A Comparative Study of AI-Driven Regression Testing Techniques of Integrated Circuit Design

VLSI design verification has become extremely complex and resource-intensive due to the system’s requirements to perform ever more complex functionalities than before. To this end, design verification has come under the AI/ML spotlight for faster and more efficient verification of these complex systems. Hence, there is a pressing need for a comprehensive analysis of various AI/ML techniques for this purpose. This paper is focused on experimental evaluation of multiple techniques that utilise AI/ML for test selection: smart regression test flow that utilise $\mathbf{k}$-nearest Neighbors (KNN) algorithm, graph neural network (GNN) assisted smart regression and test selection with reinforcement learning (RL). We performed a detailed analysis on coverage achieved, coverage points achieved, execution time, training time and model re-usability. This analysis aims in providing valuable insights on how artificial intelligence techniques can be applied to IC verification and identify the most effective technique to optimise the regression testing of an IC design. This experimental evaluation is performed under the same conditions for a 28 nm technology node with up to 10 functional blocks. Overall, 19k test cases were considered in order to perform these experiments. The findings of this paper provide practical guidance for modern verification flows, making the decision to select the most appropriate technique for a test scenario and helping future research.