Increasing testability by clock transformation (getting rid of those darn states)

Abstract

We provide a new answer to the fundamental question "Why is sequential test generation so difficult?" the presence of darn (Difficult And Really Needed) states. A darn state is both difficult to reach and required to detect some faults. We introduce a method for identifying darn stares, along with a technique to measure their detrimental effect on the fault coverage. Darn states are the result of detrimental correlation between flip-flops (FFs) sharing the same clock. We propose a novel DFT technique in which FFs are partitioned into different groups having independent clocks during resting. The goal of partitioning is to increase the fault coverage by transforming darn states into easy-to-reach states. The proposed DFT has small area overhead and no performance penalty.