Hot-carrier charge trapping and trap generation in HfO2 and Al2O3 field-effect transistorsArvind Kumar, Massimo V. Fischetti, Tak H. Ning, and Evgeni Gusev Citation: J. Appl. Phys. 94, 1728 (2003); doi: 10.1063/1.1586985 View online: http://dx.doi.org/10.1063/1.1586985 View Table of Contents: http://jap.aip.org/resource/1/JAPIAU/v94/i3 Published by the AIP Publishing LLC. Additional information on J. Appl. Phys.Journal Homepage: http://jap.aip.org/ Journal Information: http://jap.aip.org/about/about_the_journal Top downloads: http://jap.aip.org/features/most_downloaded Information for Authors: http://jap.aip.org/authors

Downloaded 20 Sep 2013 to 152.14.136.96. This article is copyrighted as indicated in the abstract. Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions

JOURNAL OF APPLIED PHYSICS VOLUME 94, NUMBER 3 1 AUGUST 2003

Hot-carrier charge trapping and trap generation in HfO 2 and Al 2O3field-effect transistors

Arvind Kumar,a) Massimo V. Fischetti, Tak H. Ning, and Evgeni GusevIBM Semiconductor Research and Development Center, IBM Research Division, Thomas J. WatsonResearch Center, P.O. Box 218, Yorktown Heights, New York 10598

~Received 5 March 2003; accepted 5 May 2003!

We present a comprehensive experimental study of hot-carrier trap generation and charging effectsin high-k dielectrics using field-effect transistors fabricated with HfO2 and Al2O3 gate insulatorstacks and polycrystalline silicon gates. The experiments utilize substrate injection of hot carriersgenerated either optically or by direct injection in the dark from a forward-biasedp–n junction.Comparison of charge-trapping measurements taken using these two techniques onn-channelfield-effect transistors (nFETs) andp-channel field-effect transistors (pFETs) finds that enhancedcharge trapping occurs when hot holes are present~in the light or in the dark inpFETs but onlyunder illumination innFETs). A fundamental understanding of the conditions for hot-carrier damagein nFETs is obtained by studying the dependence on light wavelength, temperature, and substratebias. In particular, the wavelength dependence reveals that the hot-carrier damage depends on acombination of the electron and photon energies. Study of the time dependence of the gate currentindicates the buildup of positive charge in the dielectric during stressing. The density of interfacetraps generated by hot-carrier stressing is estimated using the capacitance–voltage characteristic,and charge transfer experiments to probe the existence of slow states are performed. Finally, theexperimental findings are discussed in the context of a speculative picture in which hot holes act asa precursor to damage in the oxide. ©2003 American Institute of Physics.@DOI: 10.1063/1.1586985#

ee

emim

d

h-cattr

et

-

uto

h-d

ton

ght-

tointo

of

g in

c-thating

tal-theetuphism-

hiftma

I. INTRODUCTION

Hot-carrier effects and insulator degradation have bextensively studied in metal–oxide–semiconductor~MOS!devices to understand and predict device reliability.1,2 Con-tinued scaling of device dimensions has led to greaterphasis on such issues, and indeed fundamental limitsposed by gate leakage and intrinsic reliability are expecteprevent reduction of the thickness of the SiO2-based gatedielectric in MOS devices below;1.2 nm.2 As a result,there is currently an intense effort to study alternative higkgate dielectrics in which the insulator physical thicknessbe increased due to the higher static dielectric constank,reducing leakage while maintaining adequate gate conover the channel.3,4 Even though reduced energy band offsare a well-known property of high-k insulators~typically de-creasing with increasingk!,5 prompting concern that hotcarrier reliability could be a serious issue, to date onlypreliminary study of hot-carrier effects has been carried o6

On the other hand, effects due to charge trapping of ccarriers~those injected by the gate insulator field!, such ashysteresis in capacitance–voltage (C–V) characteristics,have been commonly observed in many studies of higkdielectrics.3,7–13Together with experimentally observed fixecharges and reduced mobilities,3 both of which may beunavoidable,14,15 they are among potential challengesadoption of high-k materials. In this work, we have chose

a!Author to whom correspondence should be addressed; electronicarvkumar@us.ibm.com

1720021-8979/2003/94(3)/1728/10/$20.00

Downloaded 20 Sep 2013 to 152.14.136.96. This article is copyrighted as indicated in the abstract.

n

--

to

n

ols

a.ld

two high-k gate dielectrics, Al2O3 and HfO2, which havebeen widely investigated as potential SiO2 replacements dueto their relatively wide bandgaps, thermal stability, and hipermittivity,3,4 to carry out a comprehensive study of hocarrier effects.

Trapping of positive charge in the oxide is thoughtplay a fundamental role in oxide degradation, but evenSiO2 the origin of the positive charge remains subjectcontroversy.2 Hydrogen release1,16 and trapped holes,17 bothof which have been used in models of SiO2 degradation,have both been invoked to explain findings in studieshigh-k dielectrics.7,9,12,18 In our earlier work,6 hot-carrierstressing was found to result in enhanced charge trappinp-channel field effect transistors (pFETs), but similar effectswere seen inn-channel field effect transistors (nFETs) onlyunder illumination, when hot holes, in addition to hot eletrons, are likely to be present. These findings suggestholes play an important role in enhanced charge trappunder hot-carrier stress.

The purpose of this work is to obtain a fundamenunderstanding of when hot-carrier effects occur in highkdielectrics and to gain some insight about the nature ofdamage created. Section II describes our experimental sand Sec. III presents the main experimental findings of tarticle. Section IV is the discussion and Sec. V is the sumary and conclusion.

