Evolutionary optimization of Markov sources for pseudo random scan BIST

Recent work by Basturkmen et al. (2002) showed that Markov sources lead to scan BIST designs of lower cost compared to earlier proposed methods in scan BIST. However the method presented by Basturkmen et al. utilizes tests generated using a deterministic test generator for target faults in synthesizing the Markov source to generate the tests. The requirement of a deterministic test generator may hinder the use of this procedure in industrial settings since the BIST tool must also include a deterministic ATPG tool that may add to the cost of the BIST tool. In this paper we investigate a procedure to synthesize BIST controllers with Markov sources for test generation using Evolutionary Algorithms (EAs). This allows us to avoid using the deterministic ATPG needed previously. Additionally we do not employ inversion logic used in by Basturkmen et al., thereby potentially reducing the hardware in the BIST controller. Nevertheless, the proposed method achieves close to 100% fault efficiency using far fewer tests than required by pseudo random tests. In this work, similar to that of Basturkmen et al., we employ Markov sources based on finite state machines (FSMs) with transitions controlled by additional inputs. Hence, transition probabilities (TPs) of the FSM are determined by the 1-probability on a corresponding input.