PL-EWMA controller to mitigate transition effects in high-mixed semiconductor production mode for stability enhancement

Semiconductors are essential for the digital economy and daily life, but their complex manufacturing process, specialized supply chain, and diverse client requirements have led to a high-mixed production mode. To reduce waste, various high-mixed controllers have been proposed in the literature. However, the transition effect, defined by large initial output deviations immediately after product-type changes, is not well addressed. This paper introduces a novel partial likelihood exponentially weighted moving average (PL-EWMA) controller designed to mitigate the transition effects observed in high-mixed production mode during product-type changes. By employing a partial likelihood method, the PL-EWMA controller accurately estimates model parameters and forecasts potential disturbances. This enables precise dynamic adjustments to the control input based on real-time data, ensuring stability and optimal performance for both frequent and infrequent product runs, leading to yield improvement and cost reduction. Extensive performance analysis was conducted across various production scenarios, including diverse parameter settings, parameter misestimation, frequent product changes, model misspecifications, and comparison with the benchmark. The results demonstrate the robust stability and superior performance of the proposed PL-EWMA controller compared to the benchmark methods. In addition, a discussion on practical implementation makes our method more applicable in real-world settings.