Optimization of Charge Pumping Technique in Polysilicon TFTs forGeometric Effect Elimination and Trap State Density Extraction

Lei Lu, Mingxiang Wang·, Man Wong l

Dept. of Microelectronics, Soochow University, Suzhou 215021, P. R. ChinaIDept. of Electronic and Computer Engineering, the Hong Kong University of Science and Technology, Hong Kong

*Email: mingxian~wang@suda.edu.cn

Abstract

Charge pumping (CP) technique is optimized tominimize the geometric component in the CP current inpolysilicon TFTs, by taking into account the pulsewaveform and its transition times. Based on theoptimization, ideal CP curves similar to those inMOSFETs are obtained. Important information on thetrap state density of polysilicon TFTs, i.e., the meanvalue as well as the energy distribution within theband-gap, can be reliably extracted in different ways.

1. Introduction

Polysilicon TFTs have been widely used inactive-matrix displays [1]. Charge pumping (CP)technique has been used as a powerful and reliablemethod for interface state density extraction inMOSFETs [2]. So far, a few works have been reported tocharacterize the trap properties of polysilicon TFTs byusing CP [3-10]. It has also been employed to investigatethe degradation of polysilicon TFTs [11, 12]. However,unlike in MOSFETs, CP never becomes a standardcharacterization technique for polysilicon TFTs, becauseit is still not sufficiently reliable and some ill-understoodphenomena are not clarified [3, 5, 7, 12], among whichthe geometric effect is an important one.

In MOSFETs CP measurement, at the pulse risingedge, some free electrons which do not have enoughtime to flow back to the source/drain will be recombinedwith holes from the substrate. Similar process alsohappens at the pulse falling edge for holes. They alsocontribute to the recombined charge per cycle (Qss)besides the "standard" CP mechanism, and provide anadditional component in the CP current (Icp) , i.e., thegeometric component current (Igeo) [2]. Geometric effectin MOSFETs was found to depend on device W/L, pulsewaveform, pulse transition time, etc [2]. In this paper,Igeo in polysilicon TFTs is eliminated (or minimized) byapplying optimized pulse waveform and transition times,which makes CP quantitatively more reliable for thedetermination of the trap state density (Dt).

2. Experimental

The experimental set-up for measuring I cp is illustrated

(a) ,.....----------,Drain (n+)

(C)·······r·····7T\···r················I···················r···7fi\·T··········r···· Vgp

*nJ\k(nmn~~hnnVnJnnnmimWn Vgb

iTrTri T iTrTwTr i Ti~a=f : T i 13=T

w

Figure 1. (a) Plane of the device. (b) Cross-sectionalview of the device and the schematic diagram of themeasurement. (c) Pulse parameters definition.

in Fig. 1 for n-channel polysilicon TFTs withW/L=30/10flm. Polysilicon is formed by the solutionbased metal-induced crystallization technique and isin Fig. 1 for n-channel polysilicon TFTs withW/L=30/10flm. Polysilicon is formed by the solutionbased metal-induced crystallization technique and islater recrystallized at 900°C [13]. The plane view (a)and cross-sectional view along the channel widthdirection (b) are shown in Fig.l. The source and drainare grounded. The p+ diffusion is the polysiliconsubstrate terminal, from which I cp is measured byapplying the square or triangular gate pulses at roomtemperature. The gate pulse parameters are defined inFig.lc, including the pulse height (Vph=3--6V), basevoltage (Vgb=-10--0V), peak voltage (Vgp= -5--7V),period (T=10--1000flS), rising time (Tr=O.l--7fls), fallingtime (TrO.l--l00flS) and the duty ratio for the triangularor square pulses (a, P=0.5). The CP measurement isperformed in two ways: keep a constant Vph whileincreasing Vgb (method A) or keep a constant Vgp whileincreasing Vph (method B).

