the blue line of 0.4 is the typical value for Al2O3 ALD. (d) XPS measurements (Ge 3d at 90 °

take-off angle) after 1 nm Al2O3 ALD on different starting surfaces, Ge: H, GeOx(3 Å)/Ge(100), and GeOx(5

Å)/Ge(100), respectively. Inset is the XPS measurements at 15 ° take-off angle. Results published in

Ref. [35].

112

Figure 3.14. Ellingham calculations for As2O3, GeO2, Ga2O3 and Al2O3, respectively. Results published in

Ref. [35].

Figure 3.15. ReaxFF-MD simulations of Al2O3 ALD using TMA+H2O dose on GeOx(5Å)/Ge(100) surface

at 1000 K. (a) Final snapshots of MD simulations for the sequential precursor dose; the MD-NVT simulations

were performed for each dose up to 500 ps; the highlighted molecules 1 and 2 are byproducts of methane

(CH4) and ethane (H3C-CH3), respectively. (b) ReaxFF-NEB scheme for TMA adsorption on GeOx/Ge

surface. (c) Number of oxygen atoms coordinated to Ge during Al2O3 ALD. (d) Oxygen origin of Al2O3

grown on GeOx/Ge surface after Al2O3 ALD. Results published in Ref. [35].

113

Figure 3.16. ReaxFF-MD simulations for oxygen diffusion from GeOx into underlying Ge subsurface for

GeOx/Ge(100) (red) and [Al2O3/GeOx]/Ge(100) (blue) interfaces, respectively. To accelerate the oxygen

diffusion effect within a limited time frame (2000 ps), a high temperature (800 K) was assigned to oxygen

atoms, while the other systems were kept at 300 K with the NVT-ensemble. The portion of oxygen diffusion

was normalized by the total number of oxygen atoms in GeOx layer.

114

3.4. Electrical characterization of Ge MOSCap devices

From the discussions above, the [Al2O3/GeOx] intermixing layer can serve as an electrically

superior interlayer between high-k dielectric and Ge surface. In the real application in device

fabrications, the other electrical performances should also be taken into account, including the

capacitance density or EOT, gate leakage current, and device reliability. So instead of material

engineering, the next step is to engineer the structure of devices. For this purpose, Ge MOSCap

devices can well serve for this purpose, with a structure shown in Fig. 3.17. It is namely a trilayer

dielectric stack using [Al2O3/GeOx] as the interlayer and HfO2 as the high-k dielectric. While the

preparation of a pristine Ge surface has been well addressed in Section 3.1, the fabrication of the

depositing dielectrics and the resultant electrical characteristics will be addressed in this section.

Fig 3.18 shows an overview of the p-Ge MOSCAPs process flow using HfO2/Al2O3/GeOx

gate stacks. Details of the optimization of individual process steps will be discussed further below.

First, p-Ge(100) substrates (Ga-doped, resistivity=1.0-5.0 /cm, by Umicore Electro-Optic

Materials) were degreased with acetone, isopropyl-alcohol and de-ionized water rinses. The

substrates were immediately transferred into the load-lock of ALD system (Kurt J. Lesker

Company ALD-150LX). The following process steps in ALD chamber were monitored by in-situ

SE (M-2000U, J. A. Woollam), which monitors the dielectric function of samples, providing real-

time information of surface modification like deposition and etching. Figs. 3.18a-d show an

example of in-situ SE monitoring the trilayer gate stack development (270 C, background pAr=1.2-

1.5 Torr). The residual native GeOx was effectively removed by in-situ RF atomic hydrogen (H*)

plasma (100 W, H2: Ar=3: 117 sccm, 30 sec), as shown in Figs. 3.18a. The GeOx passivation layer

was grown by oxygen (O*) plasma pulses (125 W, 1.75 sec/pulse, O2:Ar=3:117 sccm) Figs. 3.17b.

The Al2O3 layer was deposited by thermal ALD with tri-methyl-aluminum (TMA) and H2O Figs.

3.18b. The HfO2 layer was deposited by thermal ALD using tetrakis-dimethyl-amino-hafnium

115

(TDMAH) and H2O Figs. 3.18c. Each layer thickness in the trilayer gate stacks was precisely

controlled by in-situ SE. 60nm Ni was thermally evaporated as the gate metal on the samples, which

were then annealed in forming gas (FGA, H2:Ar=20: 1050 sccm, 330 C/10 min).

Equipped with the precise thickness information for each dielectric layer from in-situ SE,

we investigated their effects on the electrical properties of HfO2/Al2O3/GeOx MOSCAPs. The role

of GeOx was first investigated using p-Ge MOSCAPs of HfO2(24 Å)/Al2O3(10 Å)/GeOx stack with

various GeOx thicknesses. The C-V characteristics in Fig. 3.18 show that the sample with ~5 Å

GeOx passivation showed a small frequency dispersion in accumulation (C/Cmax=2.3%),

indicating an improved D-S interface quality (Dit@Ev≈5.11012 cm-2eV-1), while an insufficient

GeOx passivation (0.0 and 2.5 Å) resulted in not only a larger frequency dispersion at accumulation,

but also a higher gate leakage (JG) Fig. 3.19. Thus, we concluded that a minimum GeOx thickness

of ~5 Å was necessary to create a high-quality interface and to maintain a low gate leakage. The

physical mechanism behind this is attributed to that the GeOx passivation increased the conformity

of Al2O3 ALD nucleation (AFM results not shown here), and therefore formed a better quality of

the dielectric-Ge interface and reduced defects in the dielectrics.

The effectiveness of the Al2O3 to stabilize the interface was also studied by electrical

measurement. Figure 3.20 shows the C-V characteristics for HfO2(24 Å)/Al2O3/GeOx(5 Å) gate

stacks with various Al2O3 thicknesses or ALD cycles. As expected, a direct contact between HfO2

and GeOx resulted in an inferior dielectric/Ge interface (Fig. 3.20a); the HfO2/GeOx intermixing

may result in Hf-Ge bond formation [126], contributing to the interface and border trap

states [27,126]. In contrast, a use of 9 cycles ~5 Å Al2O3 deposition well preserved the GeOx/Ge

interface quality (Fig. 3.20c), while no further improvement was found for thicker Al2O3 (~10 Å

for 15 cycles in Fig. 3.20d). Another consequence of introducing Al2O3 stabilization is the

suppression of gate leakage (JG-V in Fig. 3.20e), attributed to the suppression of HfO2/GeOx

intermixing and therefore the reduction of electrically active defects in HfO2 [27,161].

116

The MOSCap device performances were further evaluated by the following electrical

parameters: capacitance density at accumulation (Cmax), C-V hysteresis (VFB), leakage current (JG)

at 1V overdrive, and interface trap density (Dit) at valence band top (Ev) (see Figs. 21a-e). Trends

in the electrical characteristics correlate well with the 4-stage nucleation scenario of Al2O3 ALD on

GeOx/Ge. Cmax directly reflects the capacitance density of the gate oxide and is determined as Cmax=

ox/tox, where ox and tox are the dielectric constant and thickness of the gate oxide layer, respectively.

Instead of n=0 (no Al2O3), Cmax is the highest in n=9 device. An insufficient Al2O3 ALD (n=0, 4)

will result in a direct contact between the subsequently deposited HfO2 and the unstable GeOx/Ge

interface, which has been reported to form volatile GeO diffusing into HfO2 and to oxidize Ge

sublayers via a two-step reaction mechanism. [10] The additional GeOx formed increases the

overall tox and lowers the effective ox of the gate oxide due to a lower ox of GeOx (~4.5), compared

to Al2O3 (~9) and HfO2 (~18), leading to Cmax reduction. For n=9 cycles, intermixing of

[Al2O3/GeOx] is maximized and the Al2O3 overlayer is coalesced, preventing a direct contact

between GeOx/Ge and HfO2. Due to the intermixing, Al2O3 ALD results in a small tox increase and

the resultant densification of the dielectric slightly increases the effective ox; [168] both factors

contribute to a higher Cmax for n=9 compared to n=0, 4. An overgrowth of Al2O3 ALD (n=15) only

causes additional Al2O3 thickness to the overall tox, leading to a decreased Cmax compared to n=9.

Figure 3.17. A structural schematic for a MOSCap device using [Al2O3/GeOx] as the interlayer between high-

k and Ge.

117

Figure 3.18. In-situ SE monitoring of the fabrication process: (a) H*-plasma clean of native GeOx, (b) Ge

passivation by pulsed O*-plasma, (c) Al2O3 thermal ALD, and (d) HfO2 thermal ALD, with all oxides

described as Cauchy model. Results published in Ref. [118].

Figure 3.19. C-V characteristics of HfO2(24 Å)/Al2O3(10 Å)/GeOx/p-Ge(100) MOSCAPs with varying GeOx

thickness: (a) 0.0, (b) 2.5, and (c) 5.0 Å. (d) The effect of GeOx thickness on MOSCAP performance. VFB is

the flat-band voltage, and EV is the valence band edge. Dit@Ev were extracted using conductance

method. [169]

118

Figure 3.20. C-V of HfO2(24 Å)/Al2O3/GeOx(5 Å)/p-Ge MOSCAPs with Al2O3 of (a) 0.0 Å (0 cycle), (b) 2.2

Å (4 cycles), (c) 4.9 Å (9cycles), and (d) 10.1 Å (15 cycle); (e) the corresponding gate leakages.

Figure 3.21. Electrical characterization of Ni(60 nm)/HfO2(24 Å)/[Al2O3(varied cycles)/GeOx(5 Å)]/p-

Ge(100) MOSCaps. (a) Capacitance-voltage (C-V) and gate leakage-voltage (JG-V) for the device with 9

Al2O3-ALD cycles. The Al2O3-ALD dependence for (b) maximum capacitance (Cmax) for C-V at f=1.5 MHz,

(c) gate leakage current (JG) at over-driving V=VFB-1V (VFB is the flat-band voltage), (d) C-V hysteresis (flat-

band voltage shift, VFB), (e) interface trap density (Dit) at valence band edge (E=EV). The Dit was obtained

by analyzing frequency-dependent C-V characteristics based on conductance method. [169]

119

3.5 Summary

A pristine Ge surface that is free of oxygen/organics and atomically flat has been obtained

by combining an ex-situ wet-chemical clean and H2 plasma treatment. The H2 plasma clean process

was found temperature dependent, while high temperatures (≥ 270 °C) are preferred for obtaining

a smooth Ge surface. The Al2O3 ALD nucleation kinetics on hydrogenated and oxidized Ge

surfaces have been investigated in atomic scale by combining in-situ spectroscopic ellipsometry

monitoring and the ReaxFF-MD simulations; the resultant dielectric/semiconductor interface

properties were discussed. The Al2O3 ALD on hydrogenated Ge surface showed an island growth

mode, resulting in defective dielectric/semiconductor interface. A TMA self-clean effect of GeOx

was discovered during the Al2O3 ALD on GeOx/Ge. Both experiments and simulations showed that

the self-clean effect can result in a chemical reduction of GeOx and simultaneously an intermixing

between Al2O3 and GeOx. The [Al2O3/GeOx] intermixing layer was found electrically superior on

Ge; the resultant Ge-MOSCap devices showed improved C-V characteristics and a reduced gate

leakage current. These superior properties were well addressed by the ReaxFF-MD simulations,

which showed that the incorporation of Al2O3 into GeOx can effectively suppress the O-diffusion

from GeOx to Ge subsurface, and therefore stabilize the dielectric/Ge interface properties and

reduce the defects in the dielectrics. By using the [Al2O3/GeOx] intermixing layer as the interlayer

between the high-k and Ge, Ge MOSCap devices with improved electrical performances have been

achieved, with a highly scaled sub-nm EOT (~0.85 nm) and low gate leakage (JG=0.15 mA/cm2 at

VFB-1V) in p-Ge MOSCAP using the trilayer gate stack of HfO2(24 Å)/Al2O3(5 Å)/GeOx(5 Å). The

results are compared to the other reports on Ge MOSCap devices, as shown in Fig. 3.22.

