Using Conflict-Based On-line Learning to Accelerate the Backtrace Algorithm Implemented in HW

The backtrace algorithm lies at the core of many applications, including automatic test pattern generation (ATPG). This paper describes and evaluates an approach for accelerating the backtrace algorithm using conflict-based online learning. This approach gets high performance due to large amounts of fine-grain parallelism in the implication process. Architecture of circuits performing backward determination of all input vectors to a given output one is presented as well. The results of this new method are compared with HW implementation of the basic backtrace algorithm proposed recently. The experimental results have been obtained for the ISCAS´85 benchmarks - for those largest, the average growth of an area overhead is about 49 % and the average speed up is about 25 %