Testing of hard faults in simultaneous multi-threaded processors

With the increasing circuit complexity and aggressive technology scaling, faults such as dielectric, conductor, and metallization failures are becoming more common. Traditional stuck-at testing does not detect these types of faults because these faults may be dormant and need to be stressed to manifest as "fails" (during burn-in). As voltage scaling and power consumption reduce the effectiveness of burn-in, these faults will cause failures during the useful life of the part. We propose the use of a test thread on simultaneous multi-threaded processors, to provide a means of detection of lifetime failures. In this initial study, the test thread is allowed to compete with executing programs for processor resources. When the system has only one workload thread, the test thread is able to execute with no significant impact on the workload thread. When the system has many active workload threads, the test thread has somewhat larger impact on the execution time of the workload threads.