120

Figure 3.22. Gate leakage (JG@VFB-1V) vs. EOT benchmark of Ge MOSCAPs using HfO2(24

Å)/Al2O3/GeOx(5 Å) gate stacks with 0, 4, 9, and 15 cycles of ALD Al2O3 (indicated by dash arrow). EOT

is calculated using Cmax@1.5 MHz.

Chapter 4

Atomic Layer Deposition Synthesis of 1T-TaS2

4.1 Introduction

A key challenge of applying 1T-TaS2 as the channels of CMOS devices is preparing wafer

scale thin films so that it is easier for the semiconductor industries to accommodate this new

material into the well-developed VLSI technologies. From the manufacturing point of view, a

mechanical exfoliation to obtain ultrathin layers is inefficient and unreliable, and cannot meet the

requirement of large scale integration. So a direct growth of wafer scale is becoming important.

For this purpose, atomic layer deposition (ALD) provides a possible solution. Compared to other

methods like mechanical exfoliation [54] and chemical vapor deposition (CVD), [74] ALD has not

been widely studied yet.

This chapter will discuss about a direct ALD synthesis of 1T-TaS2 thin films using TaCl5

and H2S as the precursor in a home-built ALD system. Various ALD conditions have been used to

improve the grown material qualities, including the substrate effect, precursor dosage, and growth

temperature. The physical properties of the as-obtained thin film materials were evaluated primarily

using atomic force microscopy (AFM) and Raman spectroscopy. A possible nucleation mechanism

was proposed for explaining the ALD kinetics.

Also, this chapter includes some experimental results about developing a process for ALD

synthesis of ferroelectric HfO2, which is intended to be a gate dielectric required for the future

fabrication of 1T-TaS2 based ferroelectric field-effect transistors (FeFET). The material properties

were evaluated by X-ray photoelectron spectroscopy and electrical polarization measurements.

122

4.2 ALD Growth of 1T-TaS2

4.2.1 Experimental details

A home-built ALD system was used to grow 1T-TaS2 thin films. The details about the

hardware setup has been introduced in Section 2.1.3. An in-situ quartz crystal microbalance (QCM)

metrology has been used to real-time monitor the growth kinetics and optimize the growth

parameters. The foundation of QCM is the high-Q shear-mode piezoelectric effect. The deposited

areal mass results in a small shift of the resonant frequency, as shown in Fig. 4.1. In this thesis, the

density of the material deposited was not calibrated, so instead of directly obtaining thickness

information, the ALD growth monitoring was done through checking the mass change, or even

more conveniently, the frequency shift f.

In this thesis, Si substrates with surface native oxide (~2 nm SiO2) and c-plane sapphire (c-

Al2O3) have been used as the substrates with a size of 10×10 mm2. While the Si substrates were

simply cleaned by acetone/IPA/DI-H2O sonication, the cleaning process for the sapphire substrates

involved more treatments (see Fig. 4.2); in particular, the Nanostrip rinse was intended to

thoroughly remove the organic contaminations. The cleaned substrates were immediately

transferred into the ALD chamber, which was then pumped down to the baseline pressure of ~50

mTorr by a roughing pump. Thereafter, an Ar gas flow was established to sustain a pressure of

~1.03 Torr (more details can be found in Section 2.1.3) and the power of all heaters was switched

on. The ALD growth was performed after all the heating zones of the ALD system were stabilized

in temperature.

123

Figure 4.1. The working principle of QCM for monitoring growth processes. [170]

Figure 4.2. The cleaning process for cleaning a sapphire substrate.

4.2.2 ALD growth at low-mid temperatures

ALD growth of 1T-TaS2 was firstly performed using a regular heater which allowed a low-

mid temperature range (Tmax≤ 380 °C). In-situ QCM was used to monitor the real-time thickness of

the ALD growth, as represented by the resonant frequency shift of QCM (see Fig. 4.3a). The ALD

growth rate shows an obvious temperature dependence. Fig. 3.4b shows the deposition details for

the ALD growth, which is an enlargement of the growth segment of the 380 °C growth in Fig. 3.4a.

The growth parameters for 380 °C growth can be found in Table 4.1. Each ALD cycle starts with a

TaCl5 dose, which results in a frequency shift of f 4.5 Hz in the QCM sensor and a dose pressure

of ~14 mTorr. The corresponding TaCl5 mass gain is calculated to ~55.2 ng/cm2, or an areal density

124

of 9.28×1013 cm-2 (0.928 nm-2). This is much smaller than the 3.05 nm-2 adsorption level observed

by in-situ ellipsometry in the Al2O3 ALD. After purging the residual unreacted TaCl5, a double

dose of H2S precursor was used for an overdose, due to the relatively low sticking coefficient of

H2S compared to H2O. In-between the double dose, a shorter purge time (20 sec) was used so that

there was always H2S pressure present inside the chamber (see Fig. 4.3c). After the 2nd H2S, a long

purge (40 sec) completely removed the residual unreacted H2S. Compared to the TaCl5 dose, H2S

dose did not result in mass gain but a slight decrease. This is not surprising considering the proposed

reaction mechanism:

TaCl5+ 2H2S TaS2 + 4HCl + 1/2Cl2. Eq. 4.1

In this equation, replacing five Cl atoms with two S atoms is overall losing mass. As shown in Fig.

4.3d, the growth rate is monotonically increasing with elevating temperatures within 220-380 °C,

and does not show any saturation or stop of increasing. The mechanism behind this temperature

dependence is attributed to the thermodynamics of the reaction. Fig. 4.4 shows the free energy as

a function of temperature for the reaction of Eq. 4.1. At low temperatures, positive free energy gain

(G) implies that the reaction is unfavorable. A high temperature (>453 °C) is required for a

thermodynamically favored reaction, which was not achievable with a regular substrate heater

(≤380 °C).

Since the 380 °C growth yielded the fast growth rate, it became more worthwhile for further

characterization. Fig. 4.5 shows the surface morphology after different ALD cycles on sapphire

substrates. The surface morphology indicates an obvious nucleation mode of island growth. After

250 ALD cycles, the islands are small in lateral size (<40 nm) and completely randomly scattered

on the surface, and the resultant surface coverage is low. With more ALD cycles, the islands

become larger, resulting in higher surface coverage, but the islands are irregular in shapes,

indicating an amorphous characteristic in the crystal structure. The surface coverage as a function

125

of the ALD cycles is shown in Fig. 4.5b. Even with 1000 ALD cycles (took ~27 hrs, not including

the heating and cooling time), the surface coverage is only ~61 %. Another interesting observation

is that with increasing ALD cycle number, the island height stays relatively unchanged at ~2 nm,

corresponding to ~3 monolayers of 1T-TaS2 (see the AFM line scans in Fig. 4.5a). The preference

of lateral nucleation and island growth mode has been commonly observed in the CVD growths of

TMD materials, as mentioned above. [62,171,172] This effect can be understood as a result of

strong in-plane anisotropy of TMD materials. In the out-of-plane direction, the atoms are fully

coordinated with no dangling bond available for nucleation, while the dangling bonds on the edge

of the material allows for further lateral growth. A further examination by a room temperature

Raman spectroscopy confirmed a relatively weak vibrational mode at 71.1 cm-1, which is close to

the previous report for the primary peak of 1T-TaS2 at 310 K (peaking at ~70 cm-1). Unfortunately,

due to the limitation of facility, we were unable to verify the low temperature characteristics of

Raman spectrum.

126

Figure 4.3. (a) Real-time thickness measured by in-situ QCM for ALD growth of 1T-TaS2 at low-mid

temperatures (220-380 °C). (b) An enlargement for the growth at 380 °C indicated by the black box in Figure

(a). (c) 100-cycle growth represented by QCM signals as a function of temperature. (d) Temperature

dependence for QCM signal of 100 ALD cycles.

Figure 4.4. A thermodynamic calculation of the ALD reaction. The critical temperature is estimated to be

453 °C.

127

Figure 4.5. AFM measurements for TaS2 ALD growth at T=380 °C. (a) Surface morphology at difference

ALD cycles. A line scan (white dash) is shown as an inset in each AFM image. (b) The surface coverage as

a function of ALD cycle, extracted by image processing of the AFM morphology using ImageJ software. (c)

Raman spectrum for the sample with 1000 ALD cycles.

128

4.2.3 ALD growth at high temperatures

From the low-mid temperature results, the direct ALD growth of 1T-TaS2 requires a high

temperature (>453 °C) to obtain the desired material qualities. For this purpose, the substrate heater

of the ALD system has been upgraded, using a new design as shown in Section 2.1.3. The new

heater allows for a high temperature up to 480 °C with a power input of ~70 %. An even higher

temperature application requires more power input and would significantly shorten the lifetime of

the heater (actually four heaters had been burned out before realizing this issue). Note that at high

temperatures, QCM basically loses its functionality due to: (i) the maximum temperature allowed

for the internal QCM heater is limited to (<400 °C), so the QCM signal cannot represent the reaction

process occurring on the sample surface due to the large temperature difference in-between; (ii)

even if upgrading the heater of QCM for the high temperature application, the Q-factor of QCM

starts significantly decreasing at elevated temperatures (>309 °C), [92] so one can expect that the

QCM signal will become extremely noisy and therefore unreliable at a high temperature like 480

°C.

