Mechanism of negative bias temperature instability in CMOS devices
Shared by: yho59756
Mechanism of Negative Bias Temperature Instability in CMOS Devices: Degradation, Recovery and Impact of Nitrogen S. Mahapatra'", M. A. Alam', P. Bharath Kumar', T. R. Daleil and D. Saha' 'Department of Electrical Engineering, Indian Institute of Technology Bombay, 400076, India 'School of Electrical and Computer Engineering, Purdue University, West Lafayette, IN, USA 'Phone: + 1-22-25720408, Fax: CY 1-22-25723707, Email: firstname.lastname@example.org Y Abstract of HH depends on gate current (IG)and quantum yield (QY) NBTI mechanism is discussed for a wide range of stress of 11. Fig.2 also lists the impact of experimental parameters conditions. Conditions for interface and bulk-trap generation (VG, VB, T and T P ~ Y ) HH generation. Note that stress bias on are shown. The bias, time and temperature dependence of during accelerated aging must be carefully chosen for proper interface-trap buildup and recovery are discussed using the estimation of NBTI (to minimize ANoT contribution). framework of the R-D model. The AC frequency dependence Fig.3 plots AVT versus ADm for a wide range of Eox. Due and impact of gate oxide nitridation are also discussed. to thin TPHY, stress VG remains low for the entire Eox range. Excellent correlation over a wide range of Eox (V,) suggests Introduction absence of ANoT. Otherwise, strong VGacceleration of ANoT Negative Bias Temperature Instability (NBTI) of PMOS would result in higher AVT for a given ADm with increasing parameters (threshold voltage, linear and saturation current, VG. Therefore, AVT is due to ADm alone for lower stress VG. gate-drain capacitance [ 1-31) is an important reliability issue for digital  as well as analog [ 5 ] CMOS circuits. In spite of . 16 extensive efforts in characterization (1-141 and modeling [15- IS], the basic NBTI mechanism is not yet fully understood. 1.2 Issues of interest are: (A) degradation, i.e., interface (Nm) and v, bulk (NOT)trap generation and their bias (VG), temperature 6 0.8 2 (TI, and time (t) dependence, (B) Nm and NOTrecovery and " their VG, T and AC frequency (0 dependence, and ( C ) impact h 0.45 of nitrogen on Nrr and Nm. Proper understanding of NBTI mechanism will help determine ( I ) reliability budget for any 0.0 technology node, (2) proper burn-in and test conditions, (3) TCAD and SPICE models, and (4) process parameters for 0 O ~ ' l O ' l0 1 1 0'1 051 1 O~'lOol01 1 1 0 0 O6 '1 ' 0 1 '0 ' 0' 0 0 1 1 '' ' ' ' stress time (s) NBTI control. This paper addresses the above issues (A-C). - f Fiaure 1. fLHSI Time evolution o VT shift for stress at various , , VO (symbols). Calculated ANor contributions are shown (lines). Experimental details f (RHS) Time ewlution o SILC for various stress VG. Experiments were performed on non-nitrided gate oxide MOSFETs with varying oxide thickness (TPHY) values, unless mentioned otherwise. Transfer I-V and charge pumping (CP) measurements were done by periodic interruption of stress. VT extraction was done using the constant current method. ANrr was extracted from CP current, and ADm = A",/AE, AE is the CP recombination zone in the energy band gap. To reduce measurement delay, I-V was measured for few VG I . , , values (near VT) and CP was measured at fixed pulse height. Figure 2. (LHS) PMOS energy band diagram in inversion and Definition of NBTI regime under high VG showing impact ionization. (RHS) Dependence of hot-hole generation on ewerimental parameters. Fig. 1 shows AVTand SILC (-ANOT) for various VGstress. At low VG for all time and at high VGfor early time ANoT is negligible (no SILC). AVT is governed by ANrr and shows t" (n-114) dependence. At high VGand long time AVT increases, coinciding with the appearance of SILC. Due to its strong VG acceleration ANm affects long-time AVT for high VG stress, as predicted by calculated ANoT contributions . As mentioned in TDDB literatures [16,1Y], ANoT is due to the generation and injection of hot holes (HH) into the oxide bulk. Fig.2 shows the energy band diagram of a PMOS under Figure 3. Correlation of AVr and ADIT for Stress at different inversion and high VG. Electrons tunneling from the gate can Eox. Low Stress VG (for entire Eox range) ensures negligible create impact ionization (11) and generate HH. The magnitude impact ionization and hot-holegeneration. 