Multi-Heuristic Machine Intelligence Guidance in Automatic Test Pattern Generation

We present an automatic test pattern generation (ATPG) system using the PODEM (path-oriented decision making) algorithm where backtraces and D-drive are directed by composite heuristics. In a worked out illustration, three heuristic measures are combined by an unsupervised learning procedure of principal component analysis (PCA). The three measures for each signal node are distance (d
PI
and d
PO
or minimum distances to primary inputs and outputs), COP (controllability-observability program probabilities, CC0, CC1, and CO), and SCOAP (Sandia controllabilitiy-observability analysis program combinational testability measures, SC0, SC1, and SO). PCA combines these into heuristic measures, P
0
and P
1
for directing backtraces, and P
D
to advance the D-drive. The new ATPG program for all faults of benchmark circuits shows order of magnitude reduction in total backtracks and notable reduction in CPU time in comparison to the ATPG program with any single heuristic. For individual faults, backtracks often reduced to zero. Even redundant faults were identified by the new program with fewer backtracks.