Bias temperature instability mechanisms
CHRISTOPHER L. HENDERSON, Semitracks, Inc., and DAVID W. ROSE, Keithley Instruments, Inc.
NBTI is considered to be one of the most difficult reliability challenges facing the semiconductor manufacturing community today.
Bias temperature instability (BTI), particularly negative BTI (NBTI) in p-channel transistors, affects small feature size devices and is quite difficult to reduce or eliminate. As feature sizes become smaller, the effects of NBTI become more pronounced, resulting in increases in threshold voltage and decreases in drain current and transconductance in p-channel devices. NBTI results from a positive charge buildup in p-channel transistors. It occurs at low negative gate-to-source voltages and does not result in an increase in gate leakage current. Rather, it affects off-state drain-to-source leakage and reduces the drive current. Generally, this problem is worse than standard hot carrier degradation because it results in permanent interface traps being generated, reducing device lifetime.
Although the NBTI phenomenon has been well known for years, it was a controllable problem in earlier CMOS devices. The problem resurfaced around 1990 with the introduction of dual-doped polysilicon gates, but hot carrier effects were more dominant during that timeframe. Hot carrier effects were brought under control by drain and channel engineering, and voltage reduction. Now, aggressive scaling has made NBTI more problematic than hot carrier effects.
NBTI occurs mostly in p-channel MOSFETs when either (a) a negative bias is applied to the gate, or (b) the DUT is used at an elevated temperature. The effect is much more significant when both a negative bias AND an elevated temperature occur. It results from holes migrating to the silicon-oxide interface and is manifested by the current-voltage characteristics becoming unstable. This is accompanied by reduced drain current in the linear saturation regime, slope degradation that indicates an increase in interface trapped charge, negative threshold voltage shifts that indicate positive trapped charge, and an increase in off-state leakage current in certain instances (Fig. 1).
Figure 1. Typical results before and after a bias temperature stress on a 0.25-micron channel length transistor. Stress applied at 250??C for 50 hours at an electric field of 4.3 megavolts/cm.
Holes apparently catalyze the reaction, so holes are needed at the silicon-oxide interface. Researchers believe that hydrogen in some form is involved in the degradation kinetics. This phenomenon is not as prevalent in n-channel MOSFETs because they are electron majority carriers, but they can have the less common PBTI problem. One possible reason why PBTI is less of a problem is that nitrogen is used in p-channel transistor gate oxides to block boron penetration.
NBTI typically increases with decreasing channel length. This is due to the increased contribution from the source and drain overlap regions. It can be quite pronounced in deep submicron technologies. At 90nm channel lengths, the change in threshold voltage can exceed 100mV and the change in drive current can approach 10% using the stress conditions in Fig. 1. In addition, the off-state leakage for the transistor after stressing is typically between one and two orders of magnitude higher than before stressing.
NBTI effects increase as oxide thickness decreases. For example, threshold voltage (VTH) can increase by almost an order of magnitude as oxide thickness is decreased from 40 angstroms to 20 angstroms.
Important NBTI processes
Three processes associated with NBTI and its measurement should be understood:
??? Relaxation processes
??? Fast vs. slow measurement methodologies
??? Relationship of VTH shifts with interface traps
Through a charge pumping measurement technique, Vincent Huard  of ST Microelectronics has shown that relaxation of the interface traps is only a small part of the relaxation of the VTH shift. The thermal activation energy of VTH is much greater than the thermal activation of the interface traps. If the interface traps accounted for the majority of the threshold voltage shift, one would expect the two to have similar thermal processes, and therefore similar activation energies, which is not the case. This indicates that there is something else occurring involving a positive charge. As discussed below, this is closely related to measurement methodologies and conditions.
The measurement techniques for NBTI are somewhat different than for other hot carrier mechanisms. A typical measurement scenario uses a parameter analyzer, with the test taking about 10 seconds. One can then extract DVTH using SPICE models and extrapolating down from the maximum slope to the x-axis. Another technique is to hold the gate voltage constant for the majority of the measurement time, change it for a few milliseconds and add a drain bias during this time, as shown in the two graphs on the lower portion of Fig. 2. In this scenario, the charge in the gate does not have time to relax. One can measure the drain current and then extrapolate back to get the threshold voltage. If you compare the conventional DC technique with an "on the fly" measurement, the change in threshold voltage is higher by a factor of ten when using an "on the fly" method. More importantly, there is also a change in the slope. This relates to whether the mechanism involves interface traps or fixed charge.
