Decision Selection and Learning for an 'All-Solutions ATPG Engine

'All-solutions ATPG' based methods have found applications in model checking sequential circuits, and they can also improve the defect coverage of a test-suite, by generating distinct multiple-detect patterns. Conventional decision selection heuristics and learning techniques for an ATPG engine were originally developed to 'quickly' find any available (single) solution. Such decision selection heuristics may not be the best for an 'all-solutions ATPG' engine, where all the solutions need to be found. In this paper, we explore new techniques to guide an 'all-solutions ATPG engine'. We first present a new decision selection heuristic that makes use of the 'connectivity of gates' in the circuit in order to obtain a compact solution-set. Next, we analyze the 'symmetry in search-states' that was exploited in 'success-driven learning' and extend it to prune conflict sub-spaces as well. Finally, we propose a new metric that determines the use of learnt information a priori. This information is stored and used efficiently during 'success driven learning'. Experimental results show that we can compute the complete solution-set with our new heuristics for large ISCAS'89 and ITC'99 circuits, where conventional guidance heuristics fail.