II. EXPERIMENTAL SETUP

We characterize the charge trapping by plotting the sin the flatband (DVfb) or threshold voltage (DVt) as a func-il:

8 © 2003 American Institute of Physics

Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions

te-

Fa

oeh

ol

deto

traspng

Iriee

d

thisssthem-fO

i-

c-tely

ctere

vice.

wed

re-e

ringrge

ub-ntalhot-fam-ce-nt inby

derob-

lier

ca

tra

-esh-

1729J. Appl. Phys., Vol. 94, No. 3, 1 August 2003 Kumar et al.

tion of the fluence, or total charge injected into the gaQinj . The shifts (DVfb) and (DVt) are obtained from measuring the high frequency~100–500 kHz! C–V characteris-tic at preset interruptions of the constant voltage stress.DVfb , the sense conditions are chosen to include a smrange ~typically 6200 mV) aroundVfb while for DVt thesense voltage is swept up to a bias in inversion which dnot cause measurable charge trapping; our experienceshown that the effect of any discharging during sensingsmall compared to the scale of effects seen here. In ccarrier stressing, a substantial overdriveuVg2Vtu of the gatevoltageVg above the threshold voltageVt results in carrierinjection by the gate insulator field. The band diagram unstress from substrate hot electrons is shown in the inseFig. 1~a!. The band bending, which determines the energythe incident carriers, is set by a reverse bias to the subsVx,0 while the conditionVg;Vt insures that the channel inearly equipotential.~Source/drain contacts are always keat ground.! An essential feature of our hot-carrier stressitechnique is that we have used both optical generationelectron–hole pairs19 and direct injection of minority carriersfrom a nearby forward-biasedp–n junction20 in order tocompare charge trapping in the light to that in the dark.order to insure that hot carriers are dominant in hot-carstressing experiments, we require that the gate current du

FIG. 1. Summary of previous results on HfO2 ~a! nFETs and~b! pFETs,comparing flatband voltage shift under stressing conditions using coldriers, hot carriers in the dark, and hot carriers in the light. The inset of~a!shows a band diagram in which hot electrons are injected from the subs

Downloaded 20 Sep 2013 to 152.14.136.96. This article is copyrighted as indicated in the abstract.

,

orll

sas

isd-

roffte

t

of

nrto

cold carriers~i.e., with no substrate bias, illumination, anp–n injector bias! must be a small component~less than20%! of the total gate current.21

Table I shows a summary of the samples used inwork, including gate dielectric, equivalent oxide thickne~EOT!, and flatband and threshold voltages; details ofprocessing are given in Ref. 3. To understand more copletely the effects discussed here we have used both H2

and Al2O3 gate stack devices (Al2O3 pFETs were not avail-able due to boron penetration! with polysilicon ~polycrystal-line silicon! gates. Following atomic layer deposition of ether HfO2 or Al2O3 on the same ultrathin~0.5–1.0 nm!oxynitride layer thermally pregrown on~100! Si substrates,polysilicon gate electrodes were deposited. Along with juntion formation, the gate electrodes were degenera(1020 cm23) doped (n1 for nFETs; p1 for pFETs), fol-lowed by dopant activation and first level of metal contausing a conventional MOS process flow. Measurements wtaken either on 1003100mm2 or 50350mm2 FETs, andeach new stressing measurement was done on a fresh de

III. EXPERIMENTAL RESULTS

An overview of this section is as follows. First, we shothat illumination is essential to observation of enhanccharge trapping innFETs, but not inpFETs. Second, havingestablished the role of illumination innFETs, we investigatethe dependence of the charge trapping on wavelength bystricting the maximum photon energy of the light. Third, wexamine the gate current, whose increase with time duhot-electron stressing suggests the buildup of positive chain the dielectric. Fourth, we study the temperature and sstrate bias dependences. Fifth, we consider an experimetest of whether the enhanced charge trapping seen afterhole stressing in apFET represents a reversible filling opre-existing trap states at high energies or, in fact, real dage in the gate insulator. Sixth, we look in detail at interfastate generation and conduct a charge transfer experimewhich charge is moved into and out of the created statesthe application of appropriate gate bias. Finally, we consiwhether the hot-carrier effects seen here can also beserved using cold carriers at high gate insulator fields.

A. Effect of illumination: nFETs versus pFETs

Figure 1 summarizes the main results of our earstudy6 using HfO2 nFETs @Fig. 1~a!# and HfO2 pFETs @Fig.1~b!#. The figures show plots of flatband voltage shift~ob-

r-

te.

TABLE I. Characteristics of wafers used in this study, including highkdielectric and physical thickness, EOT, device type, and flatband and throld voltages. All high-k dielectrics are deposited on top of a thin~0.5–1 nm!oxynitride layer.

Wafer Gate dielectric EOT~nm! Type Vfb ~V! Vt ~V!

H1 3 nm HfO2 1.80 nFET 20.75 0.50H2,H28 3 nm HfO2 1.85 pFET 0.40 20.90A1 3 nm Al2O3 1.68 nFET 20.55 0.85S1 2.2 nm SiO2 2.20 nFET 21 0.24

Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions

,eth

i-so

ab

ia

oheecd-ho

ietharo

u

.rsro

nkceois-

rtlym

ss-o-

eroedrtuch

y

e r

up-

rrent

1730 J. Appl. Phys., Vol. 94, No. 3, 1 August 2003 Kumar et al.

tained from 500 kHzC–V) as a function of injected chargecomparing trapping due to cold carriers, hot carriers genated by light, and hot carriers generated in the dark usingforward-biasedp–n injector. In the cold stressing experments~with no substrate bias!, the stressing gate voltage wachosen to give approximately the same gate current astained under the hot stressing conditions so that comparamounts of charge would be injected. Figure 1~a! also showsthat under comparable hot stressing conditions, essentno damage is observed in the SiO2 control wafer~wafer S1!.

The important features to note are:

~1! For pFETs, the flatband shifts are much larger for hcarriers than for cold carriers, independent of whetthe hot carriers are generated by light or by direct injtion in the dark. The shifts under illumination are moestly smaller than those in the dark, possibly due to ptodetrapping.

~2! For nFETs, the enhanced charge trapping of hot carroccurs only when the carriers are generated by light;shifts obtained by hot-electron stressing in the darkcomparable to or even smaller than those due to ccarriers.

~3! For pFETs, there is a very strong dependence on sstrate biasVx , with markedly stronger charging atVx

53 V than atVx52 V, using hot carriers in the darkFor nFETs, little difference is observed for hot carriein the dark, even as the substrate bias is changed fVx522 V to Vx526 V.

Each of the key features noted for HfO2 nFETs has also beeobserved in Al2O3 nFETs, as shown in Ref. 6. Also, to checthat illumination alone was not responsible for the enhantrapping, no enhanced trapping was observed under ccarrier stressing with the light on; this point is further dcussed in Sec. III G.