Fig. 4.6 shows the surface morphology after 500 cycles of ALD growth using TaCl5+ H2S

at T=480 °C. Scattered bar-like islands can be found, with a lateral size of ~40 nm × 170 nm. The

line scan in the inset shows an island height of ~17.5 nm, while the ~33 nm height may be a double

stack. This quasi-1D characteristic cannot from an amorphous material, so indicates a different

crystalline structure. Previous reports showed that in Tantalum-Sulfur family, there is another

member, TaS3, which is also quasi-1D crystal structured material. [47,173,174] So the bar-shape

material grown by high temperature ALD is possibly TaS3. If this assumption stands true, it

indicates that the ALD growth is sulfur rich or in other words, Ta deficient. A Ta deficiency is not

surprising because regarding the high temperature of 480 °C used during the growth, a strong

129

thermal desorption of TaCl5 can be expected, which has been well identified generally in other

ALD growths. [31]

To address the possible desorption issue of TaCl5, before the TaCl5+ H2S ALD, an ultrathin

Ta2O5 seed layer was prepared by 10 ALD cycles using TaCl5 and H2O as the precursors at 480 °C.

Thereafter, 500 cycles of TaCl5+ H2S ALD were performed. The surface morphology of the as-

obtained sample is shown in Fig. 4.7. The first important characteristic is the observation of

hexagonal shapes in conjunction with a few triangles. The lateral size of these shapes varies from

~50 nm to ~300 nm, while the AFM line scans reveal multiple thicknesses, ranging from ~4.3 nm

to ~10.7 nm. Regarding to the aspect ratio between the lateral and vertical size, the as-grown

material is more like 2D. The overall surface coverage estimated by ImageJ processing gives ~74

%. Another characteristic is the observation of the string like feature, more prominent in the lower

half of the AFM image in Fig. 4.7. These string like features are different from the above-mentioned

quasi-1D shapes, and are more like nanowires.

Fig. 4.8 shows the Raman spectrum of the sample with 500 cycles of TaCl5+H2S ALD on

a Ta2O5/sapphire substrate. The 1T-TaS2-like peak around 70 cm-1 is still well identified, and shows

a much stronger intensity compared to the 380 °C ALD result, indicating an increase of the

crystallinity and/or the material volume ratio at the sample surface. Besides 1T-TaS2-like peak and

those sharp peaks indicated by blue arrows, there are another two unidentified characteristics in

Fig. 4.8, represented by open circle and rhombus, respectively. The two broad and low intensity

peaks indicated by rhombus locate around ~1300 and ~1600 cm-1, respectively. These wavenumber

shifts are close to the previous reports about Raman spectra for carbon nanotubes, 1352 and 1580

cm-1. [175] This correlation to carbon nanotubes may explain the strange nanowire like features

observation in AFM image of Fig. 4.7. The second unidentified characteristic is the high intensity

and broad peak (or maybe a few peaks) indicated by a circle around 300 cm-1. Since this

characteristic is so broad, it is difficult to find the peak position. To address this issue, a post

130

deposition anneal in a H2S/Ar atmosphere was performed in order to improve the crystallinity of

all the possible materials in the sample, and therefore yield shaper Raman peaks. The annealing

was performed using a quartz tube furnace. Before elevating the temperature, the furnace chamber

was pump and then flush with Ar gas for three times, so as to remove the residual air inside the

chamber. Then a mixed H2S/Ar flow was established to sustain a 1 atm. pressure (H2S partial

pressure ~100 mTorr) and the chamber was heated up. The steady stay at 900 °C was 20 min, and

thereafter the sample was cooled down inside the chamber with only flowing Ar. Before the

annealing, the ALD grown film on the substrate appeared black. But the high temperature anneal

seems removed most of the material on the substrate, leaving a sapphire like sample afterwards.

This was confirmed by the Raman measurement after anneal (Fig. 4.9b). The characteristic Raman

peak for 1T-TaS2 is significantly weakened, and the carbon nanotube peaks are almost gone. The

unidentified peak around 300 cm-1 now becomes easier to distinguish, which splits into three peaks

at 225, 289, and 337 cm-1, respectively. By comparing these peaks with a previous report, their

structure looks similar to that of TaS3. [176] So from this observation, we can say that the broad

“circle” peak may originate from TaS3, meaning a coexistence of 1T-TaS2 and TaS3.

In summary, 1T-TaS2 thin films has been synthesized using ALD approach. However, the

existence of impurity phase TaS3 and the issue of relatively small crystal sizes still need to be

addressed by further optimizing the ALD process.

131

Figure 4.6. AFM image after 500 cycles of TaCl5+ H2S ALD growth on a c-sapphire substrate.

Figure 4.7. AFM image after 500 cycles of TaCl5+ H2S ALD growth on a Ta2O5/c-sapphire substrate.

132

Figure 4.8. Raman spectrum for the sample with 500 cycles of TaCl5+H2S ALD on a Ta2O5/sapphire

substrate.

Figure 4.9. (a) Setup for H2S annealing the sample with 500 cycles of TaCl5+H2S ALD on a Ta2O5/sapphire

substrate. The sample photo on the left is before anneal and the one on the right is after anneal. (b) The Raman

spectrum of the sample after anneal. (c) Raman spectrum for TaS3. [176]

133

4.2.4 Synthesis of ferroelectric HfO2

As mentioned above, the application of 1T-TaS2 as channel material in a transistor requires

a modulation of the channel conductance by electrical input. Recent researches have successfully

utilized electric current to trigger the metal-insulator transition (MIT). [177,178] Figs. 4.10a, b

show the switching for 10 nm and monolayer 1T-TaS2, respectively. While the results in Fig. 4.10

harness the phase transition that occurs around 200 K, Fig. 4.10b utilizes the high temperature

transition. In terms of application in VLSI, the electric current switching is not favorable because

of the large power consumption, similar to the case of bipolar transistors. So it is necessary to

develop an electric field control which is much more energy saving. As mentioned above, the large

carrier density in 1T-TaS2 (~×1021 cm-3) requires a high gate capacitance as high as ~24 C/cm2 for

just one monolayer 1T-TaS2. This high capacitance is achievable using a ferroelectric gate insulator.

To accommodate into the VLSI manufacturing, an ALD synthesis approach is used in thesis. The

material is doped HfO2 with Al.

The key of forming ferroelectric doped HfO2 is to sustain its ferroelectric phase during

crystallization, which is usually stable only at high temperatures. [59,179–197] A generally applied

approach is through mechanical straining using TiN electrodes to form a metal-insulator-metal

(MIM) structure. For example, to obtain ferroelectric Hf0.5Zr0.5O2, the MIM structure of TiN/

Hf0.5Zr0.5O2/TiN is crystallized at a high temperature (500 °C). [195] During the cooling, the

interfacial coupling between top/bottom TiN layers and the in-between oxide layer offer a

mechanical strain that suppresses the transition from a tetragonal (ferroelectric) to monoclinic (non-

ferroelectric) phase (see Fig. 4.12). Instead, an orthorhombic phase is established, whose

ferroelectricity can be awaken by electric filed. There are two reasons for using TiN: (i) TiN is

well-known for its excellent mechanical strength, which offers a strong pinning effect; (ii) TiN is

134

conducting, can be used as electrodes and does not contribute to the capacitance/polarization of the

MIM structure.

The process flow of developing ferroelectric Al-doped HfO2 is shown in Fig. 4.12. A

heavily doped p-type Si was used as the substrate, because its high conductivity (~103 S/cm)

enables it to be the bottom electrode. The substrate was firstly degreased with acetone/IPA by

sonication. A hydrofluoric acid (HF 5%) treatment removed the surface native oxide, followed by

a DI-H2O rinse. Right after the wet-chemical clean, the Si substrates were transferred into the load

lock of a magnetron sputtering chamber, which was thereafter pumped down to ~10-6 Torr.

A 10 nm titanium nitride (TiN) thin film was deposited by a reactive sputtering using a

N2(30%) /Ar mixed gas at a 5 mTorr process pressure. The TiN growth rate was determined to be

1.12 nm/min, by AFM measurement of thickness step created using a shadow mask (Fig. 4.14a).

The TiN sputtering slight roughens the surface from ~0.1 nm to 0.563 nm (Fig. 4.14b). With the

thickness measured by AFM, a spectroscopic ellipsometry (SE) measurement allows to extract the

dielectric function of the sputtered TiN, which is close to that of bulk TiN (Fig. 4.14c). The

stoichiometric ratio of Ti: N was determined to be ~0.94 (Fig. 4.14d); regarding a typical ~5 %

error of XPS measurement in chemical composition, TiN films grown by sputtering is nearly

stoichiometric.

10 nm Al doped HfO2 films were deposited by ALD at 300 C. The stoichiometry of Al: Hf

ratio was controlled by varying the ALD cycles number of Al2O3 and HfO2. The thickness was in-

situ monitored by SE and fixed to be ~10 nm for all samples (Fig. 4.15). The growth parameters of

Al2O3 and HfO2 were the same as Table 2.1 in Section 2.1.2. Due to a higher temperature (300 °C)

for synthesizing ferroelectric HfO2, the growth rates (GPCs) were also slightly different (see Table

4.2). The stoichiometric ratio of Al: Hf was controlled by varying the cycle number ratio of Al2O3:

HfO2. The Al atomic percentage (Al atm %) can be calculated as follows:

135

2 3

2 3 2

Al O cycle # 1.406 Al atm %=

Al O cycle # 1.406 HfO cycle #

. Eq. 4.2

Samples with three different stoichiometric ratios were prepared in this thesis with purge HfO2, Al:

Hf=1:20 cycles (~6.6 atm% Al), and Al: Hf=1:10 cycles (~12.3 atm% Al).

After the ALD growth, the top TiN layer was deposited by the reactive sputtering, and then

the samples were annealed in a rapid thermal anneal (RTA) chamber at 900 °C/30 sec with a

ramping rate of 20 °C/sec and quenching rate of 240 °C/sec. At the later stage of cooling (<200 °C),

the temperature was able to follow the set point of cooling prole, and therefore cooled down

naturally.

Fig. 4.16 shows the process for defining the top electrode pattern. A Ti(5 nm)/Pt(50 nm)

layer was deposited by e-beam evaporation, with the pattern defined by a shadow mask. The

electrode size of the pattern was Ø=100 m circular shape. The pattern of top TiN layer was

developed using a SC1 solution (NH4OH:H2O2:H2O=1:2:7, 60 °C). The SC1 development of TiN

was optimized using an ex-situ ellipsometry (Fig. 4.17). The optimum time was determined to be

~2 min.

The electrical performance of the MIM structures were tested using a polarization vs

electric field measurement, or PE hysteresis loop with a frequency of 10 kHz. The MIM structure

using pure HfO2 as the insulator shows a significant characteristic of leakage current (Fig. 4.18a).