Figure 2. Large differences in DVTH shifts are observed between "fast" and "slow" measurements. V. Huard, IRPS 2004.
In contrast to a power law dependency, sometimes a linear-log time dependence in DVTH is observed. This eliminates the possibility of a drift component, which researchers initially thought was occurring. Mickael Denais  discovered that the threshold voltage shift is correlated with the amount of hole current injected. This can be measured through carrier separation techniques. The results of his study point to a positive charge rather than interface traps. This charge could be protons, positively charged silicon, or even H3O+. Denais also verified relaxation of the VTH shift, the magnitude of which is greater at higher positive voltage but eventually reaching a saturation level, indicating this is most likely a positive charge process.
Implications on measurement methods
Dhanoop Varghese  and his co-workers have studied the effects of measurement conditions on NBTI and its VTH degradation. It was found that for a fixed gate voltage, the slope of the DVTH vs. stress time curve varies with the drain voltage used in the measurement, stress temperature, and stress-measure delay time. The slope is greater for higher temperatures, longer delay times, and lower drain (measurement) voltages. A similar relationship is also apparent when examining the interface traps, but the effect is much less pronounced.
Varghese also found there is an Arrhenius dependency in this behavior by measuring the time delay effect at two different temperatures. He modeled this phenomenon using a Reaction-Diffusion simulation with H2 hydrogen, showing there is good correlation between the data and the simulation. This indicates that the underlying mechanism is not hole trapping or charge trapping but rather some form of diffusion. The slopes of the lines that fit the data indicate molecular hydrogen or H2 as the species.
Figure 3. Stress-measurement delay time and test temperature clearly affect results.
These results underscore the need to make ultra-fast measurements. This can be seen clearly in Fig. 3, which shows the threshold voltage shift over many orders of magnitude of time. Notice that there is a change in slope below 2 to 4 seconds. The increase in slope suggests a dispersive-type process or some type of trapping and detrapping. When the time is longer, there is more of a diffusion phase, i.e., NBTI is a reaction-diffusion process. Holes react at the interface, releasing something, most likely hydrogen, which then diffuses. The process is diffusion limited rather than reaction limited. The Power Law takes over for the diffusion process. The process follows Arrhenius behavior at longer times, but the temperature effect is nonlinear.
Varghese and his group reached the conclusion that under most conditions these phenomena are strictly related to interface traps. Arrhenius plots of data collected for devices having different oxide thicknesses, different charge doses, different electric fields, and different doses of nitrogen, all fall nicely on a line that indicates a thermal activation energy of 0.6 eV. They also get an n value of about 1/6, which is consistent with molecular hydrogen diffusion.
Researchers agree on several aspects of NBTI. NBTI is a hole catalyzed reaction that has a pronounced effect on p-channel transistors, and is quite difficult to reduce or eliminate, particularly in devices with short channel lengths. NBTI field dependence is due to the reaction phase, not a field drift component. The damage increases the threshold voltage shift. Interface traps are certainly involved, but there may be another component such as fixed charge as well.
1. V. Huard et al, International Reliability Physics Symposium Proceedings (42nd Annual IEEE IRPS), 2004, pp 40-45, IEEE International.
2. M. Denais et al, presented at IEDM 2004, in IEEE Transactions on Device Materials Reliability, vol. 4, 2004. pp 715-722, IEEE International.
3. D. Varghese et al, "Material dependence of hydrogen diffusion: Implications for NBTI degradation," in IEDM Technical Digest, 2005, p. 705, IEEE International.
CHRISTOPHER L. HENDERSON, Semitracks, Inc., and DAVID W. ROSE, Keithley Instruments, Inc.
Solid State Technology, Volume 55, Issue 5, June 2012