Heuristics Based Test Overhead Reduction Techniques in VLSI Circuits

Abstract

The electronic industry has evolved at a mindboggling pace over the last five decades. Moore’s Law [1] has enabled the chip makers to push the limits of the physics to shrink the feature sizes on Silicon (Si) wafers over the years. A constant push for power-performance-area (PPA) optimization has driven the higher transistor density trends. The defect density in advanced process nodes has posed a challenge in achieving sustainable yield. Maintaining a low Defect-per-Million (DPM) target for a product to be viable with stringent Time-to-Market (TTM) has become one of the most important aspects of the chip manufacturing process. Design-for-Test (DFT) plays an instrumental role in enabling low DPM. DFT however impacts the PPA of a chip. This research describes an approach of minimizing the scan test overhead in a chip based on circuit topology heuristics. These heuristics are applied on a full-scan design to convert a subset of the scan flip-flops (SFF) into D flip-flops (DFF). The K Longest Path per Gate (KLPG) [2] automatic test pattern generation (ATPG) algorithm is used to generate tests for robust paths in the circuit. Observability driven multi cycle path generation [3][4] and test are used in this work to minimize coverage loss caused by the SFF conversion process. The presence of memory arrays in a design exacerbates the coverage loss due to the shadow cast by the array on its neighboring logic. A specialized behavioral modeling for the memory array is required to enable test coverage of the shadow logic. This work develops a memory model integrated into the ATPG engine for this purpose. Multiple clock domains pose challenges in the path generation process. The inter-domain clocking relationship and corresponding logic sensitization are modeled in our work to generate synchronous inter-domain paths over multiple clock cycles. Results are demonstrated on ISCAS89 and ITC99 benchmark circuits. Power saving benefit is quantified using an open-source standard-cell library.