This can be attributed to the formation of grain boundary by the high temperature annealing, since

the grain boundary has a higher conductance. A 6.6% Al doping results in a hysteresis loop with a

remnant polarization Pr=1.8 C/cm2 and a coercivity EC=0.86 MV/cm. At high electric field, the

sample still shows a signature of leakage, which results in a drop in the polarization. A higher Al

doping of 12.3 % can further reduce the leakage current, but the PE hysteresis characteristic

disappears, while the small hysteresis may be caused by space charge trapped in the dielectric. The

dielectric constants of all three MIM structures are shown in Fig. 4.18d. The undoped sample shows

136

a value close to that of an amorphous HfO2 dielectric, indicating that the thermal anneal does not

improve the polarizability of pure HfO2. The dielectric constant is almost doubled (~39.8) for the

6.6% doped sample, which shows the PE hysteresis behavior (Fig. 4.18b), while a higher Al doping

does not increase the dielectric constant, but a small decrease. So the dielectric constant of all the

three dielectrics with different doping levels can be correlated to the onset of ferroelectric behavior.

In summary, ferroelectric HfO2 has been successfully synthesized using a low level Al

doping, while further increase in doping can suppress the leakage current but diminishes the

ferroelectricity. Due to the intrinsic limitation of Al dopant, a relatively low polarization has been

achieved. [180] To obtain better electrical performance, it is necessary to try other more promising

dopants like La, Y, and Zr. [197]

Table 4.1. Parameters for ALD growth of 1T-TaS2.

Precursor Source

Temperature (°C)

Carrier Gas Flow (sccm)

Dose Time (sec)

Dose Pressure (mTorr)

Purge Time (sec)

TaCl5 120 40 0.3 14 30

H2O 20 40 0.015 120 20

H2S 20 40 0.015 1st dose: 310 20

2nd dose: 120 40

Table 4.2. Growth rates for Al2O3 and HfO2 at 300 °C.

Material GPC (Å/cycle) Atomic density (cm-3) GPC (atm/cm2/cycle) Al: Hf ratio

Al2O3 0.885 3.54E+22 3.13E+14 1.406

HfO2 0.805 2.77E+22 2.23E+14

137

Figure 4.10. Electric current switching of MIT transition in (a) 10 nm thick 1T-TaS2, [198] and (b) monolayer

1T-TaS2. [178]

Figure 4.11. Bulk carrier density of 1T-TaS2 as a function of temperature. [199] The table on the right

compares the monolayer carrier density for various materials.

138

Figure 4.12. The mechanism of forming ferroelectric HfO2 at room temperature.

Figure 4.13. The process flow for developing Al doped HfO2 ferroelectrics.

139

Figure 4.14. Characterizations of TiN sputtering. The sputtering parameters are: CMS18-#1 sputter system

in PSU Nanofab, 300Watt DC/15Watt AC, Ar:N2= 14:4.5 sccm, 5 mTorr, 3000 sec. (a) AFM measurement

of the thickness step created by using a shadow mask during the sputtering. (b) Surface morphology of

sputtered TiN films (56.23 nm) on c-plane sapphire (RMS 0.1 nm). (c) The dielectric function extracted

from spectroscopic ellipsometry, assuming a thickness of 56.23 nm. (d) XPS depth profiling of TiN film

deposited on a p+-Si substrate.

140

Figure 4.15. In-situ SE monitoring of 300 °C ALD growth for (a) Al2O3 calibration, (b) HfO2 calibration, and

(c) Al doped HfO2 with 1:20 cycle ratio.

Figure 4.16. Process of defining electrode pattern for measuring MIM.

141

Figure 4.17. Ex-situ SE monitoring the SC1 treatment on TiN/Si samples. (a) as a function of treatment

time. (b) as a function of treatment time. (c) A schematic for SC1 treatment. (d) at E=2.5 eV as a

function of treatment time.

142

Figure 4.18. Electrical polarization as a function of electric field for MIM structures using an insulator of (a)

purge HfO2, (b) 6.6% Al doped HfO2, and (c) 12.3% Al doped HfO2. (d) Dielectric constants for the insulators

as a function of Al doping.

Chapter 5

Conclusions, Future Work and Outlook

In this thesis, the high-k deposition on Ge surface and synthesis of 1T-TaS2 thin films by

ALD approach have been addressed. For the first part, a combinatory method using wet-chemical

clean and in-situ H2 plasma has firstly been developed and optimized to obtain a pristine Ge surface.

Then starting from this clean surface, the surface chemistry dependence of the ALD nucleation has

been identified by combining in-situ probing, ex-situ metrologies and atomic scale simulations. The

understanding of the ALD mechanism enables the fabrication of a HfO2/Al2O3/GeOx trilayer gate

stack for Ge MOSCap devices, whose optimization has resulted in superior electrical performances.

For the second part, the temperature dependence of the growth kinetics has been discovered for 1T-

TaS2 ALD, and the resultant materials structures have been characterized. Also, a fabrication

process of ferroelectric HfO2 has been developed as a necessary element to be integrated in the

future field effect transistor fabrication. In particular, the following implications are highlighted:

Chapter 3 By comparing the results of the optimized Ge results discussed in this

thesis or published elsewhere, the electrical performance of Ge devices particularly their reliability

is still not lining up with that of Si. A deep thinking about this issue is necessary. For many related

phenomena, the origin is mostly pointing to the thermodynamics of the materials. A good example

is the significant difference in the interface quality between SiO2/Si and GeO2/Si. Unlike GeO2/Ge

interface which starts degrading at T>450 °C, the generation of SiO at SiO2/Si is extremely difficult

and requires a high activation temperature >1700 °C. [200] In this sense, at a process temperature

of 270 °C, the SiO2/Si interface is ~exp[(1700-450)/543]10 times more than GeO2/Ge. To

overcome this fundamental difference in thermodynamics, there can be two routes. One is to use

144

“external support” like introducing N, Al atoms into GeO2 and forming an intermixing network,

like Ge-O-Al or O-Ge-N (or Ge-O-N), which chemically strengthens the material stability. This

approach potentially inhibits the formation of gas phase GeO, but does not necessarily suppress the

Ge-O bond breaking in the dielectric and at the interface. This probably explains that rather than

healing the in-gap trap states in the semiconductor side, the intermixing [Al2O3/GeOx] interlayer is

more effective to address the border trap states in the dielectric. Another possible solution is to

directly use an oxygen free surface passivation like Ge3N4, which has resulted in even better

interface qualities. [11] However, the relatively low band gap (≤4.0 eV) [201,202] and low

dielectric constant (k≤ 6.27) [202] of Ge3N4 have limited its application as a gate insulator. Thus, a

thickness downscaling without compromising the interface quality is necessary, which still remains

an unclarified issue. In summary, in the future research of Ge transistors, the interface stabilization

will continue as one of major topics.

Another implication is about the ALD nucleation of dielectrics. As confirmed in this thesis,

a direct ALD on a H-terminated Ge surface is highly localized growth and yields a non-conformal

and defective dielectric, as indicated by the high leakage current and high density of border trap

states. Similar effect has been also confirmed on Ge using other thermodynamically stable

dielectrics like ZrO2, [18] as well as in Al2O3/Si MOSCaps. [203] In the case of Si, the interface

and dielectric properties can be effectively healed by a high temperature anneal, which however is

not favorable for Ge. One of the major successes of employing an ultrathin GeOx layer as the

surface passivation is boosting the initial nucleation of the subsequent ALD, allowing for a

boundary/pinhole free dielectric. In this sense, when developing Ge devices with new structures

made of new materials, the nonlinear initial nucleation behavior and the resultant material

properties should always be considered as the investigation of first priority, rather than simply using

a process matrix that only controls the ALD cycles or the so-called “thickness”.

145

Chapter 4 The synthesis of 1T-TaS2 generally requires a high thermal energy to

obtain high quality materials. This trend has been well reflected in the temperature dependence of

ALD growth, while low temperature growth only yielded nearly amorphous material, much better

defined crystals were observed after a high temperature ALD. The high temperature requirement

can be even better clarified by comparing the ALD results with the other high temperature results

produced by a CVD approach using very similar reaction precursors. [74] The CVD growth yielded

large crystal size (~10 m) with high crystallinity, and more importantly, well behaved electrical

characteristics. However, just like most of CVD growth of other 2D materials, the low surface

coverage remains as a challenging issue, limiting the potential batch fabrication of integrating 1T-

TaS2 in VLSI technology. So far, there are very few reports about synthesizing wafer-scale,

coalesced, and ultrathin 2D material films, [64] Thus, the synthesizing wafer-scale and high-quality

1T-TaS2 is still a non-trivial research topic in the future.

In going forward, there are a few ideas or research projects that can be imagined, as follows:

Further optimization of [Al2O3/GeOx]/Ge interface. Early trial experiments using an O2

plasma enhanced HfO2 ALD directly on a clean Ge surface has yielded a low Dit interface right

after the device fabrication, but the properties degraded quickly due to the unstable HfO2/GeO2/Ge

structure in the devices. So from this point of view, instead of growing stabilization oxide on a

GeOx/Ge surface, a direct O2 plasma enhanced Al2O3 or ZrO2 ALD on Ge is expected to create a

superior and stable dielectric/Ge interface. The ZrO2 PEALD would be more straightforward,

because the gate dielectric growth could be finished in one deposition process; also it might yield

better performance in terms of EOT scaling, due to its high dielectric constant (k~24).

Integrating ultrathin Ge3N4 as the surface passivation on Ge. Ge3N4 could be used as

the interlayer between the high-k dielectric and Ge substrate for further improvement of electrical

performance. The Ge3N4 could potentially be prepared by the in-situ plasma source using N2 as the

reactive gas. Process optimization and material characterization of Ge3N4 would be necessary. The

146

subsequent ALD nucleation mechanism should be carefully addressed using the in-situ

spectroscopic ellipsometry, which would entail the thickness evaluation and modeling of the

material optical constants. The goal is to obtain a highly scaled (EOT ≤0.5 nm) and electrically

reliable Ge devices (MOSCaps, MOSFETs, or even FinFETs).

Mechanism study of ALD growth of 1T-TaS2. The results presented in this thesis has

shown that the nucleation of 1T-TaS2 can be assisted by using a Ta2O5 seed layer. The detailed

mechanism remains not crystal-clear yet. A possible experiment that can be done to figure out the

mechanism is to use a patterned Ta2O5 layer on sapphire substrates with various sizes, separations,

geometries, and thickness as the starting surface in the TaCl5 +H2S ALD.

Improving electrical performance of ferroelectric HfO2. While the Al doping HfO2 is

intrinsically limited in the polarization performance, other dopants like La, Zr, and Y have been

reported to induce much larger polarization (>20 C/cm2). [197] An even more interesting idea is

to utilize the antiferroelectric gate insulator, which does not show the unwanted large hysteresis at

low electric fields, but still sustains a sufficiently high capacitance density at high electric

fields. [195]

Fabricating ferroelectric field effect transistors (FeFETs) using 1T-TaS2 channel.

With the 1T-TaS2 growth and high capacitance gate insulator addressed above, the FeFETs

fabrication of 1T-TaS2 becomes possible. A plausible device structure is using a back-gated thin

film transistor (TFT) design. While a direct ALD growth of gate insulator on TMD materials has

been proven difficult, the TFT structure allows for the following process: (i) firstly MIM synthesis

of ferroelectric HfO2 first, (ii) removal of top TiN to expose the gate insulator, (iii) growth of 1T-

TaS2 thin film on top of gate insulator, and (iv) growth of the protection capping layer (can be

anything without concerning EOT scaling and therefore can be very thick).

