Content area
Abstract
Functional tests are developed during design verification to ensure the correctness of design. They can be reused in manufacturing testing to alleviate the cost of alternative testing methods and increase product quality. For large industrial circuits, there can be thousands of functional tests, each ranging from thousands to millions of cycles. The reuse thus poses many computational challenges. This dissertation addresses these challenges and the associated diagnosis problem to locate the root cause of failure.
Because of the large number of functional tests, it is impractical to apply all of them to test every fabricated device. The tests that provide good manufacturing defect screening value should be selected. Existing metrics for test selection either suffer from high computational cost, or lead to poor selection quality. An efficient register transfer level coverage metric is proposed which has high selectivity and low cost. As a high level metric, it also allows evaluation of testability issues earlier in the design cycle.
The testability of the selected tests can be further improved by additional observation points. Existing techniques require non-fault-dropping fault simulation which is very expensive. An efficient fault dropping technique is proposed based on the number of locations to which a fault gets propagated. Experimental results demonstrate the effectiveness of the method in achieving close to optimal results in the size of the selected subset with an order of magnitude less time.
The use of functional tests helps screen out defective devices. Failure analysis may be further carried out on defective devices to find the cause of the failure. Fault diagnosis helps failure analysis by reducing the number of candidate locations. Existing diagnosis techniques based on a fault model give erroneous results when a real defect is not accurately modeled. A symbolic approach is proposed which models the fault propagation condition at every gate as a Boolean equation. The solutions of the equations represent the set of all single and multiple causes of the observed behavior, thus give exact diagnosis results.
