X Marks the Spot: Scan-Flip-Flop Clustering for Faster-than-at-Speed Test

Faster-than-at-Speed Test (FAST) can be used to detect Hidden Small Delay Faults (HDFs), which are escaping a conventional at-speed test. However, due to the overclocking of the Circuit Under Test (CUT), a high number of X-values are introduced into the test responses, since the corresponding outputs have not yet stabilized at the target observation time. Furthermore, the X-ratio varies significantly with the observation time, consequently posing a great challenge to compactor designs. In this paper, a novel Scan-Flip-Flop (SFF) clustering method is presented to generate scan-chains which support an efficient response compaction for FAST. Clustering is done both by analyzing the circuit's topology and by timing-accurate simulation of a test set. An efficient graph-based algorithm clusters the SFFs, such that the X-values are accumulated in only few scan-chains.Experimental results verify that blocking the chains with highest X-density can significantly decrease the X-ratios of the remaining chains while maintaining a high fault efficiency. Compared to a random configuration the observed improvement can be up to 4 times higher.Since no actual ordering is imposed on the SFFs inside a scan-chain, the proposed algorithm still leaves a high degree of freedom for further scan-chain optimization with respect to different cost metrics.