A new EDA algorithm combined with Q-learning for semiconductor final testing scheduling problem

Abstract

In recent years, the Semiconductor Final Testing Scheduling Problem (SFTSP), recognized as a unique multi-resource scheduling challenge, attracts increasingly attention of academia and industry in the semiconductor manufacturing process. In this paper, a novel estimation of distribution algorithm combined with Q-learning (QEDA) is proposed to solve the SFTSP. According to the characteristics of the used operation encoding, a new probability matrix update mechanism is proposed for enhancing the priority relationships among operations. Considering that the traditional EDA is not in favor of local exploitation compared with its global exploration, a reinforcement learning is designed to improve the performance of the proposed algorithm. Furthermore, for the challenge of the resource allocation in SFTSP, four actions are introduced based on the variance of individual objectives. Extensive numerical simulations and comparative experiments show that the proposed QEDA algorithm exhibits much better performance than the state-of-the-art algorithms in the literature for solving the SFTSP.