Unit level predicted yield: a method of identifying high defect density die at wafer sort

Previous studies have demonstrated both theoretically and empirically that defect density at wafer sort is an effective predictor of burn-in failures on packaged semiconductor devices. Therefore, efficient methods of measuring sort defect density are of great interest. This paper explores optimal methods for measuring sort defect density at the die level, as opposed to the lot or wafer level, and using this measure to screen reliability defects. The authors developed a simple measure called unit level predicted yield, based on the yield of other die in the same fab lot, and demonstrated its strong correlation to burn-in failures on packaged units. The method makes physical sense, and has been validated on more than 80 million units on three fab processes, including Intel´s 0.8 μm, 6 layer metal CMOS logic process. It is applicable to known good die programs, or anywhere that die level reliability differentiation is desirable. It is shown to be approximately twice as efficient as wafer level methods at highlighting die with high latent defect density