Accelerating Electron Diffraction Analysis with Machine Learning Inference on FPGAs.

• Pranav Srinivas Murali (University of Washington (US))

Room: ECE-M306 Abstract: This thesis presents a novel approach to enhance the analysis of crystal structures using Reflection High-Energy Electron Diffraction (RHEED) in material science. Addressing the current bottleneck of lengthy analysis times in existing RHEED systems, this study introduces a high-speed camera-based system design capable of real-time crystal analysis. The proposed system leverages a LeNet5-based neural network deployed on an FPGA-based framegrabber to accelerate image processing. Through meticulous optimization efforts including neural network quantization and Vivado optimization directives, the system achieves latencies of 450-750 us, facilitating RHEED image analysis at speeds of up to 1000 Hz. This advancement promises researchers real-time insights into crystal growth processes, empowering timely interventions for desired experimental outcomes.