After looking backward the VLSI history, the Moore’s law has been surviving for more

than 50 years and resulted state-of-art technologies. Whenever, the technology encountered

147

bottlenecks, researchers in academia and industries always came up with genius ideas to

successfully tackle the issue. With the 5 nm node commercialized today, moving further towards 3

nm and even 2 nm technology is challenging but inevitable. Among all the possible future routes

of the technology development, integrating novel channel materials appears the most promising.

Maybe it is too arbitrary to say: “long live Moore’s law”, but the industry people generally believe

the law is not dead yet.

Reference

[1] Y. Taur and T. H. Ning, Fundamentals of Modern VLSI Devices, 2nd ed. (2009).

[2] Https://en.wikipedia.org/wiki/Bipolar_junction_transistor

[3] Http://electroiq.com/blog/2015/11/what-lies-beneath-50-years-of-enabling-moores-law/

[4] Https://arstechnica.com/gadgets/2017/06/ibm-5nm-chip/

[5] S.-H. Lo, D. A. Buchanan, Y. Taur, and W. Wang, IEEE Electron Device Lett. 18, 209 (1997).

[6] Https://web.stanford.edu/class/ee410/AdvCMOS.pdf

[7] Https://archive.cnx.org/contents/6d846a17-ee19-4663-afb7-b49448117e99@1/sspd-chapter-2-1-3-

nano-technology-era

[8] Http://archive.eetasia.com/www.eetasia.com/ART_8800667494_499489_NT_02adb57c.HTM

[9] Https://web.stanford.edu/class/ee311/NOTES/TrendsSlides.pdf

[10] Y. Kamata, Mater. Today 11, 30 (2008).

[11] S. Takagi, T. Maeda, N. Taoka, M. Nishizawa, Y. Morita, K. Ikeda, Y. Yamashita, M. Nishikawa,

H. Kumagai, R. Nakane, S. Sugahara, and N. Sugiyama, Microelectron. Eng. 84, 2314 (2007).

[12] S. K. Wang, K. Kita, C. H. Lee, T. Tabata, T. Nishimura, K. Nagashio, and A. Toriumi, J. Appl.

Phys. 108, 54104 (2010).

[13] K. Prabhakaran, F. Maeda, Y. Watanabe, and T. Ogino, Appl. Phys. Lett. 76, 2244 (2000).

[14] Y. Kamata, Y. Kamimuta, T. Ino, and A. Nishiyama, Jpn. J. Appl. Phys. 44, 2323 (2005).

[15] A. Toriumi, T. Tabata, C. Hyun Lee, T. Nishimura, K. Kita, and K. Nagashio, Microelectron. Eng.

86, 1571 (2009).

[16] S. N. a Murad, P. T. Baine, D. W. McNeill, S. J. N. Mitchell, B. M. Armstrong, M. Modreanu, G.

Hughes, and R. K. Chellappan, Solid. State. Electron. 78, 136 (2012).

[17] S. Abermann, O. Bethge, C. Henkel, and E. Bertagnolli, Appl. Phys. Lett. 94, 262904 (2009).

[18] C. O. Chui, S. Ramanathan, B. B. Triplett, P. C. McIntyre, and K. C. Saraswat, IEEE Electron Device

Lett. 23, 473 (2002).

149

[19] V. V. Afanas’ev, Y. G. Fedorenko, and A. Stesmans, Appl. Phys. Lett. 87, 32107 (2005).

[20] K. Devloo-Casier, J. Dendooven, K. F. Ludwig, G. Lekens, J. D’Haen, and C. Detavernier, Appl.

Phys. Lett. 98, 231905 (2011).

[21] Y. Oshima, Y. Sun, D. Kuzum, T. Sugawara, K. C. Saraswat, P. Pianetta, and P. C. McIntyre, J.

Electrochem. Soc. 155, G304 (2008).

[22] E. P. Gusev, H. Shang, M. Copel, M. Gribelyuk, C. D’Emic, P. Kozlowski, and T. Zabel, Appl. Phys.

Lett. 85, 2334 (2004).

[23] F. Gao, S. J. Lee, J. S. Pan, L. J. Tang, and D. L. Kwong, Appl. Phys. Lett. 86, 1 (2005).

[24] R. Zhang, T. Iwasaki, N. Taoka, M. Takenaka, and S. Takagi, J. Electrochem. Soc. 158, G178 (2011).

[25] F. G. N. W. C. X. Z. J. S. P. L. J. T. D. L. K. S. J. Whang S. J. Lee, Iedm2004 12.6.1 (2004).

[26] R. Zhang, T. Iwasaki, N. Taoka, M. Takenaka, and S. Takagi, in VLSI Technol. (VLSIT), 2011 Symp.

(IEEE, 2011), pp. 56–57.

[27] R. Zhang, P. Huang, N. Taoka, M. Takenaka, and S. Takagi, Symp. VLSI Technol. 161 (2012).

[28] T. Maeda, M. Nishizawa, Y. Morita, and S. Takagi, Appl. Phys. Lett. 90, 72911 (2007).

[29] R. Zhang, P. C. Huang, J. C. Lin, M. Takenaka, and S. Takagi, Tech. Dig. - Int. Electron Devices

Meet. IEDM 1, 371 (2012).

[30] O. Sneh, R. B. Clark-Phelps, A. R. Londergan, J. Winkler, and T. E. Seidel, Thin Solid Films 402,

248 (2002).

[31] S. M. George, Chem. Rev. 110, 111 (2010).

[32] Http://www.ioffe.ru/SVA/NSM/Semicond/, (n.d.).

[33] I. Barin, Thermochemical Data of Pure Substances, Thermochemical Data of Pure Substances

(Wiley-VCH, 1997).

[34] C. Kittel, Introduction to Solid State Physics (Wiley, 2005).

[35] Y. Zheng, S. Hong, G. Psofogiannakis, G. B. Rayner Jr, S. Datta, A. C. T. van Duin, R. Engel-

Herbert, G. B. Rayner, S. Datta, A. C. T. Van Duin, and R. Engel-Herbert, ACS Appl. Mater.

Interfaces 9, 15848 (2017).

[36] K. S. Novoselov, D. Jiang, F. Schedin, T. J. Booth, V. V Khotkevich, S. V Morozov, and A. K. Geim,

150

Proc. Natl. Acad. Sci. U. S. A. 102, 10451 (2005).

[37] K. S. Novoselov, A. K. Geim, S. V Morozov, D. Jiang, Y. Zhang, S. V Dubonos, I. V Grigorieva,

and A. A. Firsov, Science (80-. ). 306, 666 (2004).

[38] A. Winchester, S. Ghosh, S. Feng, A. L. Elias, T. Mallouk, M. Terrones, and S. Talapatra, ACS Appl.

Mater. Interfaces 6, 2125 (2014).

[39] A. Catellani, M. Posternak, A. Baldereschi, H. J. F. Jansen, and A. J. Freeman, Phys. Rev. B 32, 6997

(1985).

[40] R. Kershaw, M. Vlasse, and A. Wold, Inorg. Chem. 6, 1599 (1967).

[41] F. J. Di Salvo and J. E. Graebner, Solid State Commun. 23, 825 (1977).

[42] B. Sipos, a F. Kusmartseva, A. Akrap, H. Berger, L. Forró, and E. Tutis, Nat. Mater. 7, 960 (2008).

[43] F. J. Di Salvo, J. a. Wilson, B. G. Bagley, and J. V. Waszczak, Phys. Rev. B 12, 2220 (1975).

[44] H. Mutka, L. Zuppiroli, P. Molinié, and J. C. Bourgoin, Phys. Rev. B 23, 5030 (1981).

[45] Y. Zhou and S. Ramanathan, Crit. Rev. Solid State Mater. Sci. 38, 286 (2013).

[46] P. Darancet, A. J. Millis, and C. a. Marianetti, Phys. Rev. B 90, 2 (2014).

[47] F. Jellinek, J. Less-Common Met. 4, 9 (1962).

[48] P. Schmidt, J. Kröger, B. M. Murphy, and R. Berndt, New J. Phys. 10, 13022 (2008).

[49] A. Spijkerman, J. L. de Boer, A. Meetsma, G. A. Wiegers, and S. van Smaalen, Phys. Rev. B 56,

13757 (1997).

[50] K. Rossnagel and N. V Smith, Phys. Rev. B 73, 73106 (2006).

[51] M. Capone, L. Capriotti, F. Becca, and S. Caprara, 63, 5 (2000).

[52] N. F. Mott, Philos. Mag. 6, 287 (1961).

[53] P. P. Edwards and M. J. Sienko, Acc. Chem. Res. 15, 87 (1982).

[54] Y. Yu, F. Yang, X. F. Lu, Y. J. Yan, Y. Cho, L. Ma, X. Niu, S. Kim, Y. Son, D. Feng, S. Li, S.

Cheong, X. H. Chen, and Y. Zhang, Nat. Nanotechnol. 10, 1 (2015).

[55] M. Imada, A. Fujimori, and Y. Tokura, Rev. Mod. Phys. 70, 1039 (1998).

[56] Y. W. Yin, J. D. Burton, Y.-M. Kim, A. Y. Borisevich, S. J. Pennycook, S. M. Yang, T. W. Noh, A.

Gruverman, X. G. Li, E. Y. Tsymbal, and Q. Li, Nat. Mater. 12, 1 (2013).

151

[57] H. Fujishiro, T. Fukase, and M. Ikebe, J. Phys. Soc. Japan 70, 628 (2001).

[58] Y. Liu, R. Ang, W. J. Lu, W. H. Song, L. J. Li, and Y. P. Sun, Appl. Phys. Lett. 102, 1 (2013).

[59] D. Martin, E. Yurchuk, S. Müller, J. Müller, J. Paul, J. Sundquist, S. Slesazeck, T. Schlösser, R. Van

Bentum, M. Trentzsch, U. Schröder, and T. Mikolajick, Solid. State. Electron. 88, 65 (2013).

[60] N. V Smith, S. D. Kevan, and F. J. DiSalvo, J. Phys. C Solid State Phys. 18, 3175 (1985).

[61] X. L. Wu and C. M. Lieber, Science (80-. ). 243, 1703 (1989).

[62] K. Xu, L. Yin, Y. Huang, T. A. Shifa, J. Chu, F. Wang, R. Cheng, Z. Wang, and J. He, Nanoscale 8,

16802 (2016).

[63] Y. Gong, Z. Liu, A. R. Lupini, G. Shi, J. Lin, S. Najmaei, Z. Lin, A. Laura, A. Berkdemir, G. You,

H. Terrones, M. Terrones, R. Vajtai, S. T. Pantelides, S. J. Pennycook, J. Lou, W. Zhou, and P. M.

Ajayan, Nano Lett. 14, 442 (2014).