978-1-4244-2186-2/08/$25.00 ©2008 IEEE

4

2.5 JL Tr TrVph=5V .O.lJJS l.0JJS

2.0 T=300J.!s • l.0JJS O.lJJS

Tr=Tr

~1.5~ O.l).lS

O.3JJS'-'"

~_~r-- O.5JJS~

~ 1.0 --0-- 1.OJJS3.0J.lS

0.5 5.0J.lS7.0JJS

5......-----------------=---./~

Vgp=O n /~/

T=300~s ~ L /~Tr=Tf .•',.//

~3 -- O.1~s ~/~ -.- O.51ls /./ -- .-~

S2 • I.OIlS /~ .• -••--./- .-~-~ 5.0IlS.'••-.~--..--.- ... -.. ..-....- ....--- ../f:'/J-'...---~-

~</~- ~ -T-'" --....- '" --'" -"'- '" -..- "'-'"

O~• • --1"246 8

Vph(V)Figure 4. I cp versus Vph for square gate pulse withdifferent 1',. and If· using method B.

or 1Jls but different Tr are quite different. It means thatthe observed Icp is controlled by Tr only.

In polysilicon TFTs there is no substrate bottomcontact, a side contact has to be used for the lcp

measurement, as shown in Fig.la. Since W»L, duringthe pulse rising edge, time needed for the accumulatedholes to transit back to the substrate must be muchlonger than that of inverted electrons to drift back to thesource/drain during the pulse falling edge. Furthermore,electrons can quickly drift along the low resistanceinversion channel, while holes must diffuse through thesubstrate bulk to the side contact. Hence, if 1',. is not longenough, recombination of the remaining holes in thechannel will contribute to the 1geo of polysilicon TFTs.From Fig.3, transit time of holes can be estimated as>'"1Jls while that of electrons should be less than 0.1 Jls.

Shown in Fig.4 are lcp curves measured by method Busing square pulse. lcp continuously decreases byincreasing Tr=T[,. When the transition time is longenough (>=lJls); 1geo becomes significantly suppressedand typical lcp saturation is observed, agreeing well with

-8 -4 -2 aVgb(V)

Figure 3. Icp versus Vgb for square gate pulse withdifferent Tr and 1'.['using method A.

·-~A

Vph=5VT=300J.1sTr==150J.1sTF150~s

.--..

'.

•

---SLVph=5VT=300~s

Tr==O.1~s

TFO.1~s0.5

1.0

2.5

3.0......------------------.

•

\.~ '

O.0 L..--....J~._.~._.~-L..------L_=__--"--~-.-.--.l.------±----l-8 -6 -4 -2 0

Vgb(V)Figure 2. Icp versus Vgb for square and triangular gatepulses using method A. I cp for triangular pulses ismultiplied by 5.

3. CP optimization for geometric effect elimination

In MOSFETs the geometric effect is supposed to beminimized by using device having W/L»1 [2]. In Fig.2,the Icp is shown with W/L=30/10Jlm when using themethod A for square and triangular pulses. One cannotice that the I cp peak value obtained by applyingtriangular pulses is much smaller than that of squarepulses. Furthermore, the I cp under square pulses isasymmetric and largely shifted toward the negative,while that under triangular pulses is well shaped and issimilar to those for MOSFETs. It clearly demonstratesthat the geometric effect in polysilicon TFTs can not besuppressed by increasing WIL. A previous study [5]proposed that the substrate resistance of polysiliconTFTs is much higher than that of MOSFETs due to thepresence of grain boundaries, resulting in a large I geo inlcp. The transition times (1',.=Tf ) of triangular pulses aremuch longer than those of square pulses, so electrons (orholes) can have enough time to flow back to thesourceldrain (or substrate) before recombination occursand therefore 1geo is minimized.

In Fig.3, by increasing transition times (Tr=Tf ) ofsquare pulses from 0.1 "'7Jls, lcp is continuouslydecreased. In the meanwhile, the distortion of I cp curveby I geo in the negative V gb region becomes less and lesssignificant. A typically symmetrical I cp curve is obtainedwhen transition times is longer than I JlS, with awell-defined saturated I cp values. It clearly demonstratesthat I geo is minimized or eliminated by increasing thepulse transition times. It can be well understood by thepreviously proposed mechanism [5]. To further clarifythe influence of Tr and 1'.1' also shown in Fig.3, Icp

measured under symmetric pulses (open symbols) arecompared with those under asymmetric ones (solidsymbols). Interestingly, Icp curves under the same 1',. ofeither 0.1 or 1Jls but very different 1'.[' coincide with eachother; while those curves under the same 1'.[' of either 0.1

4. D t extraction

Three approaches have been proposed to extract D(

14..----------------~

12 fo=2.18HzSlope=2 .34 x 1O-14C

10 Dt=2.16xI012cm-2eV-1

P 8 ..JanO'p=4.85 x IO-17cm2

o-; 6'-'"

(/.)

