Real-time synchronization-error prediction using position measurements for reference-side feedforward compensation under main-secondary control

Abstract In wafer scanners, the synchronization error between the wafer stage (WS) and the reticle stage (RS) is measured in real time by position sensors and directly determines lithographic imaging quality. A main--secondary control (MSC) framework is commonly adopted for WS--RS synchronization. Within the MSC, simultaneously meeting the requirements of real-time execution, cross-trajectory generalization, disturbance robustness, and nonlinear error-mapping capability remains challenging. Trajectory compensation is an effective means of improving accuracy, but existing methods generally require the reference trajectory to be known in advance, making them difficult to apply to MSC, where the synchronization reference is generated online. This paper proposes SyncTRC-FF, a synchronization reference-side feedforward compensation method tailored to such an online reference. Its feedforward predictor, termed SyncTRC-Net, is a lightweight temporal residual convolutional model specifically designed for synchronization-error prediction. It takes the past and present WS position measurements and reference trajectory derivatives as input, predicts the synchronization error online, and injects the prediction into the synchronization reference. Because external disturbances affect the measured WS position, the compensation also suppresses the disturbance-induced synchronization error. The method leaves the existing feedback loops unchanged, and its prediction module is proved bounded-input bounded-output (BIBO) stable. Experiments on a synchronization platform with two motion stages emulating the WS--RS configuration show that, on repetitive references, SyncTRC-FF approaches the accuracy of fully converged iterative learning control (ILC) without requiring any trajectory-specific iteration, while it clearly outperforms ILC in trajectory generalization and disturbance rejection and meets real-time deployment requirements.