Adaptive weight generation

Weighted random testing is gaining popularity as an economical method for external as well as for built-in-self-testing (BIST). In weighted testing we have to satisfy two competing requirements: to keep both the test length and the hardware cost low. Usually multiple sets of weights are required. To reduce their number and, as a consequence, the hardware cost, we quantize the weights to values which can be easily generated in hardware. The concept of weight quantization, its theory and its application to reducing the number of weight sets to be implemented in hardware, are studied. When weighted test patterns are generated it is likely that the fault coverage of the corresponding test sets thus obtained overlap. To compensate for this we apply the weight sets selectively in reverse order, thereby reducing the test length and the hardware cost. We develop a new method of weight generation, adaptive weight generation, which applies our methods of selective reverse simulation and weight quantization; in this way, both the test lengths and the number of weight sets are reduced. Simulation experiments indicate that significant improvements, both in terms of time and hardware cost, are afforded by our technique