TRI3DYN Modelling and MEIS Measurements of Arsenic Dopant Profiles in FinFETSJonathan England

Varian Semiconductor EquipmentApplied Materials Implant Division

Gloucester MA USAJonathan_England@amat.com

Won Ja MinKorea Materials and Analysis Corporation

Daejon, South Koreawj.min@kmac.com

Lucio dos Santos RosaSurrey Ion Beam Centre

Advanced Technology InstituteUniversity of Surrey

Guildford, UKl.dossantosrosa@surrey.ac.uk

Jwasoon KimKorea Materials and Analysis Corporation

Daejon, South Koreajwas.kim@kmac.com

Abstract—Implant processes that produced a variety of arsenic dopant profiles in 130nm tall, 110nm pitch FinFET test structures were modelled using the dynamic, binary collision approximation code TRI3DYN and measured using a commercial time of flight medium energy ion scattering (TOF-MEIS) tool. Dopant profiles in the tops and sidewalls of the FinFETS were extracted from MEIS spectra using the 3D capability of the POWERMEIS code. The qualitative agreement between TRI3DYN predictions and MEIS measurements showed the need for POWERMEIS to account for the effects of multiple scattering (which has yet to be implemented) before more quantitative conclusions regarding the absolute accuracy of TRI3DYN can be made. Nevertheless, this work has shown the promise of TRI3DYN and TOF-MEIS as tools to predict and characterise the outcomes of implantation into 3D structures.

Keywords—Ion-implantation, ion beam modelling, TRI3DYN, plasma doping, PLAD, medium energy ion scattering, TOF-MEIS, POWERMEIS, FinFET

I. INTRODUCTION

Plasma doping (PLAD) is a leading contender to enable the conformal doping of 3D devices such as FinFETS. Previous work [1-3] used medium energy ion scattering (MEIS) measurements to characterise a PLAD process and subsequent resist clean and anneal production steps. MEIS was used to avoid the issue of ion beam mixing that hinders accurate interpretation of SIMS, the metrology more commonly used in the semiconductor industry. In that work, PLAD was modelled using a dynamic binary collision approximation (bca) code TRIDYN [4] which, in contrast to the static bca code TRIM [5], dynamically changes a planar substrate’s depth profile to account for direct implantation, deposition and ion beam mixing that must be accounted for during PLAD. The combination of MEIS dopant profile metrology and TRIDYN modelling allowed quantification of the various ion and neutral fluxes present and illustrated the particle substrate interactions for an arsenic PLAD process on bare, planar wafers. This approach has been applied to arsenic PLAD processing of FinFET like structures to investigate if dynamic, bca modelling could correctly predict dopant profiles in 3D structures and if MEIS metrology has sufficient sensitivity and could be interpreted to measure 3D dopant profiles.

Since the FinFET structures only occurred in small, localised regions of a full wafer, this study required the use of TOF- MEIS equipment that was able to identify regions of interest and collect spectra only from those regions. The

measured MEIS backscattered spectra then had to be interpreted accounting for the 3D nature of the substrate to generate the doping profiles in the fin tops, sidewalls and trench bottoms. POWERMEIS [6] was the only analysis package that had this capability. The dopant profiles extracted from the MEIS measurements were compared to TEM/EDS measurements and the outputs of a model generated using the 3D dynamic, bca code, TRI3DYN [7]. This modelling is also novel, so consistency between modelling and metrology could indicate the validity of the TRI3DYN code and assumptions of the magnitudes and angular distributions of the particle fluxes input to the model. A comparison between TRI3DYN modelling and TOF-MEIS metrology reported before [1] indicated consistency between the models and metrology, but the study was limited as only post anneal data was available at the time.

II. ARSENIC PLAD IMPLANTATION OF FINFETS

This study used die cut from test wafers that included 250 um x 250 um square regions exclusively containing 130 nm tall, 50nm wide FinFETs on a 110 nm pitch. To produce a variety of dopant profiles in the tops, sidewalls and trench bottoms of the FinFETs, die were either implanted with arsenic in a VIISTA TRIDENT high current beamline implanter at 3keV to a fluence of 7x1014 ions cm-2 using a 15° bimode recipe; a VIISTA PLAD at a bias of 2 keV to a total ion fluence of 5x1015 cm-2; or a combination of PLAD followed by beamline implants. It should be noted that the PLAD conditions were not chosen to produce the best conformality, but to give a range of different dose profiles for this study. Some of the implanted die were given sulphuric peroxide mixture (SPM) wet cleans to compare the effects of cleans on 3D structures to those observed on planar samples [1-3].