[64] K. Kang, S. Xie, L. Huang, Y. Han, P. Y. Huang, K. F. Mak, C.-J. Kim, D. Muller, and J. Park,

Nature 520, 656 (2015).

[65] D. Liu, W. Zhang, D. Mou, J. He, Y.-B. Ou, Q.-Y. Wang, Z. Li, L. Wang, L. Zhao, S. He, Y. Peng,

X. Liu, C. Chen, L. Yu, G. Liu, X. Dong, J. Zhang, C. Chen, Z. Xu, J. Hu, X. Chen, X. Ma, Q. Xue,

and X. J. Zhou, Nat. Commun. 3, 931 (2012).

[66] J. Lauth, F. E. S. Gorris, M. Samadi Khoshkhoo, T. Chassé, W. Friedrich, V. Lebedeva, A. Meyer,

C. Klinke, A. Kornowski, M. Scheele, and H. Weller, Chem. Mater. 28, 1728 (2016).

[67] M. J. Hollander, Y. Liu, W.-J. Lu, J. Li, Y.-P. Sun, J. a. Robinson, and S. Datta, Nano Lett. 15, 1861

(2015).

[68] A. Luican-Mayer, J. R. Guest, and S.-W. Hla, arXiv Prepr. arXiv1506.04102 (2015).

[69] X. Yu, T. Jia, and L. Zou, arXiv 1407, 1407 (2014).

[70] M. Yoshida, Y. Zhang, J. Ye, R. Suzuki, Y. Imai, S. Kimura, A. Fujiwara, and Y. Iwasa, Sci. Rep. 4,

1 (2014).

[71] Y. Fujisawa, T. Shimabukuro, H. Kojima, K. Kobayashi, S. Demura, and H. Sakata, IOP Conf. Ser.

J. Phys. Conf. Ser. 871, 12003 (2017).

[72] S. Tanda, T. Sambongi, T. Tani, and S. Tanaka, J. Phys. Socitey Japan 53, 476 (1984).

152

[73] S. Dunning-Riley and S. Merryfield, Nursing (Lond). 44, 8 (2014).

[74] W. Fu, Y. Chen, J. Lin, X. Wang, Q. Zeng, J. Zhou, L. Zheng, H. Wang, Y. He, H. He, Q. Fu, K.

Suenaga, T. Yu, and Z. Liu, Chem. Mater. 28, 7613 (2016).

[75] P. Schmidt, M. Binnewies, R. Glaum, and M. Schmidt, in Adv. Top. Cryst. Growth (InTech, 2013).

[76] A. Ubaldini and E. Giannini, J. Cryst. Growth 401, 878 (2014).

[77] Y. Shi, W. Zhou, A.-Y. Lu, W. Fang, Y.-H. Lee, A. L. Hsu, S. M. Kim, K. K. Kim, H. Y. Yang, L.-

J. Li, J.-C. Idrobo, and J. Kong, Nano Lett. 12, 2784 (2012).

[78] L. K. Tan, B. Liu, J. H. Teng, S. Guo, H. Y. Low, and K. P. Loh, Nanoscale 6, 10584 (2014).

[79] E. Ahvenniemi, A. R. Akbashev, S. Ali, M. Bechelany, M. Berdova, S. Boyadjiev, D. C. Cameron,

R. Chen, M. Chubarov, V. Cremers, A. Devi, V. Drozd, L. Elnikova, G. Gottardi, K. Grigoras, D. M.

Hausmann, C. S. Hwang, S.-H. Jen, T. Kallio, J. Kanervo, I. Khmelnitskiy, D. H. Kim, L. Klibanov,

Y. Koshtyal, A. O. I. Krause, J. Kuhs, I. Kärkkänen, M.-L. Kääriäinen, T. Kääriäinen, L. Lamagna,

A. A. Łapicki, M. Leskelä, H. Lipsanen, J. Lyytinen, A. Malkov, A. Malygin, A. Mennad, C. Militzer,

J. Molarius, M. Norek, Ç. Özgit-Akgün, M. Panov, H. Pedersen, F. Piallat, G. Popov, R. L. Puurunen,

G. Rampelberg, R. H. A. Ras, E. Rauwel, F. Roozeboom, T. Sajavaara, H. Salami, H. Savin, N.

Schneider, T. E. Seidel, J. Sundqvist, D. B. Suyatin, T. Törndahl, J. R. van Ommen, C. Wiemer, O.

M. E. Ylivaara, and O. Yurkevich, J. Vac. Sci. Technol. A Vacuum, Surfaces, Film. 35, 10801 (2017).

[80] M. Leskelä and M. Ritala, Angew. Chemie Int. Ed. 42, 5548 (2003).

[81] International Technology Roadmap for Semiconductors, 2007 Edition (n.d.).

[82] a. C. Dillon, A. W. Ott, J. D. Way, and S. M. George, Surf. Sci. 322, 230 (1995).

[83] J. D. Ferguson, A. W. Weimer, and S. M. George, Thin Solid Films 371, 95 (2000).

[84] A. W. W. Ott, J. W. W. Klaus, J. M. M. Johnson, and S. M. M. George, Thin Solid Films 292, 135

(1997).

[85] Y. Widjaja and C. B. Musgrave, Appl. Phys. Lett. 80, 3304 (2002).

[86] Y. Xu and C. B. Musgrave, Chem. Mater. 16, 646 (2004).

[87] G. S. Higashi and C. G. Fleming, Appl. Phys. Lett. 55, 1963 (1989).

[88] B. Lux, C. Colombier, H. Altena, and K. Stjernberg, Thin Solid Films 138, 49 (1986).

153

[89] M. D. Groner, F. H. Fabreguette, J. W. Elam, and S. M. George, Chem. Mater. 16, 639 (2004).

[90] R. Puurunen, Chem. Vap. Depos. 9, 249 (2003).

[91] Z. Chen, H. Wang, X. Wang, P. Chen, Y. Liu, H. Zhao, Y. Zhao, Y. Duan, S. Y. Chen, C. F. Lai, R.

H. Hwang, Y. H. Lai, M. S. Wang, A. Forkan, I. Khalil, Z. Tari, J. Lee, Z. T. Zhang, L. Liu, Y. Yu,

C. Yan, K. Li, Z. Zheng, S. Reineke, T. H. Han, H. Aziz, Z. D. Popovic, N. X. Hu, A. M. Hor, G.

Xu, M. S. Xu, J. B. Xu, H. Z. Chen, M. Wang, A. B. Chwang, A. P. Ghosh, L. J. Gerenser, C. M.

Jarman, J. E. Fornalik, L. H. Kim, M. Li, D. W. Choi, F. Nehm, C. Y. Park, J. S. An, H. J. Jang, J.

H. Lee, B. H. Choi, Y. Y. Lin, Y. N. Chang, M. H. Tseng, C. C. Wang, F. Y. Tsai, W. M. Yun, Y.

Duan, P. Sundberg, A. Sood, X. Liu, M. Karppinen, B. Yoon, B. H. Lee, S. M. George, J. Meyer, H.

Schmidt, W. Kowalsky, T. Riedl, A. Kahn, Z. Giedraityte, P. Sundberg, M. Karppinen, D. S. Han,

D. K. Choi, J. W. Park, S. Feng-Bo, B. H. Lee, K. H. Lee, S. Im, M. M. Sung, P. Saint-Cast, H. C.

Knoops, P. Sundberg, M. Karppinen, B. Yoon, D. Seghete, A. S. Cavanagh, S. M. George, Y.

Tomczak, K. Knapas, M. Sundberg, M. Leskela, M. Ritala, R. L. Puurunen, V. Miikkulainen, M.

Leskela, M. Ritala, R. L. Puurunen, K. Ali, K. H. Choi, C. Hossbach, and S. Lee, Sci. Rep. 7, 40061

(2017).

[92] T. J. Larrabee, A Quantified Dosing ALD Reactor with in-Situ Diagnostics for Surface Chemistry

Studies, The Pennsylvania State University, 2012.

[93] D. R. Sadoway and S. N. Flengas, Can. J. Chem. 54, 1692 (1976).

[94] J. N. Hilfiker and R. A. Synowicki, Solid State Technol. 39, 157 (1996).

[95] E. A. I. (Editors) R.W.Collins, D.E.Aspnes, Solid Thin Film. 313–314, 1 (1998).

[96] H. G. Tompkins and E. A. Irene, Handbook of Ellipsometry (William Andrew, 2005).

[97] H. Fujiwara, Spectroscopic Ellipsometry: Principles and Applications (John Wiley & Sons, 2007).

[98] G. N. Maracas, J. L. Edwards, K. Shiralagi, K. Y. Choi, R. Droopad, B. Johs, and J. A. Woolam, J.

Vac. Sci. Technol. A 10, 1832 (1992).

[99] G. E. J. Jr and F. A. Modine, Appl. Phys. Lett 69, 371 (1996).

[100] G. Bemski, Proc. IRE 46, 990 (1958).

[101] M. Houssa, G. Pourtois, M. Caymax, M. Meuris, and M. M. Heyns, Surf. Sci. 602, L25 (2008).

154

[102] W. Shockley and W. T. Read, Phys. Rev. 87, 835 (1952).

[103] E. H. Nicollian and A. Goetzberger, Bell Syst. Tech. J. XLVI, 1055 (1967).

[104] Y. Yuan, L. Wang, B. Yu, B. Shin, J. Ahn, P. C. Mcintyre, P. M. Asbeck, M. J. W. Rodwell, and Y.

Taur, Electron Device Lett. IEEE 32, 485 (2011).

[105] D. M. Fleetwood, P. S. Winokur, R. a. Reber, T. L. Meisenheimer, J. R. Schwank, M. R. Shaneyfelt,

and L. C. Riewe, J. Appl. Phys. 73, 5058 (1993).

[106] I. K. Oh, M. K. Kim, J. S. Lee, C. W. Lee, C. Lansalot-Matras, W. Noh, J. Park, A. Noori, D.

Thompson, S. Chu, W. J. Maeng, and H. Kim, Appl. Surf. Sci. 287, 349 (2013).

[107] K. Martens, Electrical Characterization and Modeling of Ge/III-V-Dielectric Interfaces, 2009.

[108] Http://ftps.zdnet.com.cn/files/2/19192.pdf, (n.d.).

[109] J. S. S. K. Gupta and J. A. S. K. Gupta, Physics and Technology of Silicon Carbide Devices (InTech,

2012).

[110] J. Lin, K. Xiong, and J. Robertson, Appl. Phys. Lett. 97, 3 (2010).

[111] E. Mateo Marti, C. Methivier, and C. M. Pradier, Langmuir 20, 10223 (2004).

