edfas.org ELECTRONIC DEVICE FAILURE ANALYSIS | VOLUME 28 NO. 1 16 ENHANCING SOFT DEFECT LOCALIZATION WITH SOFTWARE AUTOMATED INTELLIGENT LASER SCANNING (SAILS) AND ARRAY-BASED IMAGE RECONSTRUCTION Arun Karunanithi1, Benny Hsu1, Kent Erington1, Joseph Caroselli1, Winston Gao1, Calder Wilson1, and Aaron Liao2 1Advanced Micro Devices Inc., Austin, Texas 2University of Texas at Austin, Austin, Texas arun.karunanithi@amd.com EDFAAO (2026) 1:16-21 1537-0755/$19.00 ©ASM International® INTRODUCTION Soft defect localization (SDL) is a fault isolation (FI) technique used to root cause device marginalities and/ or defects. The variety of test modes and their marginalities that can be solved with SDL is increasing as new ways are found to utilize this technique. However, SDL analysis can be time consuming if the test times are slow, because millions of test executions are needed to get a statistically significant result. To solve this problem, the authors propose an improvement named software automated intelligent laser scanning (SAILS) to modulate laser dwell time on the fly. This software and hardware implementation can be applied to any test method, now or in the future, and to any of the SDL tools available in the market. The approach utilizes an array-based image reconstruction methodology alongside an automated SDL workflow for improved site detection accuracy of both weak and strong sites. This paper describes the successful implementation of this approach and shows its ability to judiciously mask out sites in real-time and use the mask to modulate laser dwell time. Soft defect localization involves setting the device under test (DUT) at a marginal pass/fail state while looping on the test indefinitely.[1] A near infrared laser beam is used to perturb sensitive circuitry and alter the DUT’s pass/fail behavior. The pass/fail result is mapped on to the raster area and is overlaid on a laser scanning microscopy (LSM) image as shown in Fig. 1. Since the technique involves looping the marginal test indefinitely, loop time for the test becomes critical for SDL data collection. TEST LOOP TIME CHALLENGE FOR SDL SDL application for typical root cause analysis for SCAN test mode is relatively simple in terms of test loop length because only a subset of pattern vectors can be run. This ensures the test loop length is on the order of microseconds. Industry studies have reported that a reasonable test loop should be about 1 millisecond; 100 milliseconds is almost impossible.[2] However, successful SDL analysis with loop times over 100 milliseconds by creating asymmetric loops has been reported.[3] When asymmetric loops are not possible, a different solution using long loop times is needed. Solutions exist in the industry to solve this including DXGlue.[4] In this paper, the authors propose one such solution based on real time laser dwell time modulation using variable synchronization pulses. This enhancement will be referred to as SAILS. Fig. 1 A typical raw SDL image on the left showing regions of white and black pixels where laser caused the test to fail and pass, respectively. Overlay SDL on LSM is shown on the right.
RkJQdWJsaXNoZXIy MTYyMzk3NQ==