Compression-free Checksum-based Fault-Detection Schemes for Pipelined Processors

Inhalt:
We propose a fault-detection scheme for pipelined, multithreaded processors. The scheme is based on checksums and improves on previous schemes in terms of fault coverage and detection latency by not using compression but storing complete checksums from several pipeline stages. We validate the scheme experimentally and derive checksum polynomials that lead to perfect fault coverage.