[112] H. Adhikari, S. Sun, P. Pianetta, C. E. D. Chidsey, and P. C. McIntyre, Surface Passivation of

Germanium Nanowires (Stanford Linear Accelerator Center (United States). Funding organisation:

US Department of Energy (United States), 2005).

[113] R. Cao, X. Yang, J. Terry, and P. Pianetta, Phys. Rev. B 45, 13749 (1992).

[114] D. Knapp, Chemistry and Electronics of the Ge(111) Surface, California Institute of Technology,

2011.

[115] S. Tanuma, C. J. Powell, and D. R. Penn, Surf. Interface Anal. 21, 165 (1994).

[116] Https://tools.thermofisher.com/content/sfs/brochures/D16069~.pdf, (n.d.).

[117] J. C. Vickerman and I. Gilmore, Surface Analysis: The Principal Techniques (John Wiley & Sons,

2011).

[118] Y. Zheng, A. Agrawal, G. B. Rayner, M. J. Barth, K. Ahmed, S. Datta, and R. Engel-Herbert,

Electron Device Lett. IEEE 36, 881 (2015).

[119] P. Ponath, A. B. Posadas, and A. A. Demkov, Appl. Phys. Rev. 4, 21308 (2017).

155

[120] T. P. Schneider, D. A. Aldrich, J. Cho, and R. J. Nemanich, MRS Online Proc. Libr. Arch. 220,

(1991).

[121] M. Walker, M. S. Tedder, J. D. Palmer, J. J. Mudd, and C. F. Mcconville, Appl. Surf. Sci. 379, 1

(2016).

[122] J. Cho and R. J. R. Nemanich, Phys. Rev. B 46, 12421 (1992).

[123] E. Landemark, C. Karlsson, L. Johansson, and R. Uhrberg, Phys. Rev. B 49, 16523 (1994).

[124] T. Kaufman-Osborn, K. Kiantaj, C.-P. Chang, and A. C. Kummel, Surf. Sci. 630, 254 (2014).

[125] J. Cho, T. P. Schneider, and R. J. Nemanich, in MRS Proc. (Cambridge Univ Press, 1992), p. 237.

[126] M. Houssa, G. Pourtois, M. Caymax, M. Meuris, and M. M. Heyns, Appl. Phys. Lett. 92, 242101

(2008).

[127] J. Cho, T. P. Schneider, and R. J. Nemanich, in MRS Proc. (Cambridge Univ Press, 1992), p. 237.

[128] C. Förster, F. Schnabel, P. Weih, T. Stauden, O. Ambacher, and J. Pezoldt, Thin Solid Films 455–

456, 695 (2004).

[129] J. H. Parker Jr, D. W. Feldman, and M. Ashkin, Phys. Rev. 155, 712 (1967).

[130] J. Wagner and M. Cardona, Phys. Rev. B 32, 8071 (1985).

[131] N. Fukata, K. Sato, M. Mitome, Y. Bando, T. Sekiguchi, M. Kirkham, J. Hong, Ќ. Z. L. Wang, R. L.

S. Ќ, Z. L. Wang, and R. L. Snyder, ACS Nano 4, 3807 (2010).

[132] F. Evangelisti, M. Garozzo, and G. Conte, J. Appl. Phys. 53, 7390 (1982).

[133] H. S. Chen and D. Turnbull, J. Appl. Phys. 40, 4214 (1969).

[134] C. Y. Tsao, J. W. Weber, P. Campbell, P. I. Widenborg, D. Song, and M. A. Green, Appl. Surf. Sci.

255, 7028 (2009).

[135] M. Wojtaszek, R. Zuzak, S. Godlewski, M. Kolmer, J. Lis, B. Such, and M. Szymonski, J. Appl.

Phys. 118, 1 (2015).

[136] M. Wojtaszek, J. Lis, R. Zuzak, B. Such, and M. Szymonski, Appl. Phys. Lett. 105, (2014).

[137] P. J. Fensham, K. Tamaru, M. Boudart, and H. Taylor, J. Phys. Chem. 59, 806 (1955).

[138] M. S. Gordon, D. R. Gano, J. S. Binkley, and M. J. Frisch, J. Am. Chem. Soc. 108, 2191 (1986).

[139] K. Hwang, E. Yoon, K. Whang, and J. Lee, J. Electrochem. Soc. 144, 335 (1997).

156

[140] A. Paranjpe, S. Gopinath, T. Omstead, and R. Bubber, J. Electrochem. Soc. 148, G465 (2001).

[141] K.-E. Elers, V. Saanila, P. J. Soininen, W.-M. Li, J. T. Kostamo, S. Haukka, J. Juhanoja, and W. F.

a. Besling, Chem. Vap. Depos. 8, 149 (2002).

[142] L. Niinisto, Proc. CAS Int. Semicond. Conf. 33 (2000).

[143] K. Ahmad and K. Schuegraf, IEEE Spectr. 50, (2011).

[144] C. Pang, C. Lee, and K. Suh, J. Appl. Polym. Sci. 130, 1429 (2013).

[145] a. Purniawan, P. J. French, G. Pandraud, and P. M. Sarro, Procedia Eng. 5, 1131 (2010).

[146] S. P. Adiga, L. a Curtiss, J. W. Elam, M. J. Pellin, C. Shih, C. Shih, S. Lin, Y. Su, S. D. Gittard, J.

Zhang, and R. J. Narayan, J. Miner. 60, 26 (2008).

[147] M. Knez, K. Nielsch, and L. Niinistö, Adv. Mater. 19, 3425 (2007).

[148] R. L. Puurunen, J. Appl. Phys. 97, 121301 (2005).

[149] V. Miikkulainen, M. Leskelä, M. Ritala, and R. L. Puurunen, J. Appl. Phys. 113, 21301 (2013).

[150] H. Kim, J. Vac. Sci. Technol. B 21, 2231 (2003).

[151] M. Leskelä and M. Ritala, Thin Solid Films 409, 138 (2002).

[152] J. A. Van Delft, D. Garcia-Alonso, and W. M. M. Kessels, Semicond. Sci. Technol. 27, 74002 (2012).

[153] T. J. Knisley, L. C. Kalutarage, and C. H. Winter, Coord. Chem. Rev. 257, 3222 (2013).

[154] J. R. Bakke, K. L. Pickrahn, T. P. Brennan, and S. F. Bent, Nanoscale 3, 3482 (2011).

[155] S. Swaminathan, M. Shandalov, Y. Oshima, and P. C. McIntyre, Appl. Phys. Lett. 96, 82904 (2010).

[156] A. Delabie, F. Bellenger, M. Houssa, T. Conard, S. Van Elshocht, M. Caymax, M. Heyns, and M.

Meuris, Appl. Phys. Lett. 91, 82904 (2007).

[157] F. Bellenger, M. Houssa, A. Delabie, V. Afanasiev, T. Conard, M. Caymax, M. Meuris, K. De Meyer,

and M. M. Heyns, J. Electrochem. Soc. 155, G33 (2008).

[158] S. Iwauchi and T. Tanaka, Jpn. J. Appl. Phys. 10, 260 (1971).

[159] M. Caymax, G. Eneman, F. Bellenger, C. Merckling, A. Delabie, G. Wang, R. Loo, E. Simoen, J.

Mitard, B. De Jaeger, G. Hellings, K. De Meyer, M. Meuris, and M. Heyns, Tech. Dig. - Int. Electron

Devices Meet. IEDM 461 (2009).

[160] E. Simoen, J. Mitard, G. Hellings, G. Eneman, B. De Jaeger, L. Witters, B. Vincent, R. Loo, A.

157

Delabie, S. Sioncke, M. Caymax, and C. Claeys, Mater. Sci. Semicond. Process. 15, 588 (2012).

[161] Q. Xie, S. Deng, M. Schaekers, D. Lin, M. Caymax, A. Delabie, X.-P. Qu, Y.-L. Jiang, D.

Deduytsche, and C. Detavernier, Semicond. Sci. Technol. 27, 74012 (2012).

[162] D. A. Newsome, D. Sengupta, H. Foroutan, M. F. Russo, and A. C. T. van Duin, J. Phys. Chem. C

116, 16111 (2012).

[163] G. D. Wilk, R. M. Wallace, and J. M. Anthony, J. Appl. Phys. 89, 5243 (2001).

[164] A. Delabie, A. Alian, F. Bellenger, M. Caymax, T. Conard, A. Franquet, S. Sioncke, S. Van Elshocht,

M. M. Heyns, and M. Meuris, J. Electrochem. Soc. 156, G163 (2009).

[165] M. L. Huang, Y. C. Chang, C. H. Chang, Y. J. Lee, P. Chang, J. Kwo, T. B. Wu, and M. Hong, Appl.

Phys. Lett. 87, 252104 (2005).

[166] C. L. Hinkle, a. M. Sonnet, E. M. Vogel, S. McDonnell, G. J. Hughes, M. Milojevic, B. Lee, F. S.

Aguirre-Tostado, K. J. Choi, H. C. Kim, J. Kim, and R. M. Wallace, Appl. Phys. Lett. 92, 71901

(2008).

[167] C. H. Chang, Y. K. Chiou, Y. C. Chang, K. Y. Lee, T. D. Lin, T. B. Wu, M. Hong, and J. Kwo, Appl.

Phys. Lett. 89, 242911 (2006).

[168] C. R. Pollock, Fundamentals of Optoelectronics (Irwin Chicago, 1995).

[169] R. Engel-Herbert, Y. Hwang, and S. Stemmer, J. Appl. Phys. 108, 0 (2010).

[170] Https://en.wikipedia.org/wiki/Quartz_crystal_microbalance, (n.d.).

[171] M. Akhtar, G. Anderson, R. Zhao, A. Alruqi, J. E. Mroczkowska, G. Sumanasekera, and J. B.

Jasinski, Npj 2D Mater. Appl. 1, 5 (2017).

[172] Y. M. Shi, H. N. Li, and L. J. Li, Chem Soc Rev 44, 2744 (2015).

[173] J. M. E. Harper, T. H. Geballe, and F. J. DiSalvo, Phys. Rev. B 15, 2943 (1977).

[174] G. Mihaly, N. Housseau, H. Mutka, L. Zuppiroli, J. Pelissier, P. Gressier, A. Meerschaut, and J.

Rouxel, J. Phys. Lettres 42, 263 (1981).

[175] M. Bansal, R. Srivastava, C. Lal, M. N. Kamalasanan, and L. S. Tanwar, Nanoscale 2, 1171 (2010).

[176] J. C. Tsang, C. Hermann, and M. W. Shafer, Phys. Rev. Lett. 40, 1528 (1978).

[177] M. J. Hollander, Y. Liu, W.-J. J. Lu, J. Li, Y.-P. P. Sun, J. a. Robinson, S. Datta, L. J. Li, Y.-P. P.

158

Sun, J. a. Robinson, and S. Datta, Nano Lett. 15, 1861 (2015).

