Statistical Test Compaction Using Binary Decision Trees

Because of the significant cost of explicitly testing an integrated, heterogeneous device for all its specifications, there is a need for a test methodology that minimizes test cost while maintaining product quality and limiting yield loss. The authors are developing a decision-tree-based statistical-learning methodology to compact the complete specification-based test set of an integrated device by eliminating redundant tests. A test is deemed redundant if its output can be reliably predicted using other tests that are not eliminated. To ensure the required accuracy for commercial devices, the authors employ a number of modeling and data-massaging techniques to reduce prediction error. Test compaction results produced for a commercial MEMS accelerometer are promising in that they indicate it is possible to eliminate an expensive mechanical test.