III. TRI3DYN PROCESS MODELLING

The modelling of arsenic PLAD implants into bulk wafers and 3D structures using TRI3DYN has been described before [1]. In summary, the starting substrate was represented by voxels which described the concentration of the various elements present and then ions and neutrals were launched at the substrate. The collision cascades initiated by the energetic particles were calculated and the voxels dynamically adjusted to account for atoms gained or lost during the cascades. Dopant profiles were extracted from the voxel contents upon completion of the model. The particle fluxes were calibrated from planar MEIS measurements of PLAD implants into a bare wafer. The FinFET structures

XXX-X-XXXX-XXXX-X/XX/$XX.00 ©20XX IEEE

50nm

30 40 50 60 70 80100

101

102

103

104

Scat

tere

d Io

n In

tens

ity (c

ount

s)

Energy (keV)

50nm

30 40 50 60 70 80100

101

102

103

104

Scat

tere

d Io

n In

tens

ity (c

ount

s)

Energy (keV)

50nm

30 40 50 60 70 80100

101

102

103

104

Scat

tere

d Io

n In

tens

ity (c

ount

s)

Energy (keV)

were small compared to the plasma Debye length so that the sheath would not be perturbed by the microstructure. The angles of incidence of ions, being accelerated across the plasma sheath, were therefore assumed to be normal to the wafer surface. The neutrals could arise from species sputtered from the chamber walls (which could potentially cause any angular distribution depending on the location from where the sputtering occurred) and from the gas (which would be expected to lead to a cosine form distribution [8]). In the models presented here, neutrals were assumed to have normal incidence with zero divergence.

IV. TOF-MEIS METROLOGY

The FinFETS were analysed by Korean Materials Analysis Corporation using a time of flight TOF-MEIS system [9]. The system included an optical microscope system to locate the region of interest to be analysed and a two electrostatic lens system focused a 100keV He primary ion beam to a diameter of 50μm. He ions backscattered through 130° were analysed using a time of flight (TOF) detector. Spectra were collected for a total primary beam charge of 300 nC; a typical measurement took about 90 minutes to acquire. The TOF detector allowed fast data acquisition which is particularly important as the focussed, high current density beams necessary to ensure that spectra are only collected from the region of interest can quickly modify the sample.

To obtain dopant profiles, samples were measured at three primary incidence angles [10]: 25º and 65º relative to a normal to the wafer surface across the FinFETs (i.e. in the plane shown in the TEM images in Fig. 1) and 25º along the FinFETs (i.e. perpendicular to the plane in Fig 1.) Simultaneous successful simulations of all three spectra should have given an unambiguous measurement of the dopant profiles. The samples were all twisted by 7º to avoid channelling.

Cross sectional TEM images of the PLAD and combined PLAD and beamline processed structures were made and EDS profiles taken across the FinFET tops and sidewalls. It should be noted that TEM/EDS measures a profile in one FinFET, in contrast to MEIS that measures profiles averaged over many structures. The cross sectional TEM images shown in Fig.1 show light grey areas which indicate where As has been deposited and implanted and Si amorphised. Interpreted in isolation, the very thin light areas on the sidewalls of the (low dose) PLAD only case could be interpreted as consequences of the implant or surface oxidation. The EDS measurements were able to measure As profiles in the fin tops and As was just visible for the combined PLAD and implant case but no As was detected in the sidewall for the PLAD only case.

V. POWERMEIS SIMULATION OF TOF-MEIS SPECTRA

Measured TOF-MEIS spectra were simulated using POWERMEIS with parameters described elsewhere [2] to produce fits such as those shown by the red lines in Fig.1. Using a similar approach to TRI3DYN, POWERMEIS launched primary He ions at a 3D substrate described using voxels of varying composition. The number and final energies of the scattered primary ions were calculated from considering the electronic energy loss as the ions travelled through voxels along their inward paths; the kinematics of a (single) primary / substrate atom collision that scattered the ions back towards the detector; and the electronic energy loss as the ions travelled through voxels along their outward paths. A measured MEIS spectrum was simulated by manually adjusting the atomic contents of the substrate voxels until the POWERMEIS simulated spectra at all angles matched the measured MEIS spectra by eye (even though chi squared statistics could be calculated). Each simulation took a few minutes to set up manually and calculate; unfortunately automated routines to vary the substrate and run simulations to optimise the fits were not available. Dopant profiles were extracted from the contents of the voxels that gave the best fit. Generating substrates where each voxel could have a different elemental composition was not straightforward. In this early study, dopant profiles were described by defining a

