Quality Prediction in Multi-Stage Manufacturing with Hybrid TCN-Transformer

Quality prediction in multi-stage manufacturing processes, such as semiconductor production, is critical for ensuring operational stability and minimizing costs. These processes generate high-dimensional multivariate time-series data from sensors, encapsulating complex temporal dynamics and inter-stage dependencies. Existing approaches suffer from computational inefficiency and limited capability to capture long-range temporal patterns in high-dimensional sensor data. To overcome these challenges, this paper presents the Multi-Stage Wafer Quality Predictor (MSWQP), a novel framework that efficiently integrates multi-stage temporal information while reducing computational overhead. Evaluated on a real-world semiconductor dataset across two stages (Sputtering and Rapid Thermal Processing), MSWQP outperforms baselines—including Bi-LSTM, standard TCN, and Transformer models—achieving an MSE of 0.0868, MAE of 0.1919, and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$R^{2}$</tex> of 0.5243. These results highlight MSWQP's superior predictive accuracy and efficiency, offering a scalable solution for quality control in complex industrial systems.