Atomic Layer Deposition Optimization via Targeted Adaptive Design.

Abstract

Atomic Layer Deposition (ALD) is a commonly employed process for producing conformal nanoscale coatings in the microelectronics and energy materials industries. ALD processes are composed of cycles of sequential self-limiting chemical reactions followed by purges with an inert gas to produce atomically thin coatings. At the end of each cycle, the Growth Per Cycle (GPC) which corresponds to net mass or thickness change from the previous ALD cycle is determined. Optimizing ALD processes for stable and uniform GPC for a new combination of reactants is challenging as the optimal combination of gas timings, temperature, and gas partial pressures spans a large multidimensional space and in-situ characterization is typically performed with a limited number of mass sensors. In this work, we use Targeted Adaptive Design (TAD), a Gaussian Process (GP)-based probabilistic machine learning framework that aims at efficiently and autonomously locating control parameters that would yield a desired target within specified tolerance, to optimize simulated ALD processes.