Maximizing nontarget defect detection using conventional stuck-at fault-based automated test pattern generation tools

Testing plays a vital role in determining the success or failure of today's competitive integrated circuit manufacturing business. The cost of test generation, the quality of the applied tests, and the cost of applying the test determine if a product can be profitable in today's marketplace. With product cycle times decreasing and package density increasing, the need for efficient, high quality testing continues to grow. The applied post-manufacturing tests generally consist of a number of different types of tests including conductivity, functional, IDDQ, and static tests. This dissertation will focus upon the static testing of integrated circuits. Accepted commercial practice is to use the stuck-at fault coverage as the metric to measure a test set's quality. As a result, the test pattern length is determined by how many tests are required to attain the desired stuck-at fault coverage. Although a good approximation, test generation based solely upon the stuck-at fault coverage metric is misdirecting the test generation resource. The quality of a test set should not only be based upon the stuck-at fault coverage, but also on its effectiveness at detecting other defect types seen in the real manufacturing process. Although there has been research based upon using more exotic fault models that mimic these real manufacturing defects, test generation using these fault models is too costly in time and memory requirements.; Valuable information is lost when using conventional ATPG tools. Our research focuses upon enhancing existing test generation tools by targeting those sites which make the greatest contribution to overall defective part level during the test generation process. For the first time, test pattern generation techniques that maximize non-target defect detection have been successfully used to test a real, 100% scanned, commercial chip consisting of 75,000 logic gates. The results of the experiment proved that the new test generation method yielded defective part levels significantly lower than those achieved by tests generated using today's state of the art tools and techniques.