We study the performance of thread-parallel SAT-based ATPG tool TIGUAN in a multi-processor, multi-core compute server environment. We explore the trade-off between the additional processing power available in the system and communication overhead. Determining the optimal amount of resources is complicated by the heterogeneity of ATPG instances: test generation is done quickly for the majority of faults while a smaller number of hard faults requires intensive computations. We present first experimental results for standard stuck-at tests and high-quality gate-exhaustive tests for large industrial circuits.
展开▼
