Integrated modeling and simulation incorporating machine learning methods: case studies and perspectives on usefulness

Plasma processes are the core of new device process technology. Realization of new device technology includes challenges. Device structures are 3D with strict topography and material property tunability requirements. Process hardware options have increased degrees of freedom. They also introduce mechanistic complexity that burdens engineering development. Pathfinding for process and hardware development is the most critical phase. Empirical methods are insufficient to grapple with “pathfinding” phase challenges. At this phase data is at best sparse; pathfinding needs to exploit physical models. System level models are problematic due to size and complexity and the sparsity of parameter space they cover. At the pathfinding phase, they may be inaccurate. Integrative methods extend concurrent engineering. In this approach, a combination of first principles simulation and experiments targets important unknowns of process and hardware development rather than seek a system level (“simulation-of-everything”) description. Clarity to the least understood aspects of critical processes is the desired outcome. A vehicle for understanding what species flux-surface reactions result in structures, and films then what processes achieve them. Intermediate level theory models like KMC and microkinetic approaches weave insights about surface process through to models that can describe process parameters. Even with a modeling infrastructure, diagnostics can never exit the integrative framework. Unknown parameters may be more easily/quickly measured than computed. Computational cost and sparsity of parameter space coverage remain issues. Machine learning methods can help as interpretive, or optimization tools or means to obtain a system level model. This presentation will survey the basic elements of integrative models including how they support new multicomponent dielectric material deposition. We then explain the fundamental aspects of machine learning methods. Case studies describe the utility of different machine learning approaches. The aim of the talk is to show how value addition results from plasma unit process development with first principles approaches augmented by machine learning methods at the earliest concept and feasibility phases.