5.1.1 0-7803-8684- 1/04/$20.00 02004 IEEE IEDM 04-105 Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on December 3, 2008 at 00:38 from IEEE Xplore. Restrictions apply. Role of electric field and inversion holes D: diffusivity of H species) . Time exponent n depends Fig.4 (LHS) correlates AVT (-ANIT) to stress Eox and VG on the type of H species, its trapping and release in the oxide for different TPHY. ANm shows good correlation with Eox but bulk [ 181. ANm rate changes after H diffusion front reaches not VG. As a separate proof, Fig.4 (RHS) plots normalized the Si02-poly interface. Either H absorption into poly ensures drain current shift (AI,) for stress at p-MOS inversion (PI), n- faster H removal and higher ANrr rate 1161, or, H reflection MOS accumulation (NA) and n-MOS inversion (NI). AIDwas from poly would result in ANm saturation [IO]. Finally, ANm measured at a fixed gate overdrive and reflects ANn induced should eventually saturate when all Si-H bonds are broken. mobility degradation. PI and NA show similar AI, under The measured time dependence of ANn together with that identical Eox (VGIA=VGII due to Vm difference) and not -lV calculated using the R-D model for successive stress and under identical Vc. Furthermore under identical Eox, PI recovery phases are also shown. The predictive nature of the (holes near the interface) shows higher AIo than NI (holes R-D model can be readily observed. tunnel from gate). This clearly shows the importance of Signature of R-D model: Stress experiments inversion layer boles behind NBTI degradation . Fig.6 shows ADm for stress at various Eox and T (low VG, high T). ADrr shows t" (n-1/5) dependence at short time that increases at longer time. Break time (tBREAK) decreases, post- break slope (SBREA~) increases at higher T. tBREAKand SBREAK are insensitive to Eox. This long-time ADmincrease is unlike that observed at high Vc, and is due to the absorption of H species into poly once the H diffusion front hits the Si02-poly interface (at t=tBREAK) The time exponent n for t < tBnEAK . suggests the H species as neutral @ or H2 [IS]. Moreover, Eox and T dependent data fort < tBREAK be can scaled along Y and X axis directions respectively to universal relations . Since H species is neutral (D does not contain Figure 4. (LHS) Dependence of AVT (in N,r contributim rqime) any field dependence), Eox scaling will affect the reaction on stress v0 and for different T ~ (RHS) . Normalized AI; ~ term SNonly. By assuming similar T activation for kF,,and kR measured at fixed overdrive for different stress configuratims. (SNdoes not contain any T dependence), T scaling will only Mechanism of Nm generation: R-D model affect the diffusion term D. Therefore, X and Y axis scaling of pre-break T and Eox dependent data would yield activation F i g 3 illustrates the Reaction-Diffusion (R-D) model for energy (EA)of difusion and field acceleration of reaction. EA Nm generation . Hole-assisted reaction breaks interfacial Si-H bonds into Si- (Nm) and H species. Nm buildup equals for ADm is obtained by scaling T dependent (Eox constant) total number of released H species. Dissociation of Si-H and data along Y axis direction. According to the R-D model, if diffusion of H species away from the interface determine Nr EA(km) - E A (kR), then EA(ADm)= EA(D) * n. buildup rate for very short and longer times respectively. For Fig.7 (LHS) plots the T activation of ADm, l/tBREAK and typical measurements, only the later phase is observed where D. EA(D) suggests the H species as neutral H2'. Similar EA for D and lltBREAK confirms that the T activated enhanced ANm(t)=[kFo.P.N&R]? (Dt)" = SN(Dt)" (km, kR:forward and reverse reaction rate; P hole density; No: initial Si-H density; ADm is a diffusion related phenomenon. Fig. 7 (RHS) plots 9, Si-H + 2h+ = 2Si++ H2 0.165 - 0.25 H+ 0 measured -simulated Distance into oxide 1 Nn = Ibn.P.N&Sn.lM = SN IDt)" 0 0 4WO 2 W 4030 6M: Stresslrmvew time 1s) I " I ~~~I Figure 5. (I) Schematic of reaction-diffusion mcdei for NIT generation (151. (il) Various reaction pathways can release different H species (181. (Ill)Time evolution of H diffusion profiles in the oxide. H concentration at the Si-Si02 interface reduces in time. (IV) Various phases of NI, buildup. 1: reaction-limited, 2: quasi-equilibrium,3: diffusion-limited,4: enhancement due to poly absorption, 5 : saturation due to poly reflection, 6: final saturation. (V) Phase-3 time dependence depends on H species (181. (VI) Measured and calculated ANn during stress and recovery. 