中文摘要：

针对已有的测试向量选择方法采用串行程序实现，难以应对测试程序时间及测试数据量迅速增加的问题，提出一种基于GPU的测试向量选择方法，用于高效地从大测试向量集（n倍检测的测试向量集或随机的测试向量集）中选择出较高测试质量的测试向量．在考虑受限的测试时间／成本的条件下，采用GPU编程将测试向量选择过程并行化，以最大化1～n倍检测覆盖率为目的，将测试向量按照故障检测能力从大到小排序，从而在实际芯片测试时能够尽快淘汰故障芯片，减少测试时间．实验结果表明，与国际上考虑”倍检测的测试选择工作相比，该方法获得了21．9倍加速；与商业工具产生的同样大小的测试集相比，该方法得到的测试集具有更好的1～n倍检测覆盖率（平均提升3．2％～8．3％），同时也能获得更加陡峭的故障覆盖率曲线．

英文摘要：

Previous test selection methods are based on serial algorithms which cannot meet the continuously increasing CPU runtime and test data volume. This paper proposes a Graphic Processing Unit （GPU） based Max 1-to-n detection test selection method to produce a high quality test set from any possible pattern sources, either generated randomly or deterministically. We adapt GPU to parallelize the test selection which considers the limitation of the test cost and select the tests that could maximize the 1-to-n detection fault coverage. The tests are ordered by their contribution to n- detection fault coverage so that they can detect more defected chips in the early time of test application and reduce the test cost. Experiments demonstrate the proposed method offers a 21. 9X speedup compared with the previous n-detecion-aware test selection method. Moreover, in comparison with the test set generated by commercial tools, under the same test set size, the selected test set produced by the proposed method could achieve a better n-detection fault coverage improvement on average） and a steeper fault coverage curve.