Low-Power and Error-Resilient VLSI Circuits and Systems

Efficient low-power operation is critically important for the success of the next-generation signal processing applications. Device and supply voltage have been continuously scaled to meet a more constrained power envelope, but scaling has created resiliency challenges, including increasing timing faults and soft errors. Our research aims at designing low-power and robust circuits and systems for signal processing by drawing circuit, architecture, and algorithm approaches.
To gain an insight into the system faults due to supply voltage reduction, we researched the two primary effects that determine the minimum supply voltage (VMIN) in Intel's tri-gate CMOS technology, namely process variations and gate-dielectric soft breakdown. We determined that voltage scaling increases the timing window that sequential circuits are vulnerable. Thus, we proposed a new hold-time violation metric to define hold-time VMIN, which has been adopted as a new design standard.
Device scaling increases soft errors which affect circuit reliability. Through extensive soft error characterization using two 65nm CMOS test chips, we studied the soft error mechanisms and its dependence on supply voltage and clock frequency. This study laid the foundation of the first 65nm DSP chip design for a NASA spaceflight project. To mitigate such random errors, we proposed a new confidence-driven architecture that effectively enhances the error resiliency of deeply scaled CMOS and post-CMOS circuits.
Designing low-power resilient systems can effectively leverage application-specific algorithmic approaches. To explore design opportunities in the algorithmic domain, we demonstrate an application-specific detection and decoding processor for multiple-input multiple-output (MIMO) wireless communication. To enhance the receive error rate for a robust wireless communication, we designed a joint detection and decoding technique by enclosing detection and decoding in an iterative loop to enhance both interference cancellation and error reduction. A proof-of-concept chip design was fabricated for the next-generation 4×4 256QAM MIMO systems. Through algorithm-architecture optimizations and low-power circuit techniques, our design achieves significant improvements in throughput, energy efficiency and error rate, paving the way for future developments in this area.