IN DEEP-SUBMICRON IC PROCESSES and products, traditional defect-based test methods using stuck-at test vectors and I DDQ patterns cannot sufficiently maintain quality and reliability goals, because of the increase of delay-sensitive (resistive) defects. Structural test methods based on path delay or rise and fall transitions can counteract this trend and avoid expensive functional testing.Frequency-dependent defects pose a significant challenge to IC suppliers. At-speed functional testing, which requires the generation of test patterns and the ATE to support them, is cost prohibitive. Moreover, suppliers must apply functional tests at the fully specified temperature and voltage ranges.For ASIC testing, structural test methods such as ATPG scan or BIST are generally accepted replacements for full at-speed testing, as long as several conditions are met: I The methods' stuck-at and delay fault coverage are at or above functional-test levels. I The fabrication process is controlled well within the modeled parametric specifications. I The on-chip IP (cores, memory, and I/O) is well characterized across worst-case PVT (process, voltage, and temperature).However, once manufacturers discard at-speed functional testing in favor of structural testing, testing's sole purpose becomes to screen defects, and therefore, the approach's success depends on high defect coverage. To fully realize high defect coverage and thus guarantee IC quality and reliability, designers must augment high fault coverage with additional defect-based screening methods.
Defect screeningBoth defect-based test methods-such as transition delay fault (TDF) testing, minimum operating voltage (min V DD ), and very low voltage (VLV)-and temperature testing are important methods for screening resistive path and delay defects. 1-3 The major challenge for designers wanting to use these test methods is identifying and screening the defective dies in production without losing too many defect-free dies or consuming excessive amounts of test time in the search routines. Figures 1 and 2 give examples of resistive path defects as outliers in raw data collected from the ATE. Figure 1 plots the min V DD for two silicon wafer lots against process speed, which we measured with a process monitor. Although correlated to the part's operating frequency, min V DD does not measure the operating frequency of the device under test. Most dies in each lot show the expected intrinsic behavior of a higher min V DD as the process slows; however, outliers to the intrinsic distribution show a higher min V DD than expected for the measured process speed. Clearly, applying a single volt-