MSU electrical engineering research into AI technology to predict and refine diamond growth toward commercial use of diamond material
Large silicon crystals are the basis for semiconductor computer chips and switching devices for electric grid applications. The efficiencies of today’s electronic devices are dependent on a crystal’s perfection and its ability to better control electrons without loss.
Diamond semiconductor crystals outperform silicon by orders of magnitude but are essentially unavailable for commercial use in the current marketplace.
What if artificial intelligence (AI) could improve process efficiencies and grow large-area defect-free crystals for manufacturing?
That’s what Dr. Elias Garratt of Michigan State University will explore with the help of a $500,000 grant from the National Science Foundation under the Future Manufacturing program.
The Assistant Professor of Electrical and Computer Engineering and Materials Science is collaborating with Fraunhofer USA Center for Coatings and Diamond Technologies (CCD) and Fraunhofer USA Center for Experimental Software Engineering (CESE) to develop AI technology to predict and refine diamond growth for manufacturing.
Dr. Garratt stated the research team hopes to leverage the vast amounts of data generated during the crystal-growth process instead of analyzing data once experiments are completed.
“Development and integration of deep learning artificial intelligence architectures in the Chemical Vapor Deposition process will make growth predictions more accurate and add defect assessment to the prediction for manufacturing of diamond materials,” Dr. Garratt explained. “Outcome of the project will accelerate the development cycles and reduce costs for manufacturing processes – making them adaptable to a broad range of crystal-growth processes for electronics,” he added.
The team at MSU and Fraunhofer USA, will also develop a course in AI-based manufacturing aimed at preparing vocational workers to run crystal deposition/growth equipment in an AI-driven manufacturing setting. Concepts developed in the project will also be integrated into existing courses, capstone projects will be designed for students, along with the education modules for training operators.
The abstract of the grant can be found at: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2036737&HistoricalAwards=false
AI prediction of diamond shape (macro-feature) ~6 hours into the future. From the prediction model inputs (5 input images + numerical reactor variables of temperature, pressure, etc.) the AI generated two output images. Not only was the final shape of the diamond reasonably predicted, but polycrystalline material was also predicted to grow on the edge of the pocket holder the diamond rests in. This secondary result is highly significant as polycrystalline diamond material strongly impacts the final shape, size and microscopic defect content of diamond due to how its growth changes the environment.
Representation of AI-guided material process development cycle. Reactive approaches rely on the design of new experiments (design of experiments) based on results obtained at the conclusion of the process run – the process is run to completion before assessment is performed serially. The guided approach envisioned inserts AI models between the start and end of the process to leverage the large amounts of data generated during synthesis. AI models trained on data distributions from in-situ, reactor, and ex-situ data can achieve accurate predictions of future material states, detect the formation of defects in those states, and be used as a closed-loop control. This generates a new cycle within a single process. This approach is expected to make process development more efficient by reducing the number of experiments needed to achieve a result, e.g. fewer defects in a material.