Innovative Advanced Materials, Inc. STTR Phase II Award, September 2020

Innovative Advanced Materials (IAM), Inc., founded by Dr. W. Alan Doolittle, has developed machine learning methods and algorithms capable of predicting material properties for MBE grown materials such as Ternary Alloy Composition, Electron Mobility, Surface Roughness, and other material parameters during growth, in real-time.  These techniques provide the fundamental basis for managing the MBE growth process and pursuing methods that could lead to Materials Discovery technologies.  Materials Discovery exists at the intersection of the fields of Materials Science and Computational Design, including materials synthesis and characterization as it relates to the discovery and design of new materials.  IAM proposes to demonstrate machine learning methods for MBE growth that will lead to in-situ characterization of materials in real-time.  IAM will use its Ubiquitous Control™ software, built based on MATLAB technologies, for eventual control of the MBE process utilizing in-situ sensors, including RHEED image analysis and Auger chemical analysis.  The proposed Deep Learning Neural Network architecture allows for multiple data streams using a parallel Convolution Neural Network and Artificial Neural Network model that provides continuous material property prediction in real-time.  While the machine learning algorithm does not “know the underlying physics” the RHEED 2D diffraction data is already known to contain physical information about lattice spacing and thus, composition, surface roughness, faceting (connected to surface energy), relative degrees of crystallinity (i.e., amorphous, polycrystalline, crystalline), ratios of chemical species (e.g., 2x4 and 4x2 reconstructions), and relative temperature ranges via observed RHEED patterns and growth mode (Frank–van der Merwe or FM, Volmer–Weber or VW, or Stranski–Krastanov or SK modes).  The focus of this program is to demonstrate control schemes capable of managing the MBE growth via advanced machine learning technologies and multi-sensor in-situ materials characterization.