Speaker
Description
Edge computing for future high-energy physics experiments will require the utilization of cutting-edge technology in both hardware and software. Data handling and processing, including particle identification and characterization, is a crucial element of modern experiments in which precise, repeatable calculations must be performed within nanosecond timescales with high data rates. The NEUROPix project aims to redefine this process through the incorporation of spiking neural networks in FPGA-based firmware units and define a path toward a neuromorphic computing ASIC, which would process raw data from Timepix4-like sensors. Like all machine learning techniques, the first step is the creation of a large, detailed, and realistic dataset for training for which we employ Allpix Squared. This talk will cover: strategies used for the creation of this dataset, including the expression of Allpix Squared output into a raw-data-like format and methods for creating large training datasets from a smaller set of well-defined simulation outputs; an overview of constraints inherited from the spiking neural network application; and plans for the NEUROPix project. We also seek community input to refine the simulation’s output and discuss other machine learning applications.