[178] G. Liu, B. Debnath, T. R. Pope, T. T. Salguero, R. K. Lake, and A. A. Balandin, arXiv Prepr.

arXiv1602.05147 5 (2016).

[179] E. D. Grimley, T. Schenk, X. Sang, M. Pešić, U. Schroeder, T. Mikolajick, and J. M. LeBeau, Adv.

Electron. Mater. (2016).

[180] S. Mueller, J. Mueller, A. Singh, S. Riedel, J. Sundqvist, U. Schroeder, and T. Mikolajick, Adv.

Funct. Mater. 22, 2412 (2012).

[181] T. Olsen, U. Schröder, S. Müller, A. Krause, D. Martin, A. Singh, J. Müller, M. Geidel, and T.

Mikolajick, Appl. Phys. Lett. 101, 0 (2012).

[182] P. Polakowski and J. Müller, Appl. Phys. Lett. 106, 232905 (2015).

[183] T. S. Böscke, J. Müller, D. Bräuhaus, U. Schröder, and U. Böttger, Appl. Phys. Lett. 99, 2012 (2011).

[184] F. Bohra, B. Jiang, and J.-M. Zuo, Appl. Phys. Lett. 90, 161917 (2007).

[185] R. Chen, H. Kim, P. C. McIntyre, and S. F. Bent, Appl. Phys. Lett. 84, 4017 (2004).

[186] M. M. Atalla, E. Tannenbaum, and E. J. Scheibner, Bell Labs Tech. J. 38, 749 (1959).

[187] S. Mueller, C. Adelmann, A. Singh, S. Van Elshocht, U. Schroeder, and T. Mikolajick, ECS J. Solid

State Sci. Technol. 1, N123 (2012).

[188] J. Müller, T. S. Böscke, U. Schröder, R. Hoffmann, T. Mikolajick, and L. Frey, IEEE Electron Device

Lett. 33, 185 (2012).

[189] H. Chen and F. Lu, J. Vac. Sci. Technol. 1006, 1006 (2005).

[190] Y. Liu, T. Kamei, K. Endo, S. O&apos;uchi, J. Tsukada, H. Yamauchi, T. Hayashida, Y. Ishikawa,

T. Matsukawa, K. Sakamoto, A. Ogura, and M. Masahara, Jpn. J. Appl. Phys. 49, (2010).

[191] J. Müller, U. Schröder, T. S. Böscke, I. Müller, U. Böttger, L. Wilde, J. Sundqvist, M. Lemberger, P.

Kücher, T. Mikolajick, and L. Frey, J. Appl. Phys. 110, 114113 (2011).

[192] A. Didden, H. Battjes, R. MacHunze, B. Dam, and R. Van De Krol, J. Appl. Phys. 110, 1 (2011).

[193] R. Chowdhury, R. D. Vispute, K. Jagannadham, and J. Narayan, J. Mater. Res. 11, 1458 (1996).

[194] R. Ruh, H. J. Garrett, R. F. Domagala, and N. M. Tallan, J. Am. Ceram. Soc. 51, 23 (1968).

[195] J. Müller, T. S. Böscke, U. Schröder, S. Mueller, D. Bräuhaus, U. Böttger, L. Frey, and T. Mikolajick,

159

Nano Lett. 12, 4318 (2012).

[196] G. Gerra, a. K. Tagantsev, N. Setter, and K. Parlinski, Phys. Rev. Lett. 96, 1 (2006).

[197] J. Muller, T. S. Boscke, S. Muller, E. Yurchuk, P. Polakowski, J. Paul, D. Martin, T. Schenk, K.

Khullar, A. Kersch, W. Weinreich, S. Riedel, K. Seidel, A. Kumar, T. M. Arruda, S. V. Kalinin, T.

Schlosser, R. Boschke, R. Van Bentum, U. Schroder, and T. Mikolajick, Tech. Dig. - Int. Electron

Devices Meet. IEDM 280 (2013).

[198] M. J. Hollander, Y. Liu, W.-J. Lu, L.-J. Li, Y.-P. Sun, J. A. Robinson, and S. Datta, Nano Lett. 15,

1861 (2015).

[199] R. Inada, Y. Ōnuki, and S. Tanuma, Phys. B+C 99, 188 (1980).

[200] J. W. Mellor, A Comprehensive Treatise on Inorganic and Theoretical Chemistry (Longmans, Green,

1922).

[201] E. Nitrides, F. Oba, K. Tatsumi, I. Tanaka, and H. Adachi, J. Am. Chem. Soc. 85, 97 (2002).

[202] M. Yang, S. J. Wang, Y. P. Feng, G. W. Peng, and Y. Y. Sun, J. Appl. Phys. 102, (2007).

[203] I. S. Jeon, J. Park, D. Eom, C. S. Hwang, H. J. Kim, C. J. Park, H. Y. Cho, J. H. Lee, N. I. Lee, and

H. K. Kang, Japanese J. Appl. Physics, Part 1 Regul. Pap. Short Notes Rev. Pap. 42, 1222 (2003).

VITA Yuanxia Zheng

Education

PENNSYLVANIA STATE UNIVERSITY, UNIVERSITY PARK, PA

PH.D. IN PHYSICS. GPA=3.90/4.00. AUG 2011-DEC 2017

NANJING UNIVERSITY, NANJING, CHINA

M.S. IN PHYSICS. GPA=3.36/4.00. AUG 2008-JUN 2011

NANJING UNIVERSITY, NANJING, CHINA

B.S. IN PHYSICS. GPA=3.50/4.00. AUG 2004-JUN 2008

Publications 1. Y. Zheng, J.M. Lapano, G.B. Rayner Jr, R. Engel-Herbert, Native oxide removal from Ge surfaces by hydrogen

plasma, Appl. Phys. Lett. (under review).

2. Y. Zheng, S. Hong, G. Psofogiannakis, G.B. Rayner Jr, S. Datta, A.C.T. van Duin, et al., Modeling and in Situ Probing

of Surface Reactions in Atomic Layer Deposition, ACS Appl. Mater. Interfaces. 9 (2017) 15848–15856.

3. H. Zhang, L. Guo, G. Stone, L. Zhang, Y. Zheng, E. Freeman, et al., Imprinting of Local Metallic States into VO2

with Ultraviolet Light, Adv. Funct. Mater. 26 (2016) 6612–6618.

4. T.P. Senftle, S. Hong, M.M. Islam, S.B. Kylasa, Y. Zheng, Y.K. Shin, et al., The ReaxFF reactive force-field:

development, applications and future directions, Npj Comput. Mater. 2 (2016) 15011.

5. L. Zhang, Y. Zhou, L. Guo, W. Zhao, A. Barnes, H.-T. Zhang, et al., Correlated metals as transparent conductors, Nat.

Mater. 15 (2016) 204–210.

6. Y. Zheng, A. Agrawal, G.B. Rayner, M.J. Barth, K. Ahmed, S. Datta, et al., In-situ process control of trilayer gate

stacks on p-Germanium with 0.85 nm EOT, Electron Device Lett. IEEE. 36 (2015) 881–883. 7. A. Agrawal, M. Barth, G.B. Rayner, V.T. Arun, C. Eichfeld, G. Lavallee, S.Y. Yu, X Sang, S. Brookes, Y. Zheng,

Y.J. Lee, et al, Enhancement mode strained (1.3%) germanium quantum well FinFET (WFin= 20nm) with high mobility

(μHole= 700 cm2/V-s), low EOT (~0.7 nm) on bulk silicon substrate, IEDM 2013. 8. H.-T. Zhang, L. Zhang, D. Mukherjee, Y.-X. Zheng, R.C. Haislmaier, N. Alem, et al., Wafer-scale growth of VO2

thin films using a combinatorial approach, Nat. Commun. 6 (2015) 8475. 9. H.C. Xuan, L.Y. Wang, Y.X. Zheng, Q.Q. Cao, Y. Deng, D.H. Wang, et al., Large converse magnetoelectric effect

in ferromagnetic shape memory alloy Ni49Fe18Ga27Co6 and Pb(Zr0.52Ti0.48)O3 laminates, J. Alloys Compd. 519 (2012) 97–

100. 10. S.C. Ma, Y.X. Zheng, H.C. Xuan, L.J. Shen, Q.Q. Cao, D.H. Wang, et al., Large roomtemperature magnetocaloric

effect with negligible magnetic hysteresis losses in Mn1− xVxCoGe alloys, J. Magn. Magn. Mater. 324 (2012) 135–139. 11. L. Wang, D. Wang, Q. Cao, Y. Zheng, H. Xuan, J. Gao, et al., Electric control of magnetism at room temperature,

Sci. Rep. 2 (2012) 223. 12. S.Y. Chen, Y.X. Zheng, Q.Y. Ye, H.C. Xuan, Q.Q. Cao, Y. Deng, et al., Electric field-modulated Hall resistivity

and magnetization in magnetoelectric Ni–Mn–Co–Sn/PMN–PT laminate, J. Alloys Compd. 509 (2011) 8885–8887.

13. Y.X. Zheng, Q.Q. Cao, C.L. Zhang, H.C. Xuan, L.Y. Wang, D.H. Wang, et al., Study of uniaxial magnetism and

enhanced magnetostriction in magnetic-annealed polycrystalline CoFe2O4, J. Appl. Phys. 110 (2011) 43908. 14. H.C. Xuan, L.Y. Wang, Y.X. Zheng, Y.L. Li, Q.Q. Cao, S.Y. Chen, et al., Electric field control of magnetism without

magnetic bias field in the Ni/Pb(Mg1/3Nb2/3)O3-PbTiO3/Ni composite, Appl. Phys. Lett. 99 (2011) 32509. 15. C.L. Zhang, Y.X. Zheng, H.C. Xuan, S.C. Ma, Q.Q. Cao, D.H. Wang, et al., Large and highly reversible magnetic

field-induced strains in textured Co1−xNixMnSi alloys at room temperature, J. Phys. D. Appl. Phys. 44 (2011) 135003. 16. H.C. Xuan, L.Y. Wang, S.C. Ma, Y.X. Zheng, Q.Q. Cao, D.H. Wang, et al., Large converse magnetoelectric effect

in Metglas FeCoBSi and 0.7Pb(Mg1/3Nb2/3)O3-0.3PbTiO3 laminated composite, Appl. Phys. Lett. 98 (2011) 52505.

17. H.C. Xuan, Y.X. Zheng, S.C. Ma, Q.Q. Cao, D.H. Wang, Y.W. Du, The martensitic transformation, magnetocaloric

effect, and magnetoresistance in high-Mn content Mn47+xNi43-xSn10 ferromagnetic shape memory alloys, J. Appl. Phys.

108 (2010) 103920. 18. S. Chen, L. Wang, H. Xuan, Y. Zheng, D. Wang, J. Wu, et al., Multiferroic properties and converse magnetoelectric

effect in Bi1-xCaxFeO3 ceramics, J. Alloys Compd. 506 (2010) 537–540.