An On-line Variable Speed Scanning Method with Machine Learning Based Feedforward Control for Atomic Force Microscopy

Atomic force microscopy (AFM) is a powerful instrument that has the ability to characterize sample topography on nanoscale resolution and is widely used in different fields, such as nanotechnology, semiconductor, MEMS, bioscience, etc. However, the main drawback of AFM is the low scanning speed, which restricts the applications in some cases. Conventionally, the scan trajectory is raster pattern under uniform speed which is hard for fast scanning and is inefficient. In this paper, we propose a fast scanning algorithm combining on-line variable speed scan and machine learning based feedforward control. The feedforward signal is based on a Gaussian process model which can be trained from previous image and is able to predict sample topography. Thus, the feedback controller only needs to correct difference between the predictions and real scan data, which can increase scanning speed. The on-line variable speed algorithm would adjust scanning speed based on the instant error and the predictions of the Gaussian process model. Therefore, the scanning process would be sped up at regions with small height change and slow down before where height changes dramatically, such as edges and sidewalls. The experiment results demonstrate that our proposed method can increase image performance and save scanning time remarkably.