Fig. 1. Spectra measured by TOF-MEIS at 25° across the FinFETS (black lines) with their POWERMEIS fits (red lines) and bright-field post process TEM cross sections for: top) beamline implant (TEM image is of unprocessed FinFETS); middle) PLAD only; bottom) a combination of PLAD followed by a beamline implant.

limited number of layers uniformly along the fin top, sidewalls and trench bottoms (as shown in Fig. 2). The POWERMEIS optimisation consisted of varying the atomic contents in these layers. Transferring output from TRI3DYN directly into POWERMEIS was not evaluated, and capabilities such as describing variations in the dopant profiles around the fin corners were not available.

VI. INTERPRETATION OF MEIS SPECTRA

Several regions of interest can be identified in the MEIS spectra shown Fig. 1:

The As edge is at 83 keV. Primary ions scattered off atoms in the top of the fin populate the peak observed below this edge. Primary ions scattered off arsenic atoms in the sidewall loose further energy to traverse the fin, and so have energies below the peak. Ions scattered from the trench bottoms loose the most energy, and may even overlap with the Si signal.

The Si edge (the energy at which He ions are scattered from Si atoms on the sample surface) is at 63 keV. The shape of the large peak below 50 keV, due to scattering from Si within the FinFET, is characteristic of the finFET geometry [10].

Oxygen would be observed below 43 keV but is hidden by the Si signal; oxygen have not been included in the POWERMEIS analyses.

The fits to the lowest energy part of spectra shown in Fig. 1 were made using a combination of ~75% doped FinFETs and ~25% planar doped Si. This would represent the cases of the primary beam drifting, or exhibiting a beam halo, over regions of the planar Si surrounding the FinFETs. Measurements made by dwelling the beam for several minutes to create a mark on the sample confirmed that the beam was exclusively sampling the region of interest. It is thought that the inclusion of multiple scattering in the POWERMEIS code will improve the low energy fit assuming scattering exclusively from FinFETs.

Fig. 2 compares post implant arsenic profiles in the FinFETS predicted by TRI3DYN modelling and extracted from measured TOF-MEIS spectra using POWERMEIS for the three cases studied. The qualitative agreements for all cases are superficially pleasing, which suggests that the modelling and metrology approaches are worth pursuing, but clearly improvements are required for better quantitative agreement. It is evident that the fin widths of the POWERMEIS solutions are wider than those measured by TEM. This is consistent with POWERMEIS predicting less energy loss in the substrate than is actually occurring. This can be explained by the omission of multiple scattering in the calculations as the electronic energy loss should be well known under these experimental conditions. Inclusion of multiple scattering should improve the fit for more realistic, narrower fin shapes.

The beamline implanted profiles should be the most accurate as the process is pure ion implantation with well characterised ion beam current and total fluence. The agreement in dose profile shape between TRIDYN and MEIS in the fin tops for all cases is particularly close, but the values are different. The fidelity of the POWERMEIS spectral fits between 65keV and 80keV, sensitive to the sidewall doping (Fig. 1) is poor. Improving the spectral fits there (for all three angles of MEIS measurement), would also

improve the agreement between the POWERMEIS and TRI3DYN sidewall profiles shown in Fig. 2. It will also be

Fig. 2. Arsenic profiles predicted by TRI3DYN and extracted from TOF-MEIS measurements using POWERMEIS are plotted over TEM cross section images for: top) beamline implant; middle) PLAD only; bottom) a combination of PLAD and beamline implants. The graphs show profiles averaged in the fin tops and sidewalls at locations indicated by the red and blue boxes.

interesting to see how the inclusion of multiple scattering into POWERMEIS will affects the extracted As profiles.

Fig 3 shows the effect on TOF-MEIS measured spectra of the SPM clean after the PLAD only process. The study of the planar PLAD process [1-3] gives the expectation that the SPM clean should remove As from the surfaces of the FinFETs. This is indeed observed in the spectra, which shows As loss from the fin top and sidewall. It is interesting to note that the clean did not remove any As after the beamline only implants, and only removed As from the sidewall after the combined PLAD and beamline process.