5.1.2 106-IEDM 04 Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on December 3, 2008 at 00:38 from IEEE Xplore. Restrictions apply. as tgREAK a function of TPHY. tgREAKreduces with TpHydue to Moreover, good AVT versus ADrr correlation seen during faster arrival of Hz front into Si02-poly interface for thinner recovery suggests entire VT recovery is due to recovery of Drr TPHY.Furthermore, Fig.7 also shows that EAfor NBTI (AD,) only. This was also reported in , and expected since ANoT is related to EA (D) by the time exponent n (-l/5), according and hole trapping is negligible for low stress VG values . to that predicted by the R-D model. AC stress Signature of R-D model: Recovery experiments Fig19 shows the pulse frequency dependence of ANrr. Fig.8 shows fractional ADn recovery (after NBTI stress) Degradation under AC is always lower than that under DC and the correlation of AVT versus AD, during recovery for NBTI [9,14,17,18]. AN, is frequency independent for lower f successive stress and relaxation phases. Stress Eox was kept and the reduction (w.r.1 DC) is due to recovery effects. At constant and recovery parameters (stress-recovery sequence, higher f, ANn reduces further due to non-equilibrium reaction VG and T during recovery) were varied. Stress and recovery effects [ 17,181. Significant lifetime improvement is achieved were performed at identical T. Such post-stress ADn recovery for circuits under switching conditions. is anticipated by the R-D model due to H diffusion towards 1, , 0.6 Si-Si02 interface and re-passivation of Si-H bonds [17,18]. io-' ioo i o ' ioz io3 10' io-' i o o i o ' i o 2 i o 3 io' . , , , o<AVi and ADlT during recove,, obtained for different recovery stress time fl .~ .. . s - \-, ~ parameters (sequence, Vo and T) but identical stress Eox. Figure 6. Time evolution of ADlr for various Eox and T stress. Pre-break Eox and T dependent data is scaled by factors 10' shown in brackets along Y- and X-axis resoectivelv (shown by arrows) to an arbitrary reference level. 4 10' ,U lo" IO' Id Id I' O Id IO6 loo 20 22 24 26 /",. T..... IA'l I Figure 7. (LHS) Temperature activation of I/tsnsm. D (of t i Figure 9. Effect of frequency on NITbuildup. Upper Fig data Species in SiO,) and AD,, (NBTI activation). D is obtained from [le],symbol: experiment, line: R-D model. from X-axis scaling of T dependent data. (RHS) tsREan a as function of Tpw. Role of Nitrogen Fractional recovery is similar for successive relaxation Fig.10 shows AVT for different amount of N2 content in phases, lower at higher T but higher when a positive VGis the gate oxide. Data obtained from [12,13] for films grown by applied in the recovery phase. Of importance is the reduced RTO and followed by RTN, having two different TPHY. AVT recovery at higher T, and is triggered by the loss of H species always increases with increasing N2 content (under identical into poly as explained in . Once the H-diffusion front hits stress condition). Fig. 10 also shows the calculated reaction . the SiOz-poly interface, H species are "absorbed" into the energy (ER)of Si-H bond dissociation for different interface poly due to its faster H diffusivity. This leads to higher AD, structures [131. According to 1131, larger N2 content reduces ERand therefore, NITgeneration and NBTI are favored. ._ at hieher T as seen before 181. However., it also reduces the ~ ~~ ~~~~~~~~~~~ number of H species available for recovery once fie stress is Fig.11 (LHS) shows AVT and ANT correlation for thermal stopped, Faster diffusion in poly at higher results in and nitrided oxides having different Nz content. Data from higher fractional H loss and lower fractional ADr recovery. [ l l l shows AVT increases for a given ANn as N2 content is 5.1.3 IEDM 04-107 Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on December 3, 2008 at 00:38 from IEEE Xplore. Restrictions apply. increased. Increase in AVT seen with increasing N2 content is ANrr during stress and recovery is reaction limited for short due to increased hole trapping in the oxide hulk, as per [ I I]. time and diffusion limited for long time. Fig.11 (RHS) shows trapped hole density versus time for ANm shows a power-law with stress time whose exponent various stress VG. Data from [ I l l and shows hole trapping is determined by the diffusing H species. Time exponent and increases at higher stress VC. However for a given stress VG, activation energy of diffusion is used to identify the diffusion hole trapping is independent of stress time. According to species as neutral H2. A unique scaling scheme is presented [I I], time-independence of hole trapping implies negligible to determine Eox and T dependence of ANm. T activation of NOT, generation. Therefore, increased hole trapping at higher NBTI and that of diffusion are shown to be uniquely related Vc IS due to