From fabrication to functionality: a hybrid physics-based and data-driven approach

Plasma processing plays an integral role in semiconductor manufacturing. One emerging discipline is neuromorphic engineering, which strives to enable the brain’s energy efficiency for microchips through biologically inspired information processing. Technological implementations necessitate a holistic understanding and the disentanglement of fundamental mechanisms. Previous work has focused primarily on investigating the physical processes on the device level, often neglecting the dependence on the fabrication processes. In this work, machine learning facilitates a link between the involved disciplines – nanoelectronics and plasma processing. Data mining is used to correlate the characteristics of experimentally manufactured SiO_x:Cu resistive switching devices with their sputter deposition in Ar/O₂ and Ar plasmas. The plasma properties are determined using 2d3v particle-in-cell simulations, featuring an integrated data-driven module for surface chemical kinetics and sputtering. It is argued that the devices’ functionalities are primarily related to the interplay between the composition of particle fluxes and the deposited energy at the surface.