AI-guided frame prediction techniques to model single crystal diamond growth

At a Glance

Metadata	Details
Publication Date	2025-05-01
Journal	Journal of Vacuum Science & Technology A Vacuum Surfaces and Films
Authors	Rohan Reddy Mekala, Arjun Srinivasan, Matthias Muehle, Elias Garratt, Adam Porter
Institutions	Fraunhofer USA Center Mid-Atlantic CMA, Michigan State University

Abstract

Single-crystal diamond (SCD) is an attractive wide bandgap semiconductor material for a variety of applications. While lab-scale prototype demonstrations have demonstrated superior performance, utilizing diamond has been prohibitively expensive compared to more mature, commercialized semiconductor materials. The only way to overcome this is by controlling the size and quality of SCD wafers during chemical vapor deposition. We are addressing this challenge by using an artificial intelligence (AI)-based algorithm to predict SCD growth states through the use of in situ red, green, blue (RGB) images. This AI growth prediction can be incorporated as the control system to increase the SCD wafer size and quality. First, we installed a full-frame mirrorless interchangeable lens camera equipped with a macrolens to a diamond synthesis reactor. Then, a cumulative AI-based pipeline comprising three distinct and interconnected components was developed to model and guide diamond growth: (1) A feature extraction pipeline, using deep learning-based semantic segmentation models to isolate and quantify geometric growth features such as diamond contours and pocket holder regions from in situ RGB images. (2) A defect detection pipeline, employing object detection and segmentation architectures to identify and localize macroscopic defects, including polycrystalline growth and edge anomalies, during the growth process. (3) A frame prediction pipeline, which combines image sequences and reactor telemetry using a generative deep learning architecture to forecast future reactor states, including expected growth morphology and potential defect emergence. Prediction accuracies of greater than ∼99% were obtained with minimal information loss between predicted and actual outputs. This constitutes a never-before obtained result of spatiotemporal (AI) prediction of diamond shape from in situ growth data. This demonstrates the potential of these algorithms as a machine-intelligence-enabled solution for automated optimization and control of the diamond growth process.

Tech Support

Original Source

DOI: https://doi.org/10.1116/6.0004290

References

1965 - Cyclic Plastic Strain Energy and Fatigue of Metals
2004 - High accuracy optical flow estimation based on a theory for warping