Dual-Purpose Mixed-Level Test Generation Using Swarm Intelligence
Kelson Gent, Michael S. Hsiao · 2014
Automatic test pattern generation for non-scan sequential circuits is an extremely challenging task. If successful, it can offer many benefits to the EDA community, ranging from manufacturing and functional test to post-silicon validation. High-level test generators often miss the low-level details, thus missing the detection of some gate-level faults. On the other hand, gate-level test generators miss the high-level path traversal knowledge to more effectively traverse the state space. In this work, we present a fine-grain mixed-level test generator that utilizes co-simulation of register-transfer and gate levels to generate high quality vectors. The algorithm, based on an ant colony optimization, targets branch coverage at the RTL and simultaneously attempts to associate rare fault excitations with a sequence of branch activations. By weighting these sequences within the fitness function across the two levels, the algorithm is able to achieve high fault coverage in the presence of deep hard-to-reach states without scan. The result is that the test sequences obtained offer both high branch coverage as well high stuck-at coverage with low computational costs. In particular, for hard-to-test circuits such as the ITC'99 circuit b12, >98% branch coverage and >90% stuck-at coverage are achieved, vastly improving over other state of the art non-scan tools.