Understanding the impact of transistor-level BTI variability

Recent work has shown large variations due to bias-temperature instability (BTI) at the device level, and we study its impact on the behavior of larger circuits. We propose an analytical method that is over 600x faster than Monte Carlo simulation and accurate for technologies down to 16nm, and demonstrate it on circuits with up to 68,000 transistors. Results show that the impact of BTI variability at the circuit level is significantly smaller than at the device level, but increases with device downscaling.