B. Wavelength dependence in nFETs

The enhanced charge trapping innFETs observed undeillumination depends on the wavelength, or, equivalenphoton energy, of the light. We can restrict the maximuphoton energyhnc by using Long Pass Schott colored glafilters which pass all incident wavelengths above a cutwavelengthlc5c/nc .22 A study of the wavelength dependence is complicated by the fact that more restrictive~higherlc) filters require higher source intensities to generate rsonable gate currents, but the charge trapping is a stfunction of light intensity even when plotted versus injectcharge to account for differences in gate current. In ordecircumvent this dependence on source intensity and sonly the wavelength dependence, we utilize a novel tenique in which the gate current at given substrate biasVx isfixed by the p–n injector forward bias of 1 V. We thenilluminate the device with light of sufficiently weak intensitthat the gate current is independent of thelc of the filterused.

Figures 2~a!–2~c! shows results for HfO2 nFETs. Thethreshold voltage shiftDVt ~measured from theC–V nearweak inversion! is plotted as a function of injected charg

Downloaded 20 Sep 2013 to 152.14.136.96. This article is copyrighted as indicated in the abstract.

r-e

b-le

lly

tr-

-

rseeld

b-

m

dld-

,

n

a-ng

tody-

Qinj at different combinations ofVx andhnc . Since the pulsetimes used were the same for all curves in Fig. 2~sevenpulses, doubling in time, starting from 20 s!, the lack ofdependence of gate current onlc can be seen by the nea

FIG. 2. Threshold voltage shift in HfO2 nFETs at different maximum pho-ton energies for substrate voltages of:~a! 22.25 V, ~b! 22 V, and ~c!21.75 V, along with fits to a power-law behavior. Charge trapping is spressed as the maximum photon energy is reduced. The inset of~b! showsthe mean gate current during each stressing pulse, indicating concusuppression of the increase in gate current with time.

Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions

inereen

in-

y

rge

-he

rd

glueserentthebe-

(

p-rre-a

ast,ses

nd-g

ate

s

e oftheat a

u

1731J. Appl. Phys., Vol. 94, No. 3, 1 August 2003 Kumar et al.

alignment of the data points along theQinj axis. At Vx

522.25 V, enhanced charge trapping evolves from beclearly visible whenhnc52.7 eV to being close to the curvtaken in the dark when the maximum photon energy isduced tohnc51.5 eV. As the electron energy is decreasby lowering uVxu, the required photon energy to see ehanced charge trapping is correspondingly increased.

Figure 3~a! shows that similar results are obtainedAl2O3 nFETs, although theQinj scale where effects are observed is somewhat higher than in wafer H1.~As discussedin Sec. III F, DVfb'DVt for Al2O3 nFETs.) Figure 3~b!shows the cutoff behavior in a slightly different way, bcomparing curves taken using the same filter (lc5455 nm,or hnc52.7 eV) at different substrate voltages.

In order to find an energy scale for the enhanced chatrapping phenomenon, we can characterize the data bypirical power-law fits of the formAQinj

b , as shown in Figs.2–3. Figure 4 shows scatter plots ofA and b ~inset! as afunction of some measure of total energy,hnc1euVxu, forboth HfO2 and Al2O3 nFETs. For simplicity, we have neglected the voltage drop in the bulk which will reduce t

FIG. 3. Flatband voltage shift in Al2O3 nFETs ~a! at different maximumphoton energies and the same substrate biasVx522.5 V and~b! at differentsubstrate biases and the same maximum photon energyhnc52.7 eV. Sup-pression of charge trapping occurs either through reducing the maximphoton energy or through reducing the substrate bias.

Downloaded 20 Sep 2013 to 152.14.136.96. This article is copyrighted as indicated in the abstract.

g

-d-

em-

energy of the incident hot electrons beloweuVxu and eventu-ally limit it. 1 Also note that the units of the coefficientAdepend onb, a complication we ignore here sinceb changesonly weakly. The HfO2 nFETs show a clear turn-on behavioin A at aroundEtot'3.8 eV whileb also shows a weak trento increase. A weak increase inA is also observed for Al2O3

nFETs, but the scale ofA is small compared to that for HfO2nFETs.

C. Gate current as a probe of gate insulator charge inHfO2 nFETs

The inset of Fig. 2~b! plots the mean gate current durineach of the seven stressing pulses for three different vaof the maximum photon energyhnc . The enhanced chargtrapping is accompanied by an increase in the gate curwith time which can also be suppressed by restrictingmaximum photon energy. Figure 5 compares the typicalhavior of the gate current with time in a HfO2 nFET duringa stress consisting of several pulses under coldVg

51.75 V in the dark! and hot (Vx524 V in the light! stress-ing conditions. The discontinuities correspond to interrutions of the stress to perform aC–V measurement. Undecold stressing conditions, electrons which get trapped in pexisting sites repel further incident electrons, resulting inweak decrease of the gate current with time. In contrunder hot stressing conditions, the gate current increastrongly with time. We interpret this increase as correspoing to the buildup of positive charge in the dielectric durinhot-electron stressing.

In contrast to the steep increase in the hot-electron gcurrent during stress, the tunneling gate currentI g

cold as afunction of gate voltageVg is almost unchanged. This ishown in the inset of Fig. 5 which plotsI g

cold as a function ofVg before and after the hot-electron stress. The absencchange in the tunnneling gate current is consistent withobservation that the hot-electron gate current restartsmuch lower value following each interruption for aC–V

m

FIG. 4. Scatter plots of power-law coefficientA and exponentb ~inset! as afunction of total energyEtot5hnc1euVxu extracted for HfO2 and Al2O3

nFETs. Note that the HfO2 nFETs show a clear turn-on behavior inA ataroundEtot'3.8 eV.

Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions

ariu

infou

n-

,nd

m-

ngob-

andlat-

e ofO

theons,s toark

ial

bem

eheof

hsion

ndm

note

the-inuseolesing

coo

as

e

usstch

reses

1732 J. Appl. Phys., Vol. 94, No. 3, 1 August 2003 Kumar et al.

measurement, suggesting that most of the positive chexits the dielectric when the stress is removed. This dipthe hot-electron current after each interruption also helpsdeem as unlikely another possible explanation for thecreasing gate current: the formation of a defect chain,which we would expect the current to resume its previovalue following each interruption in the stress.