Fig. 1. TOF-MEIS spectra measured for the PLAD only case after implant (red) and post SPM clean (blue) showing the loss of As from the fin top (energies between 80 – 85 keV) and sidewalls (energies 65keV to 80keV)

VII. CONCLUSIONS

Small regions of repeating FinFET test structures were measured by TOF-MEIS. Measured MEIS spectra were sensitive to process variations in fin tops and sidewalls and in this study had a higher detection capability for As in the sidewall than the TEM-EDS measurements. The sensitivity of the technique to dopant in FinFETS depended not only on the concentration of dopant atoms, but also on the geometry of the structures. The signal from the top of FinFETs scales with fin width and inversely with pitch; sidewall signal decreases with pitch. More open structures (thin fins of large pitch) exhibit the highest signals from the trench bottoms. The TOF-MEIS technique requires no sample preparation and the compact TOF-MEIS system footprint of 3.5 x 4 m in size would fit into a generic process characterisation laboratory. Variations in MEIS spectra at characteristic energies would immediately indicate changes in doping, cleaning or annealing processes.

Although improvements are possible, the agreement between TRI3DYN and POWERMEIS profiles suggest that these 3D modelling and metrology approaches are worthy of further investigation. Modelling using dynamic, bca codes appears to be a valuable tool for visualising and optimising

processes on new three dimensional structures. The early capability to model 3D MEIS spectra to extract arsenic profiles was manual and time consuming, the generation of the voxel compositions input to the POWERMEIS analysis could be more flexible and input directly from TRI3DYN models utilised. Adding these capabilities and the inclusion of multiple scattering should improve the fidelity of the fits for more realistic, narrower fin shapes.

Arsenic could be considered a best case for MEIS metrology because primary ions that have scattered from As atoms are mostly well separated in energy from scattering events from lighter Si substrate atoms. Metrology for lighter elements such as H, B and P are restricted by lower cross sections and energy overlap with scattering from substrate atoms. It will be interesting to see if 3D elemental profile metrology could be extended using techniques that exploit enhanced cross sections, such as elastic back scattering (EBS), changes in detected particle energy, such as nuclear reaction analysis (NRA), or forward scattering such as elastic recoil detection (ERD).

ACKNOWLEDGMENT

JE would like to thank his colleagues in the Plasma Doping Group and Process Laboratory of Varian Semiconductor Equipment, Applied Materials Implant Division for help in processing the samples and preparing this paper; DaeWon Moon at the Daegu Gyeongbuk Institute of Science and Technology for developing the TOF-MEIS equipment and initiating the interaction with Korea Materials Analysis Corporation; Pedro Grande, Gabriel Marmitt, Igor Alencar, Henrique Trombini at Universidade Federal do Rio Grande do Sul for supporting and discussing POWERMEIS and Wolfhard Moeller of Helmholtz Zentrum Dresden Rossendorf for inspiration and support concerning TRI3DYN modelling.

REFERENCES

[1] J. England, W. Möller, J.A. van den Berg, A. Rossall, W.J. Min, J. Kim Nucl. Instr. And Meth B 409 (2017) 60–64

[2] J. England, J. A. van den Berg, A. K. Rossall, unpublished[3] J. A. van den Berg, A. K. Rossall, M. J. Hussey, J. England

unpublished[4] W. Möller, W. Eckstein, Nucl. Instr. And Meth. B2 (1984) 814. [5] J.F. Ziegler, M.D. Ziegler, J.P. Biersack, SRIM - The stopping and

range of ions in matter (2010), Nucl Instrum Meth B, 268 (2010) 1818-1823.

[6] M. Sortica, P. Grande, G. Machado and L. Miotti, J. Appl. Phys. 106 (2009) 114320

[7] W. Möller, Nucl. Instr. And Meth. B322 (2014) 23[8] “TRI3DYN, 3D Dynamic Computer Simulation of Ion Irradiation

Processes in Nanostructures Using the Binary Collision Approximation User Guide”, (June 2018), p55, Wolfhard Möller

[9] K.-W. Jung, H. Yu, W.J. Min, K.-S. Yu, M.A. Sortica, P.L. Grande, D.W. Moon, Anal. Chem. 86 (2014) 1091.

[10] W. J. Min, J. Kim, K. Park, G. Marmitt, J. England, P. Grande, D. W. Moon, unpublished

30 40 50 60 70 80100

101

102

103

104

Scat

tere

d Io

n In

tens

ity (c

ount

s)

Energy (keV)