increased tilling of pre-existing (prior to stress) by the time exponent of diffusion. traps, perhaps generated due to the nitridation process. . Once diffusing H2 front hits the Si02-poly interface, ANm 1o’t A ii I increases due to absorption of H2 in “absorbing-poly”. It is , . gl0+/f I e I /i _I1I SiOl + Si-H: 0.84eV Si303N+ Si-H: 7.81eV Si302N2+Si-H:7.65eV I anticipated that reflection of H2 from “reflecting-poly” would reduce ANm rate and can explain “saturation” seen for many Nz containing samples. However, this needs to be verified. ANm recovery fraction is shown to be slightly higher for positive Vc, lower at higher T and independent of stress-relax Si30N3 + Si-H: 7.46eV sequence. Lower fractional recovery at higher T is due to the loss of H2 due to absarption into poly. For “reflecting-poly”, -Si3N1+ Si-H: 7.34eV the recovery fraction is,anticipated to be larger. This needs to be verified as well. Under AC stress, NBTI is lower due to loo 0 4 8 12 16 PO 24 recovery effects. Further reduction at high f is observed due Nitrogen concentration (at. %) to non-equilibrium reaction effects. Recovery plays a big role Figure 10. (LHS) Effect o NI incorporation in the gate oxide cm f in exact determination of NBTI degradation and projected NBTI. Data f r a n  and . (RHS) Calculated NBTi reaction lifetime and calls for careful attention. energy (ER)for variws interface structures. Data from . Increased NBTI is seen for nitrided gate oxide MOSFETs. Possible reasons may he increased ANm due to reduction in reaction energy, or hole trapping in processing related traps. The exact mechanism is still debated and needs verification. Acknowledgement J. Vasi and R. L 1 (IIT Bombay) for useful discussions. S. a J. Hillenius (Agere Systems) for encouragement and support. References 111 K. Uwusawa, T. Yamamota and T. Mogami, in proc., Inr. Electron Devlce MeeL.p.871, 1995. I21 D. S c M c r and I. F. Babcock, I. Appl. Phys.. v.94, p.l.2003 stress time fx 1000s)  A. Krishnan et. al., in proc., Int. Electron Device Meet., p.349.2003 Figure 11. (LHS) Correlation of VT shift and normalized NI, 141 V. Reddy el. al., in proc., lnl. Reliability Phys. Symp., p.248.2002 buildup for pure SiOn and nitrided oxide samples. (RHS)  C. Schlunder et. al., in proc.. Int. Reliability Phys. Symp.. p.5.2003 Trapped hole ccncentration versus stress time measured in  N. Kimiruka et. al., in proc.. VLSl Tech. Symp., p.73, 1999 nitrided oxide for different stress VG.Data from [I I].  S . Mahapatra and M. Alam. in proc., 101. Electron Device Meet., p.505, 2002. Conclusions  S. Mahapatra, P. Bharalh Kumar and M. Alam, in proc., Int. Electron T o summarize, this paper discusses various mechanisms Device Meet., p.337, 2003. that influence NBTI for thick and thin oxide MOSFETs, for a [Y] S. Rangan. N . Mielke and E. C. C. Yeh, in proc., Int. Electron Device Meet., p.341.2003 wide range of stress hias and temperature. It is shown that [IO] S.Tsujkawa et. al., in pm.. Ini. Reliability Phys. Symp., p.?8,2004 NOTgeneration takes place for thick TpHydevices at high Vc [ I l l V. Huad and M. Denais. in proc., Int. Reliability Phys. Symp., p.40, stress. ANOThas a much stronger VG acceleration than N m 2004 generation, and influence the long-time overall VT shift. Care [I21 T. Sasaki et. al., in proc., Solid Stale Dev. Mal., p.66.2003 [I31 S. S. Tan el. al., in proc., Solid Stale De”. Mat.. p.70.2003. must be taken to properly choose stress VGduring accelerated [I41 G. Chen el. al., in proc., Int. Reliability Phys. Symp., p.lY6.2003 tests for reliable estimation of true NBTI lifetime. [I51 K. 0. Jeppson andC. M. Stensson, J.Appl. Phys., v.48.p.2004. 1977 For proper choice of stress VC, AN* determines overall [I61 M. d a m , B. Weir and P. Silverman, in proc.. Inl. Workshop on Gate Insulator, p.l0,2001. AVT especially for non-nitrided thermal oxide. Unlike ANw L171 M. Alam, in proc.. Int. Electron Device Meet., p.345.2003 that is governed by Vc and hot-holes, ANm during NBTI is 1181 S. Chakravarthi et. al.. in proc., Int. Reliability Phys. Symp., p.273, governed by Eox and inversion layer (cold) holes. The hias, T 2004 ~~~ and time dependence of Nm buildup and recovery for stress [I91 M. Alam, J. Bude and A. Ghetti, in proc.. Int. Reliability Phys. Symp.. p.2 I , 2000. and relaxation phases can be described’using the R-D model. 1201 M. L. ReedandJ. D. P l u m e r , J. Appl. Phys., v.63.p.5776. 1988 5.1.4 108-IEDM 04 Authorized licensed use limited to: INDIAN INSTITUTE OF TECHNOLOGY BOMBAY. Downloaded on December 3, 2008 at 00:38 from IEEE Xplore. Restrictions apply.