D. Temperature and substrate bias dependence inHfO2 nFETs

Figure 6~open symbols! shows the temperature depedence of charge trapping under hot stressing conditions~inthe light! for HfO2 nFETs. As a function of injected chargethe Vt shift decreases as the temperature is increased. Ucold tunneling conditions~in the dark!, shown in the inset, a

FIG. 5. Time dependence of gate current under cold and hot stressingditions. Under cold stressing conditions, the gate current decreases mtonically with time; under hot stressing conditions, the gate current increrapidly with time but dips sharply following each interruption for aC–Vmeasurement.~Inset! Tunneling gate current as a function of gate voltagshowing little change before and after the hot-carrier stress.

FIG. 6. Temperature dependence of charge trapping in HfO2 nFETs underhot and cold~inset! stressing conditions, and substrate bias dependenceder hot stressing conditions. The saturation behavior under hot streconditions as the injected charge is increased can be fitted to a streexponential function.

Downloaded 20 Sep 2013 to 152.14.136.96. This article is copyrighted as indicated in the abstract.

gens-rs

er

similar decrease of the charge trapping with increasing teperature is observed. Note also that theQinj scale is largerwhile theDVt scale is smaller, reflecting the weaker trappiunder cold stressing. Two simple explanations for theserved temperature dependence are thermal detrappingthe lowering of the electron temperature with increasingtice temperature.

Figure 6 also shows the substrate bias dependenccharge trapping under hot stressing conditions for Hf2

nFETs. The charge trapping shows a sharp increase assubstrate bias, and thus the energy of the incident electris increased. This strong dependence under illumination ibe contrasted with the weak dependence found in the d@see Fig. 1~a!#, confirming that light appears to be essentfor enhanced charge trapping.

As the substrate voltage is increased toVx523 V, Fig.6 shows that a clear saturation inDVt is observed at highQinj . As suggested in Ref. 13, we find that the data canempirically fit to a stretched exponential function of the forDVt5DVtm$12exp@2(s0Qinj /e)b#%, with three fitting pa-rameters whose values are shown in Table II.@Note from Fig.1~a! that the saturation behavior observed inDVt is not ob-served inDVfb even atVx526 V.] In Ref. 13, which stud-ied filling of pre-existing traps by cold carriers,DVtm is in-terpreted as the saturationVt shift, s0 as the average capturcross section, andb as an exponent which characterizes tdistribution in capture cross sections. Note that in the limitweak injections0Qinj /e!1, the stretched exponential witthree fitting parameters reduces to the power law expresused above withA5DVtm(s0 /e)b. We emphasize that weonly use the fits to characterize empirically our curves aremark that the physical interpretation may be different frothat in Ref. 13 since we are dealing with trap generation,just filling of pre-existing traps. As shown in Table II, wfind that s0 is an increasing function of temperature atVx

523 V and a decreasing function ofuVxu at 25 °C.

E. Trap creation in a stressed pFET

We now consider an experimental test of whetherenhanced charge trapping seen inpFETs represents a reversible filling of pre-existing trap states at high energies or,fact, real damage in the gate insulator. We choose topFETs here because we can controllably introduce hot hinto the insulator and then probe the damage by studyelectron trapping in accumulation. In our earlier work6 it was

n-no-es

,

n-inged

TABLE II. Parameters of stretched exponential fit at various temperatufor wafer H1 withVx523 V under hot stressing conditions. Uncertaintirepresent quality of fit to data.

Temperature(°C) Vx ~V! DVtm ~V! s0 (cm2) b

25 23 0.31460.002 (1.0460.05)310216 0.4360.0175 23 0.18260.002 (2.0160.13)310216 0.5260.02

120 23 0.09960.001 (6.1260.40)310216 0.5360.0225 24 0.47760.004 (8.0560.62)310217 0.3560.0125 25 0.60960.008 (2.9160.38)310217 0.2660.01

Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions

loioh

pi

or

ca

ntasain

min

g

rad

nodth

safendlyfe

th.

pingthe

gates

ofr

telyicalfees-ight

in-

rent

cese

ll

y100nce

ofe

arccu-

ect

ss

d wa

ho

b

e

.

1733J. Appl. Phys., Vol. 94, No. 3, 1 August 2003 Kumar et al.

found that the flatband voltage of apFET which had under-gone hot-hole stressing in the dark could be restored to cto its unstressed value by shining light on it. This observatsuggests that the existence of persistent damage done byhole stressing can be tested by comparing electron trapin accumulation on an unstressedpFET to that in apFETwhich has undergone hot-hole stressing followed by resttion of the flatband voltage using light.

Figure 7 shows the results of two such experiments,ried out on two slightly different wafers, H2 and H28, andunder slightly different conditions. In the first experime~open symbols!, the charge trapping in accumulation wmeasured on two different devices: an unstressed devicea device which had been first subjected to hot-hole stressThe results show clearly that the charge trapping in acculation is worse on a device subjected to hot hole stress~open circles! than on a fresh sample~open stars!. In thesecond experiment~solid symbols!, the samedevice wasused both before and after hot-hole stressing to convincinverify the increased damage after stress.~The inset shows thecharging during hot-hole stressing followed by the restotion using light.! As in the first experiment, the increaseslope ofDVfb as a function of log(Qinj) is indicative of in-creased electron trapping after hot-hole stressing. Wethat wafer H28 exhibited an initial negative flatbanshift following the first measurement in accumulation, bobefore ~approximately210 mV) and after~approximately217 mV) the hot-hole stressing, giving rise to the reverof the positions of the two curves relative to those for waH2. Also noteworthy is that on wafer H2, positive flatbashifts, to the right of the initial flatband, were commonobserved, clearly signifying new damage; however, on waH28, the flatband was not observed to shift to the right ofinitial flatband, even after long-term~48 h! measurement

FIG. 7. Comparison of electron trapping in unstressed and hot-hole streHfO2 pFETs using two different experiments. The first experiment~opensymbols! uses two different devices, an unstressed one and one stressehot holes. The second experiment uses the same device, before andhot-hole stressing. In both cases, the increased slope following hot-stressing is indicative of hot-hole induced electron traps.~Inset! Chargetrapping during hot-hole stressing followed by restoration of flatbandillumination.