0 4

2 foo t10° 101 102 103 104 105 106

f(Hz)Figure 6. Qss versus frequency used to determine (UnUp)1I2

and ~.

or the energy distribution of Dt using CP in MOSFETs[2]. In the first approach, ~ is obtained from the

frequency dependence of Qss by applying triangularpulses [2]. It allows the extraction of ~ without the

need to know surface potential dependence on Vg • Thesemi-log plot of Qss as a function of/is shown in Fig.6.A straight line is obtained. Its slope and the intercept atx-axis fa can be used to determine ~ and (UnUp) 112

(mean capture cross section) by:

ji;i; = _1_. Vph • fa (2)

n p vthn; IVjb - ~I ~a(l- a)

dQss/d Inf =2qkTAg .~ (3)

where un/up, Vth, n;, Vjb, k and T are the electron/holecapture cross section, the thermal velocity, the intrinsiccarrier density, the device flat band voltage, theBoltzman's constant and the temperature, respectively.~ and (UnUp)1I2 are determined to be 2.16xl0 l2 cm-2ey- l

and 4.85x 10-17 cm-2, respectively.

In the second approach, Dt is obtained from the linearfrequency dependence of I cp by applying square pulses [2,3]. ~ is derived by the following equation:

_ - f Iv, - Vjbl (4).Icp - 2qfAg kT . Dt In(vthn; -V G'n G'pT;Tf V )

ph

From Eq.2, (UnUp) 112 is obtained, on which ~

calculation from Eq.4 only weakly depends. I cp as afunction off is plotted in Fig.7. One obtains a straightline, whose slope is used to extract ~ by Eq.4. ~ of

6.76xI011 cm-2 ey- l is obtained.In the third approach, energy distribution of Dt is

obtained from the transition time dependence of I cp byapplying square pulses with different transition times [2,4]. When varying 1f and fixing Tn one scans the energyrange between the conduction band (Ec) and the mid-gap(E;). D t energy distribution can be obtained as

4

0.0L..L..-..-"----J.-.L---.L.-L...--..I-----L-----L_ ....I...---I-----l....--'------L-----L----I_l--.....I

-8 -6 -4 -2 0 -2 0 2Vgb(V) V gp(V)

Figure 5. I cp as functions of V gb and V gp for triangular (a)and square pulse (b) with different Vph using method A.

2.0

0.8 (a) i·::: :-.". ~~-.i +-...... /\.J1&.II. ..... ..,I ....... V V ft...... .II.

~O.6 /'.' \ T=100 I ...<. I: J.tS r~ Vph!6 5 '4 3V ~ Tr=50J.l.S I Vph '3 :.4

-0.4 "\\.Tf=50Il

S i0.2 I

~ • • ... t/·0.0 ........... ....-•• __

0.5

~1.5

J'1.0

the result in Fig.3. As for the Icp curve of Tr=TrO.1 flS, anapparent linear dependence is obtained. In MOSFETs, itis known that Igeo follows the following formula [2]:

I geo = l5. fAgCox(Vg -~) (1)

where f, Ag, Cox, Vg, Vt and t5 are the pulse frequency,gate area, the gate oxide capacitance per unit area, thegate voltage, the device threshold voltage and theproportion of recombined carriers, respectively. Fromthe slope of I cp dependence t5 can be determined to be0.89. It indicates that most holes remain in the channelafter pulse rising edge and are subsequently recombinedby the inversion electrons.

Shown in Fig.5 are Icp curves measured by optimizedwaveform for (a) triangular and (b) square pulses. I cp isplotted against V gb in the left and Vgp in the right. WithIgeo eliminated from the I cp, ideally shaped CP curveswith clear falling/rising edges and central flat regionsare obtained under both conditions. Besides, thefalling/rising edges (Vgb~1Y and Vgp~1.5Y, asindicated) of two measurement conditions coincide, sodo the positions of central flat regions (between Vgp>O

and Vgb<-2Y, as indicated). The maximum Icp is seen togradually increase with Vph, which is in agreement witha previous CP model [2]. Compared to previous CPstudies in polysilicon TFTs [3, 7], the optimized I cp

shown here is clearly more suitable for the Dt extraction.

