Performance-Aware Test Scheduling for Diagnosing Coexistent Channel Faults in Topology-Agnostic Networks-on-Chip

Abstract

High--performance multiprocessor SoCs used in practice require a complex network-on-chip (NoC) as communication architecture, and the channels therein often suffer from various manufacturing defects. Such physical defects cause a multitude of system-level failures and subsequent degradation of reliability, yield, and performance of the computing platform. Most of the existing test approaches consider mesh-based NoC channels only and do not perform well for other regular topologies such as octagons or spidergons, with regard to test time and overhead issues. This article proposes a topology-agnostic test mechanism that is capable of diagnosing on-line, coexistent channel-short, and stuck-at faults in these special NoCs as well as in traditional mesh architectures. We introduce a new test model called Damaru to decompose the network and present an efficient scheduling scheme to reduce test time without compromising resource utilization during testing. Additionally, the proposed scheduling scheme scales well with network size, channel width, and topological diversity. Simulation results show that the method achieves nearly 92% fault coverage and improves area overhead by almost 60% and test time by 98% compared to earlier approaches. As a sequel, packet latency and energy consumption are also improved by 67.05% and 54.69%, respectively, and they are further improved with increasing network size.

Self-Healing Router Approach for High-Performance Network-on-Chip

“…The traditional method for a fault-tolerant router uses spare or redundant routers, which increases the area overhead significantly. Sometimes, spare or redundant channels are allowed [11]. Another traditional method uses a fault-tolerant routing algorithm [12], [13].…”