Downloaded 20 Sep 2013 to 152.14.136.96. This article is copyrighted as indicated in the abstract.

senot-

ng

a-

r-

ndg.u-g

ly

-

te

lr

re

Thus both experiments suggest increased electron trapfollowing hot-hole stressing, although the exact nature ofdamage created is not clear.

F. Generation of interface states in nFETs

Stretchout of the high frequencyC–V characteristic is asignature of the generation of states at or near the Si–insulator interface.23 To quantify the stretchout, Fig. 8 plotthe flatband and threshold voltage shifts as a functionsubstrate bias for HfO2 and Al2O3 nFETs before and aftehot-electron stressing~using light!, with the stressing timeschosen such that the injected charge is very approximathe same for each material. The two insets compare typhigh-frequency~100 kHz! C–Vs for the dielectric stacks othe two materials. In HfO2 nFETs, a clear stretchout of thC–V curve is observed, with the left side near flatbandsentially unchanged after stress and a large shift on the rside near threshold. The stretchout shows a pronouncedcrease with increasinguVxu. In contrast, Al2O3 nFETs showa shift on both sides with much weaker stretchout appaonly at higheruVxu.

Having clearly established the generation of interfastates in HfO2 nFETs by the pronounced stretchout, we uthe combined high-low frequency capacitance method23 toestimate the interface trap densityD it before stressing as weas the changeDD it resulting from hot-electron stressing.24

Although the accuracy of this method is limited both bexclusion of traps outside our chosen frequency range ofHz–100 kHz and by errors in the accumulation capacitaCacc, we can nonetheless obtain a lower-bound estimatethe interface trap densityD it as well as some insight into thenergetic position of created traps. Figure 9 showsD it(E) asa function of position in the gapE–Ei (Ei is the intrinsiclevel at the interface!. The trap density is found to peak nethe conduction band edge. Errors in the capacitance in amulationCacccan lead to considerable uncertainty inD it(E),as indicated by the error bars in Fig. 9 which show the effof changingCacc by 63% resulting in a65% change in the

ed

ithfterle

y

FIG. 8. Flatband and threshold voltage shifts in HfO2 and Al2O3 nFETs.Insets show typicalC–V characteristics for each dielectric. Note that thHfO2 nFETs show pronounced stretchout (DVt@DVfb) following hot-electron stressing while Al2O3 nFETs show a shift and a weak stretchout

Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions

Th

dg

izt-de

ts

ac

oive

iar

rgx

a

anrso

tld

eceoei

ch

hatarlyby

de-arge

ntsd

ues

f

ultsob-toringf the

n-

ein-

bovesetforthe

nce

nclo

eri-rdes

1734 J. Appl. Phys., Vol. 94, No. 3, 1 August 2003 Kumar et al.

Si band gap from our base measured value of 1.19 eV.inset shows the net change inD it following hot-electronstressing~in the light atVx524 V for 200 s!. The generatedstates are mostly within 60 meV of the conduction band eand have an integrated density of (1.360.2)31012 cm22,where the error estimate reflects the uncertainty inCacc.

In general, the interface states can be loosely catagoras ‘‘fast’’ states, which can quickly follow the applied volage, and ‘‘slow’’ states, which give rise to hysteresis unlong-duration bias but do not cause distortion of theC–Vcharacteristic.25 We perform a charge transfer experimen25

to probe the existence of slow states, which have been aciated with ‘‘anomalous positive charge,’’25–27 bearing inmind that the interpretation of results in a bilayer gate stmay not be as straightforward as in SiO2 . We begin with anAl2O3 nFET where hot-electron stressing (Vx526 V, 200s, in the light! results in both a weak stretchout and a shiftthe C–V characteristic. We then apply a nondestructnegative bias~which causes no damage by itself! to slowlydepopulate the created states, perform a forwardC–V sweepto a nondestructive positive bias, wait at the positive ballowing some repopulation of the slow states, and do averseC–V sweep. We repeat this procedure, allowing chato be transferred in and out of the slow states. We use a ficharging hold time of 10 s at the positive bias (Vg

50.7 V), but progressively double the discharging timethe negative bias (Vg521.1 V), starting from 250 s andgoing up to 256 000 s. Figure 10 shows the flatbandthreshold voltage shifts during both the forward and reveC–V sweeps as a function of the total discharging time. Feasy comparison to the initial flatband shiftDVfb

0 createdfrom hot-electron stressing, we normalize all voltage shiftsDVfb

0 574 mV. First, we note that the normalized threshovoltage shifts start above unity, signifying stretchout. Sond, we see that the separation between forward and revsweeps increases with time for both flatband and threshThird, at long detrapping times, the flatband shift becomnegative, which could arise from the created states becom

FIG. 9. Density of interface traps in unstressed HfO2 nFET measured usingthe combined high-low frequency capacitance method. Inset: Change iterface trap density resulting from hot-carrier stressing is concentratedto the conduction band edge.

Downloaded 20 Sep 2013 to 152.14.136.96. This article is copyrighted as indicated in the abstract.

e

e

ed

r

so-

k

f

se-eed

t

der

o

-rseld.sng

positively charged. However, it must be noted that at sulong times, the detrapping voltage of21.1 V was found tocause some shifting in a virgin sample comparable to tseen in the stressed sample. Thus, although we can clesee the population and depopulation of the slow stateselectrons, we unfortunately cannot determine whether thefects, once emptied, can also assume a positive chstate.25–27

We have also tried similar charge transfer experimeusing HfO2 nFETs. Interestingly, we find that the flatbanand threshold voltages remain nearly fixed at their valfollowing hot-electron stressing~i.e., DVfb'0 and DVt

large!. Thus, in contrast to Al2O3 nFETs, the existence oslow states~or perhaps their time constants! remains an openquestion in HfO2 nFETs.

