Precise atom manipulation through deep reinforcement learning

Abstract

Atomic-scale manipulation in scanning tunneling microscopy 1 has enabled the creation of quantum states of matter based on artificial structures 2-4 and extreme miniaturization of computational circuitry based on individual atoms. [5][6][7] The ability to autonomously arrange atomic structures with precision will enable the scaling up of nanoscale fabrication 8 and expand the range of artificial structures hosting exotic quantum states. However, the a priori unknown manipulation parameters, the possibility of spontaneous tip apex changes, and the difficulty of modeling tip-atom interactions make it challenging to select manipulation parameters that can achieve atomic precision throughout extended operations. Here we use deep reinforcement learning (DRL) 9 to control the real-world atom manipulation process. Several state-of-the-art reinforcement learning techniques are used jointly to boost data efficiency. The reinforcement learning agent learns to manipulate Ag adatoms on Ag(111) surfaces with optimal precision and is integrated with path planning algorithms 10,11 to complete an autonomous

Emerging machine learning strategies for diminishing measurement uncertainty in SPM nanometrology

“…Maximizes the reward signal for corrective actions or punishes for the wrong ones Manages the tip conditioning process [135], manages an atomic-scale manipulation autonomously [167,168], optimizes PI feedback parameter [133] framework, which learned the inputs and then outputted the rating values (from 0 to 10) corresponding to the quality of the FD curves. Note that the rating value here means the fitting rate used to tilt the raw FD curve in order to remove the spikes appearing in the indentation zone of the FD curve.…”

“…Specifically, the ML model was required to find the suitable trajectories of the STM tip in which the individual molecules (3,4,9,10perylene tetracarboxylic dianhydride) can be bound to the tip and then removed from the Ag (111) surface. Likewise, Chen et al [168] proposed a promising reinforcement learning method to perform an atomic manipulation task that the STM tip that was capable of transferring a single atom (i.e., Ag and CO) to the targeted position (i.e., Ag (111)). The trained reinforcement learning model was applied to construct an artificial kagome lattice with 42 atoms.…”