Defect Detection Using Quiescent Signal Analysis

Abstract

I_DDQ or steady state current testing has been extensively used in the industry as a mainstream defect detection and reliability screen. The background leakage current has increased significantly with the advent of ultra deep submicron technologies. This increased background leakage noise makes it difficult to differentiate defect-free devices from those with defects that draw significantly small amount of currents. Therefore it is impossible to use single threshold I_DDQ testing for today’s technologies. Several techniques that improve the resolution of I_DDQ testing have been proposed to replace the single threshold detection scheme. However, even these techniques are suffering from loss of resolution that is required for detection of subtle defects in the presence of leakage currents in excess of a few mA. All these techniques use a single I_DDQ measurement for detection and thus the scalability of these techniques is limited. Quiescent Signal Analysis (QSA) is a novel I_DDQ defect detection and diagnosis technique that uses I_DDQ measurements at multiple chip supply pads. Implicit in our methodology is a leakage calibration technique that scales the total leakage current over multiple simultaneous measurements. This helps in decreasing the background leakage component in individual measurements and thus increases the resolution of this technique to subtle defects. Defect detection is accomplished by applying linear regression analysis to the multiple supply port measurements and using outlier analysis to identify defective devices. The effectiveness of this technique is demonstrated in this paper using simulation experiments on portion of a production power grid. Predicted chip size and leakage values from the International Technology Roadmap for semiconductors (ITRS) are used in these experiments. One of the other major concerns expressed in ITRS is that of significant increase in intra-die process variations. The performance of the proposed technique in presence of such variations is evaluated using three different intra-die process variation distribution models.