G. Bias and polarity dependence in the coldtunneling regime in nFETs

We devote this final subsection on experimental resto test whether enhanced charge trapping can also beserved using cold electrons, injected by the gate insulafield. Some evidence for trap creation under cold stressconditions is suggested both by the strong dependence oVt or Vfb shift on stressing gate voltage for the sameQinj

13

and by our observation of stretchout in theC–V character-istic. Figure 11~a! shows the charge trapping in the cold tuneling regime at gate biases close toVfb11.1 V for wafer H1(Vfb520.75 V) and wafer A1 (Vfb520.55 V). It is seenfor both HfO2 and Al2O3 nFETs that the onset of chargtrapping, which has a clear gate voltage dependence, isdeed close to an energy corresponding to a Si bandgap aflatband. As discussed further in Sec. IV, this voltage onwould correspond to the bare minimum energy requiredan incident cold electron to create an electron-hole pair inpolysilicon gate by impact ionization.

In addition to a bias dependence, a polarity dependehas also been observed in Al2O3

10 and ZrO27 gate stacks.

Typically a smaller shift is observed for gate injection (Vg

,0) than for substrate injection (Vg.0). In Fig. 11~b! we

in-se

FIG. 10. Flatband and threshold voltage shifts from charge transfer expment on Al2O3 nFET. The widening window of the shifts between forwaand reverseC–V sweeps illustrates filling and emptying of interface statfollowing hot-electron stressing.

Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions

sagbeicle

ineth

ed

edaucgtraatera

ic-tionofior

hole

ere.va-i-ron

of

earndtedtun-ing,

ot-oles

een

the

c-oncanor

hotigh-

ively

high

ail.hot

w-

lt-in

u-ar

1735J. Appl. Phys., Vol. 94, No. 3, 1 August 2003 Kumar et al.

show the polarity dependence for HfO2 , taken at gate biasewhich are chosen to be symmetric about the flatband voltVfb520.75 V. Again, the charge trapping is seen tostronger for substrate injection than for gate injection, whwould be consistent with the larger energy barrier for hoentering from the substrate to the oxynitride.

However, in contrast to the results for hot electronsnFETs, shining light on the sample causes no enhancemof the charge trapping under cold stressing conditions. Asgate voltage is increased, a substantial current is measurthe substrate@see inset of Fig. 11~b!#, as reported previouslyfor SiO2 .28 If indeed this current consists of generatholes,28 one might expect the charge trapping to showenhancement with light at such a bias; however, no scorrelation between substrate current and charge trappinthe cold tunneling regime was seen even when the subscurrent was significant. Thus we conclude that charge trping under cold stressing conditions is probably dominaby the filling of pre-existing traps rather than by trap genetion.

FIG. 11. ~a! Bias dependence under cold stressing conditions for HfO2 andAl2O3 nFETs showing onset of charge trapping nearVg5Vfb11.1 V. ~b!Polarity dependence for HfO2 nFETs under cold stressing conditions, shoing stronger charge trapping under substrate injection (Vg.0) than undergate injection (Vg,0). ~Inset! Substrate current as a function of gate voage. Even when the substrate current is large, enhanced charge trappnot observed under cold stressing conditions.

Downloaded 20 Sep 2013 to 152.14.136.96. This article is copyrighted as indicated in the abstract.

e

hs

nteat

nhinte

p-d-

IV. DISCUSSION

We explain our results using a tentative, qualitative pture of the chain of processes which lead to the degradaof the high-k stack. Considering that even in the caseSiO2—an insulator whose hot-carrier degradation behavhas been studied for over 3 decades—the anodeinjection17 and the hydrogen-release1,16 models are stilldebated,2 ‘‘tentative’’ and ‘‘speculative’’ are adjectives whichmust necessarily accompany the discussion presented h

We begin by recapping the key experimental obsertions in HfO2 nFETs for hot-electron stressing under illumnation. First, during hot-electron stressing, the hot-electgate current increases, which we attribute to the builduppositive charge. Second, theC–V characteristics show littleflatband shift after stressing, implying charge neutrality nflatband. Third, theC–V characteristics are stretched out ashow a large threshold shift after stressing, implying creainterface states with trapped negative charge. Fourth, theneling gate current is essentially unchanged after stressagain suggesting charge neutrality.

Figure 12 illustrates a speculative picture of how helectron stressing may be inducing damage using hot has a precursor for the case of HfO2 nFETs. We focus here onthe role of interface states, since their creation has bclearly shown; however, we strongly suspect~as discussedfurther below! that accompanying damage elsewhere inhigh-k stack is also occurring. Figure 12~a! illustrates hot-hole injection from the polysilicon gate into the gate dieletric during hot-electron stressing. After a hot electrtraverses the gate insulator stack to the gate electrode, itgenerate an electron–hole pair either by impact ionizationby excitation of a surface plasmon.29 Assisted by a photon, ahole from the gate can be injected back into the high-k di-electric where it causes damage. Alternately, the incidentelectron can absorb a photon, enabling it to generate a henergy hole. Figures 12~b!–12~c! shows newly created trapstates which are charge neutral near flatband and negatcharged~filled with electrons! near threshold. In Fig. 12~d! itis shown how these trap states can be emptied underfields in strong inversion.

We now discuss this speculative picture in greater detFirst, we consider the creation of electron–hole pairs by

g is

FIG. 12. Band diagram showing schematically:~a! positive charge trappingduring hot-electron stressing;~b! created interface states which are unoccpied ~neutral! at flatband;~c! interface states being filled by electrons nethreshold; and~d! emptying of interface states at high insulator field.

Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions

ee

o

h-e

ri

ffs

nmesiri

tco.lele

eaevee

edt

isdof

autht

-agedtndencaas-hoeno

w

n

-

ond-

oldtageerthe

ble.vegee ofthe

de

a-ts,

e-

ced

le

gea

eritin

ofon,

dhold

itiveharge

1736 J. Appl. Phys., Vol. 94, No. 3, 1 August 2003 Kumar et al.

electrons entering the anode by impact ionization. Indethe energy barrier seen by holes injected from the anodcalculated to be 3.4 eV in bulk HfO2 and 4.9 eV in bulkAl2O3

5 ~although measured as low as 3 eV in thin filmsAl2O3)18 compared to 5 eV in SiO2 .23 A second possibilityfor hole generation is via a process mediated by Si/higkinterface plasmons for which the energy of excitation dcreases with dielectric constantk as (k11)21/2.29 Althoughwe are hesitant to draw conclusions based only on comparesults from single wafers of HfO2 and Al2O3 , we note thatthe turn-on behavior of the power-law coefficientA at lowerenergies in HfO2 compared to Al2O3 ~Fig. 4! is consistentwith our expectations based on the lower valence band oand higher dielectric constant of HfO2 . Also, the estimatedthreshold energy ofEtot'3.8 eV for hot-electron damage iHfO2 should be contrasted with the 5 eV threshold for daage in SiO2 .30,31 Finally, we note that the smaller valencband offset and higher dielectric constant on the anodeof the gate insulator stack may also be related to the poladependence observed here and in Refs. 7 and 10, buinability to see enhanced charge trapping effects understressing conditions leaves this relationship inconclusive

Next, we consider the nature of the damage by hot hoHot holes in SiO2 have been widely studied as a possibprecursor for damage in SiO2 .17 In this picture, a hot holerecombines with an electron, and the resultant energy relbreaks bonds in the SiO2 , generating positive chargcenters—also called ‘‘slow states’’ or ‘‘anomalous positicharge’’25–27—on which electrons are trapped. Although wmight expect from this previous work that hot-hole inducdamage occurs in the oxynitride underlayer, damage atinternal ~oxynitride-high-k! interface is also possible asdamage in the high-k dielectric itself. Indeed, Afanas’ev anStesmans18 have recently reported that optical injectionelectron–hole pairs in Al2O3 and ZrO2 results in positivecharging, suggesting trapping of holes.

In order to further test the hot-hole hypothesis, we cask whether the amount of positive charge required to cathe increase in the gate current correlates roughly withnegative charge in the created electron traps giving rise topositive threshold shiftDVt . To do this, we convert the hypothesized positive charge to an equivalent gate voltshift, DVgeq, by finding the change in gate voltage requirunder cold stressing conditions~with all other electrodes aground! to effect the same increase in gate current seen uhot stressing conditions. A comparison of these two voltagDVgeq andDVt , can help us verify the correlation betweethe positive charge and the created electron traps. TheDVgeq<DVt would suggest a correlation whereas the cDVgeq!DVt would imply too little positive charge to account for the traps required to create the observed thresshift. Figure 13 shows the results of plotting the equivalgate voltage shift as a function of the measured threshvoltage for 23 various hot-carrier stress experiments onfer H1, with substrate voltages ranging fromVx521.5 V toVx526 V. Here the equivalent gate voltage shiftDVgeq hasbeen computed from the ratio of final to initial gate curreI gf /I gi using the experimentally obtained relationDVgeq

5(0.13 V) ln(Igf /I gi), valid for currents in the range of in

Downloaded 20 Sep 2013 to 152.14.136.96. This article is copyrighted as indicated in the abstract.

d,is

f

-

ng

et

-

detytheld

s.

se

he

nsee

he

e

ers,

see

ldtlda-

t

terest. The cluster of single points near the origin correspto low substrate biasuVxu<2.25 V, as in the wavelengthdependent experiments! while the lines connect points fromexperiments at higher substrate bias with large threshshifts measured at several intervals. We see that the volratio DVgeq/DVt varies from approximately unity near thorigin ~weak stressing! to roughly 0.3–0.5 under highestressing conditions, showing that a correlation betweenpositive charge and the electron traps is indeed reasona

We thus summarize the observations which we belieconstitute strong ‘‘circumstantial evidence’’ that the damaconsists of defects whose origin stems from the presencholes in the dielectric and whose nature is similar tonature of the slow states observed in SiO2 films:25–27

~i! the sharp increase of the gate current;~ii ! the generation of interface traps at the Si/oxynitri

interface, indicated by theC–V stretchout;~iii ! the possibility of altering quasireversibly the occup

tion of at least a fraction of the generated defectestified by the hysteresis experiments on Al2O3

nFETs ~Fig. 10!;~iv! the fact that the amount of the positive charge d

duced by the shiftDVgeq is quantitatively of the sameorder of magnitude as the negative charge dedufrom the threshold voltage shift; and

~v! the similarity of the damage observed after hot-hostressing inpFETs.

As mentioned earlier, our inability to clearly observe chartransfer in HfO2 nFETs at long time scales may indicatedifference in the nature of the slow states in HfO2 comparedto those in Al2O3 and SiO2 .

Two further aspects of the nature of the damage mdiscussion. First, previous studies of hot-hole damageSiO2 point to the generation of electron neutral traps.32,33

Although we are cautious about the physical interpretationthe fitting parameters0 used here as a capture cross secti

FIG. 13. Equivalent gate voltage shiftDVgeq due to positive charge deducefrom increase in hot-carrier gate current as a function of measured thresvoltage shiftDVt for 23 experiments on HfO2 nFETs. The ratioDVgeq/DVt

is in the range 0.3–1, suggesting that a correlation between the poscharge giving rise to an increase in the gate current and the negative cgiving rise to the threshold voltage shift is reasonable.

Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions

stth

anfla

teo

ar

ennl-

sghc idnic-a

en

pi

e,y

ctaarto

hof

p-reni

c

ar

enan

ft-Di-

u-nce

p.

ett.

J.

roe-

nd

nt.

l.

htly

ible

s

ent,

k,

1737J. Appl. Phys., Vol. 94, No. 3, 1 August 2003 Kumar et al.

we note that its 10216 cm2 range is consistent with previoustudies of neutral traps34 and their generation by hoholes.32,33 Second, based on Fig. 12, we can speculateexistence of an amphoteric defect which is filled withelectron near threshold, is empty and thus neutral nearband, and can be filled with a hole~devoid of an electron!during stress when high-energy holes are available. Alnately, a more complicated picture is that electron and htraps are independently generated. Unfortunately, our chtransfer experiments have not yielded enough informationdistinguish between these two possibilities.

