High-performance chip reliability from short-time-tests - Statistical models for optical interconnect and HCI/TDDB/NBTI deep-submicron transistor failures

In high-performance chips, both Bragg gratings (used for signal separation in multi-signal optical interconnect alternatives to copper interconnect architectures) and deep-submicron transistors fail when the stress-induced activation of the performance-enhancing hydrogen in the amorphous oxide generates enough defects to significantly degrade performance. By making an analogy to the more mature theory of Bragg gratings, disorder-induced variations in the activation (generation) energies of the defects, are shown to be a sufficient explanation for the sub-linear time dependence of HCI (hot carrier induced degradation), TDDB (time dependent dielectric [soft/hard] breakdown) and NBTI (negative bias temperature instability) deep-submicron transistor degradation modes. We then show that for all these degradation modes, Weibull (not Lognormal as is sometimes assumed) intrinsic failure-time distributions result from the variations in defect activation energies and that the short-time device degradation can be used to extract tails of these semi-symmetric Weibull failure-time distributions. This also explains why Arrhenius defect generation rates yield non-Arrhenius MTF in small devices. Combining the resulting failure statistics with novel qualification methodology, "latent failures" can be avoided through design changes implemented for reliability.