Iterative Learning Controller with Learning Gain Optimization and Online Data Adjustment for Atomic Force Microscope

Atomic force microscope (AFM) is a high precision measurement instrument under nanoscale resolution, and is commonly used in various fields, including semiconductor analysis, MEMS inspection, living samples inspection, etc. Apart from optical microscopes, the operating principle of AFM requires accurate motion control of the measuring stage to maintain a certain distance between the tip and sample surface. However, rapid topography height changes make it become a challenging task. Iterative learning controllers (ILC) are often implemented to improve tracking performance when the reference signal is periodic and the system model could be identified accurately. It is suitable to apply ILC in the z-direction actuator of AFM in cases where topography wouldn't vary much from scanning line to scanning line. Nevertheless, samples with varying topography along scanning lines would degrade the tracking performance while applying conventional ILC. Here, we propose a regression model of edge prediction to shift the feedforward signal, making ILC more adaptive to varying topography samples. Furthermore, we integrate an optimization process to tune the learning gain online for compensating the inaccuracy of system modeling. Therefore, error would converge gradually even without precise system parameter and with line varying topography. Experiment results verify that our proposed method outperforms PI and conventional ILC in terms of controller error and topography accuracy remarkably.