Modeling Custom Digital Circuits for Test

Abstract

Models meant for logic verification and simulation are often used for Automatic Test Pattern Generation (ATPG). For custom digital circuits, these models contain many tristate devices that tend to lower coverage for stuck-faults. Additionally, these tristate devices contribute to increased ATPG runtimes, fewer generated test sequences, and an overall lower test quality. The circuit under test is partitioned into channel connected sub-networks (CCSN) that consist of transistors that are connected at their source or drain terminals, except when these terminals are power, ground or primary inputs. Unlike other published work, algorithms presented in this paper analyze each CCSN in the context of its environment, thereby capturing the logical relationships among its input signals. Other algorithms presented include identification and modeling of embedded latches, clock generators and memory circuits. An abstract array model for memory that reduces the size of the model and increases simulation speed is also presented. When one specific feature of the algorithm was disabled, experimental results showed higher ATPG runtimes of about 35%, and an average decrease in fault coverage of around 15–20%. For the largest data cache, the memory modeling algorithm decreased the number of primitives from 1.23 million to 139 thousand.