Use It or Lose It: Proactive, Deterministic Longevity in Future Chip Multiprocessors
MetadataShow full item record
Ever since the VLSI process technology crossed the sub-micron threshold, there is an increased interest in design of fault-tolerant systems to mitigate the wearout of transistors. Hot Carrier Injection (HCI) and Negative Bias Temperature Instability (NBTI) are two prominent usage based transistor degradation mechanisms in the deep sub-micron process technologies. This wearout of transistors can lead to timing violations along the critical paths which will eventually lead to permanent failures of the chip. While there have been many studies which concentrate on decreasing the wearout in a single core, the failure of an individual core need not be catastrophic in the context of Chip Multi-Processors (CMPs). However, a failure in the interconnect in these CMPs can lead to the failure of entire chip as it could lead to protocol-level deadlocks, or even partition away vital components such as the memory controller or other critical I/O. Analysis of HCI and NBTI stresses caused by real workloads on interconnect microachitecture shows that wearout in the CMP on-chip interconnect is correlated with lack of load observed in the network-on-chip routers. It is proven that exercising the wearout-sensitive components of routers under low load with random inputs can decelerate the NBTI wearout. In this work, we propose a novel deterministic approach for the generation of appropriate exercise mode data to maximize the life-time improvement, ensuring design parameter targets are met. The results from this new proposed design yields ~2300x decrease in the rate of CMP wear due to NBTI compared to that of ~28x decrease shown by previous work.
Boga, Siva Bhanu (2015). Use It or Lose It: Proactive, Deterministic Longevity in Future Chip Multiprocessors. Master's thesis, Texas A & M University. Available electronically from