Identifying hard-to-detect delay defects by testing at multiple test-speeds

Physical defects that cause operational failure by adding additional signal-propagation delay are referred to as "delay defects." Delay defects are becoming more prevalent as operating speeds increase and timing margins decrease with IC technology scaling. A device with delay defects may pass production tests when the additional delay is not large enough to cause a test failure. However, the device may fail in field if the defects are sensitized to an untested long path. In order to improve future production yield and debug processes, such devices should be diagnosed to detect these "hard-to-detect" defects that escaped during production testing.;This dissertation presents two new test application techniques to enhance the diagnostic resolution of delay testing: Error Sequence Analysis (ESA) and Failing Frequency Signature Analysis (FFSA). Both techniques apply a test-speed sweep procedure---applying a test multiple times at incremental test-speeds. During the test-speed sweep procedure, ESA analyzes the order in which the scan cells fail, while FFSA analyzes the test-speeds at which each test pattern fails. These techniques can detect and diagnose defects although a given test set may not sensitize the defects to critical paths. In addition, they are effective even with variations of circuit delay, which may degrade the effectiveness of delay testing considerably.;For the test experiment, we used Stanford's ELF18 test chips designed and fabricated by Philips Semiconductor in a 0.18microm process. We tested a total of 551 test chips using a transition fault test set, which shows transition fault coverage of 97% and estimated defect level of 5,238 dpm (defects-per-million). Using the same test set, ESA detected 26 chips that passed conventional transition fault test, 14 chips of which still passed conventional debug processes that include applying thorough test sets, very-low-voltage test, and IDDQ test. FFSA detected 69 chips that passed conventional transition fault test, 24 chips of which still passed the conventional debug processes. These results demonstrate that ESA and FFSA should be an essential part of the debug process to detect and diagnose the defective devices whose failures may be difficult to reproduce using traditional test and debug techniques.