Finally, we comment on the failure to observe enhanccharge trapping in the presence of light using cold electroThis may be because in cold-electron injection, the stropolar interaction of low-energy electrons with longitudinaoptical phonons in the high-k film can prevent the carrierfrom acquiring significant kinetic energy, except at very hielectric fields. In contrast, hot electrons enter the dielectrithe regime in which scattering with LO phonons is alreareduced~because of itsE21/2 dependence on the electroenergy E)35 and thus may undergo quasiballisttransport—as seen in SiO2

36—and reach the anode with significant kinetic energy. Therefore, at the anode, the smhigh-k/Si barrier height and the quasi-thermal electronergy in the high-k insulator will result in a small electronkinetic energy in the polysilicon gate, thus effectively supressing the pair-generation process. This is the regimevestigated by Zafaret al.:13 Only a weakC–V stretchout isobserved, and the gate current decreases gently with timthat filling of pre-existing traps dominates over filling of antraps created. Thus, although contrary to our initial expetions, it is perhaps not surprising that the enhanced chtrapping with light is not observed even at high gate insulafields.

V. CONCLUSION

In conclusion, a strong case has been made forcarrier degradation in which hot holes act as a precursorcreating damage in high-k dielectric stacks. This damage apears to be qualitatively different from, and potentially mosevere than, the charge-trapping effects frequently seeexperiments using cold-carrier injection, which are domnated by filling of pre-existing traps. We expect these effeto be of more pressing importance inpFETs, where hot holescan be generated under normal biasing conditions. InnFETs,careful attention to the supply of hot holes should be givena design consideration. Of particular interest for future wowill be a comparison of the results obtained here to thosmetal-gated samples where anode holes should origionly from surface plasmons and not from impact ionizatio

Downloaded 20 Sep 2013 to 152.14.136.96. This article is copyrighted as indicated in the abstract.

e

t-

r-legeto

ds.g

ny

ll-

-n-

so

-ger

t-or

in-ts

skinte.

ACKNOWLEDGMENTS

The authors thank Barry Linder and Gil Singco for soware development; Doug Buchanan, Susan Cohen, DanMaria, Robert Miller, Jim Stathis, and Sufi Zafar for stimlating discussions; Jim Glownia and Jeff Kash for assistawith optics; and Steve Cohen for technical help.

1D. DiMaria, E. Cartier, and D. Arnold, J. Appl. Phys.73, 3367~1993!.2M. Green, E. Gusev, R. Dengraeve, and E. Garfunkel, J. Appl. Phys.90,2057 ~2001!.

3E. Gusev,et al. Tech. Dig. - Int. Electron Devices Meet.2001, 451.4G. Wilk, R. Wallace, and J. Anthony, J. Appl. Phys.89, 5243~2001!.5J. Robertson, J. Vac. Sci. Technol. B18, 1785~2000!.6A. Kumar, T. Ning, M. Fischetti, and E. Gusev, Tech. Dig. VSLI Sym2002, 152.

7M. Houssa, V. Afanas’ev, A. Stesmans, and M. Heyns, Appl. Phys. L79, 3134~2001!.

8J. Chavez, R. Devine, and L. Koltunski, J. Appl. Phys.90, 4284~2001!.9Z. Xu, M. Houssa, R. Carter, M. Naili, S. D. Gendt, and M. Heyns,Appl. Phys.91, 10127~2002!.

10A. Kerber, E. Cartier, R. Degraeve, L. Pantisano, P. Roussel, and G. Gseneken, Tech. Dig. VLSI Symp.2002, 76.

11J. Wang, S. Chiao, C. Lee, T. Lei, Y. Lin, M. Wang, S. Chen, C. Yu, aM. Liang, J. Appl. Phys.92, 3936~2002!.

12M. Houssa, J. Autran, A. Stesmans, and M. Heyns, Appl. Phys. Lett.81,709 ~2002!.

13S. Zafar, A. Callegari, E. Gusev, and M. V. Fischetti, Tech. Dig. - IElectron Devices Meet.2002, 217.

14M. Fischetti, D. Neumayer, and E. Cartier, J. Appl. Phys.90, 4587~2001!.15G. Lucovsky, T. Wu, H. Niimi, V. Misra, J. Misra, and J. Phillips, App

Phys. Lett.74, 2005~1999!.16R. Gale, F. Feigl, C. Magee, and D. Young, J. Appl. Phys.54, 6938~1983!.17S. Lai, J. Appl. Phys.54, 2540~1983!.18V. Afanas’ev and A. Stesmans, Appl. Phys. Lett.80, 1261~2002!.19T. Ning and H. Yu, J. Appl. Phys.45, 5373~1974!.20J. Verwey, J. Appl. Phys.44, 2681~1973!.21This condition sometimes requires the gate voltage to be chosen slig

below threshold during hot-carrier stressing.22The intensity of the halogen lamp source varies smoothly in the vis

range used for the filter cut-on wavelengths.23E. Nicollian and J. Brews,MOS (Metal Oxide Semiconductor) Physic

and Technology~Wiley, New York, 1982!.24The transistor source and drain are grounded during the measurem

providing a source of minority carriers in inversion.25S. Lai and D. Young, J. Appl. Phys.52, 6231~1981!.26M. Fischetti, R. Gastaldi, F. Maggioni, and A. Modelli, J. Appl. Phys.53,

3136 ~1982!.27F. Feigl, D. Young, D. DiMaria, S. Lai, and J. Calise, J. Appl. Phys.52,

5665 ~1981!.28Z. Weinberg and M. Fischetti, J. Appl. Phys.57, 443 ~1985!.29M. Fischetti, Phys. Rev. B31, 2099~1985!.30D. DiMaria and J. Stasiak, J. Appl. Phys.65, 2342~1989!.31D. DiMaria, J. Appl. Phys.86, 2100~1999!.32I. Chen, S. Holland, and C. Hu, J. Appl. Phys.61, 4544~1987!.33H. Uchida and T. Ajioka, Appl. Phys. Lett.51, 433 ~1987!.34D. DiMaria, in The Physics of SiO2 and Its Interfaces, edited by S. Pan-

telides~Pergamon, New York, 1978!, pp. 160–178.35H. Frohlich, Proc. R. Soc. London, Ser. A160, 230 ~1937!.36D. DiMaria, M. Fischetti, J. Batey, L. Dory, E. Tierney, and J. Stasia

Phys. Rev. Lett.57, 3213~1986!.

Reuse of AIP content is subject to the terms at: http://jap.aip.org/about/rights_and_permissions