A New Kind of Atomic Force Microscopy Scan Control Enabled by Artificial Intelligence:Concept for Achieving Tip and Sample Safety Through Asymmetric Control  

作　　者：Johannes Degenhardt Mohammed Wassim Bounaim Nan Deng Rainer Tutsch Gaoliang Dai 

机构地区：[1]Physikalisch-Technische Bundesanstalt,Bundesallee 100,38116 Brunswick,Germany [2]Institut Für Prozessmess-und Sensortechnik Technische Universität Ilmenau,EhrenbergstraBe 29,98693 Ilmenau,Germany [3]Institute of Production Metrology TU Braunschweig,SchleinitzstraBe 20,38106 Brunswick,Germany

出　　处：《Nanomanufacturing and Metrology》2024年第2期50-59,共10页纳米制造与计量（英文）

基　　金：funding from the EMPIR programme co-financed by the participating states and from the European Union’s Horizon 2020 research and innovation programme(20IND08‘MetExSPM’).

摘　　要：This paper introduces a paradigm shift in atomic force microscope(AFM)scan control,leveraging an artificial intelligence(AI)-based controller.In contrast to conventional control methods,which either show a limited performance,such as proportional integral differential(PID)control,or which purely focus on mathematical optimality as classical optimal control approaches,our proposed AI approach redefines the objective of control for achieving practical optimality.This presented AI controller minimizes the root-mean-square control deviations in routine scans by a factor of about 4 compared to PID control in the presented setup and also showcases a distinctive asymmetric response in complex situations,prioritizing the safety of the AFM tip and sample instead of the lowest possible control deviations.The development and testing of the AI control concept are performed on simulated AFM scans,demonstrating its huge potential.

关 键 词：Atomic force microscopy Artificial intelligence Deep reinforcement learning Optimal control 

分 类 号：TH742[机械工程—光学工程] TP18[机械工程—仪器科学与技术]