ICMCTF 2025 Session CM4-1-MoM: Simulations, Machine Learning and Data Science for Materials Design and Discovery I

The recent adoption by the general public of artificial intelligence (AI) tools such as ChatGPT has reinvigorated research into AI applied to material science.Deep learning, a subset of AI, allows computer systems to autonomously learn patterns in data and construct efficient decision rules for tasks including classification, regression, or segmentation.In material analysis, these tools have primarily been applied to techniques requiring analysis of data collected in the form of images. Electron backscatter diffraction (EBSD) is one such technique benefitting from these recent efforts to improve material analysis by leveraging deep neural networks.Within the last decade, advancements in EBSD equipment have enabled the capture of high-definition diffraction patterns at rates exceeding 3,000 Hz.This creates significant opportunities for increasing the amount of information that can be ascertained from a sample, as well as opens the door for training data intensive deep neural networks.

Deep neural network-based classification of the diffraction patterns is motivated by Hough-based EBSD’s susceptibility to structural misclassification; a failure mode that modern EBSD can encounter even when the researcher has complete knowledge of the sample prior to beginning analysis.While several methods to improve phase-differentiation have been proposed, each still requires pre-selection of phases and additional data (e.g. chemistry or simulated diffraction patterns) to be available.In contrast, deep neural network-based methods have demonstrated effective phase differentiation and identification of phases to the space group level without the need for further information.The deep learning approach to EBSD diffraction pattern analysis is capable of these more advanced analyses because it uses all information in the image when assessing a diffraction pattern, whereas traditional Hough-based EBSD pattern analysis discards a significant amount of information.

To promote adoption of AI tools, it must be determined if and when it is prone to error.To test the ideal operating conditions, the deep neural network model is trained using diffraction patterns captured with a fixed geometry and SEM settings, and a parametric study is performed to develop an understanding of model performance as several of the most common EBSD operating conditions are varied. Each time one parameter is varied, the diffraction patterns are re-collected, and the CNN used to reassess the space group identification.Ultimately, the model is found to retain a high classification accuracy even with significant changes to the diffraction conditions.

This paper presents an analytical investigation of phase separation dynamics in thin films using the one-dimensional Cahn-Hilliard equation. The main focus is on the development of perturbation solutions for Greens functions under various boundary conditions, specifically periodic, Neumann, and Dirichlet. The derived solutions provide insight into the behavior of concentration profiles, essential for understanding the phase separation process in technologically relevant thin film materials.

Thin films are widely used in various technological applications, including microelectronics and energy storage. The phase separation within these films is critical to their performance and properties. The one-dimensional Cahn-Hilliard equation serves as a fundamental model for studying the evolution of concentration profiles during phase separation. This paper aims to explore the dynamics of phase separation under different boundary conditions by deriving similarity transform and perturbation solutions to the Cahn-Hilliard equation.

Starting with the 1D Cahn Hilliard equations, Perturbation expansions are applied to the Green function solution and then evaluated for various boundary conditions. Alloys evaluated were Cu-Ni and Pd-Si.

Results showed differences for the diffferent Boundasry consditions and material parameters involved. The small perturbation parameter is chosen to be k the interface parameter. in the CH equation.

The results are hoped to ber useful in designing surface thin films, using inputs like thermal diffusivity M and the interface parameter appearing in the CH equation .

The ability to write structures at the nanoscale using lithography underpins all modern society. The electronic devices we take for granted contain integrated circuits, the key component being the field-effect transistors (FETs). They have reduced in size by a factor of two every two years for over fifty years, following “Moore’s Law”. This size reduction is dependent on the continuous development of materials and techniques that produce better line resolution. The technical program aimed to apply a simulation tool called Excalibur to materials and processing problems critical to Applied Materials Inc. The Excalibur simulation suite can model the behavior of electrons and ions in the range of 100 keV to 3.6 eV. This software allows rapid prototyping of the next-generation resists for electron beam lithography (EBL) and ion beam lithography (IBL) for the semiconductor industry. Although Excalibur currently does not simulate extreme ultraviolet (EUV) radiation, we propose that it can provide a first-order analysis and prediction of EUV based on e-beam behavior. This project provides evidence that such prediction can be well modeled using the Excalibur tool. We also provide an alternative simulation, which we call Merlin, that aims to be more accurate and faster than its predecessor.