0.0 3.0x104

6.0xl04

9.0xl04

[(Hz)Figure 7.lcp versus frequency used to determine D,.

20

15

5

o

Slope=1.90 x10-B eOt=6. 76x 101lem-2ey-l

SL YPh=5YYgb=-4YTr=TFIJls

band-gap scanned by triangular pulses. A more elaboratestudy is needed to clarify such difference.

5. Conclusions

Optimized CP is used to mInImIze the 1geo inpolysilicon TFTs, by taking into account the pulsewaveform and transition time. Based on the optimization,ideal lcp curves are obtained. Furthermore, Dt areextracted from three approaches, while D, values agree

well and typical dual exponential energy distribution ofD t is obtained. It is confirmed with 1geo minimizedoptimized CP is more reliable for Dt extraction inpolysilicon TFTs.

Acknowledgments

0.00/ 0.20 0.25 0.30 /0.50Ei E-Ei(eV) Ee

Figure 8. Dual exponential energy distribution ofD t•

follows:

D (E) - Tf . dlcp (5)t - - qAgkTf ~r

E - E; = -kT InlvthO"nn; .Tf IVjb - v;1/Vph 1 (6).

Shown in Fig.8 is the semi-log plot of the energydistribution obtained by Eq.5 and Eq.6. This dualexponential distribution has a small slope of 0.597 eV-1

for deep states and a large slope of 5.039 eV-1 for tailstates, agreeing with the typical energy distribution of Dt

determined from other measurements [14, 15]. Thecurrent result obtained by using optimized CP techniqueis in contrast to that in some previous CP studies forpolysilicon TFTs, where irregular or distorted energydistribution ofD t was obtained [5, 6]. According to Eq.6,a broader range of energy distribution can still beachieved by changing the measurement temperature.

D t values in the latter two approaches agree well,which also agree with those of high temperaturepolysilicon TFTs in previous CP studies [3, 4, 6].However, Dt derived in the first approach is relativelylarger, possibly because of different energy range in

II

10

~9

~8N

~§ 7o~

~6

c55

~ Vph=5V/ ~Vgb=-4V

Tr=IIlS TFO.l '"" IOOlls

Slope=5.03geV-1

SI~pe=0.597eV-l•. .- ...

.. This work is supported by the National NaturalScience Foundation of China.

References

[1] E. Persidis, H. Baur and F. Pieralisi et aI.,Solid-State Electron., Vol. 52, p. 455 (2008).

[2] G. Groeseneken, H. E. Maes and N. Beltran et aI.,IEEE Trans. Electron Devices, ED-31, p.42(1984).

[3] M. Koyanagi and I-W. Wu et aI., IEDM Tech. Dig.,p.836 (1990).

[4] M. Koyanagi, Y. Baba and K. Hata et aI., IEEEElectron Device Lett., Vol. 13, p.152 (1992).

[5] G. W. Lee, J. W. Lee and C. H. Han, Jpn. 1. Appl.Phys., Vol. 38, p.2656 (1999).

[6] K. 1. Kim and o. Kim, Jpn. 1. Appl. Phys., Vol. 36,p.1394 (1997).

[7] N. S. Saks, S. Batra and M. Manning,Microelectronic Engineering, Vol. 28, p.379(1995).

[8] A. Balasinski, J. Worley and J. Walters et aI., IEEEElectron Device Lett., Vol. 16, p.460 (1995).

[9] K. J. Kim, W. K. Park and S. G. KIM, Solid-StateElectron., Vol. 42, p.1897 (1998).

[10] O. Kim and K. J. Kim, Electronics Letters, Vol. 34,p.809 (1998).

[11] A. Balasinski and 1. Worley et aI., IEDM Tech.Dig., p.529 (1995).

[12] C. Y. Chen, M. W. Ma and W. C. Chen et al., IEEEElectron Device Lett., Vol. 29, p.165 (2008).

[13] B. Zhang, Z. Meng and S. Zhao et aI., IEEE Trans.Electron Devices, Vol. 54, p. 1244 (2007).

[14] L. Pichon, A. Boukhenoufa and C. Cordier et aI., J.Appl. Phys., Vol. 100, p. 54504-1 (2006).

[15] G. A. Armstrong, S. uppal and S. D. brotherton etaI., Jpn. J. Appl. Phys., Vol. 37, p. 1721 (1998).