Calendar of Events
Events Calendar
Smart Pixels: On-Chip Data Reduction with Machine Learning
Speaker: Jennet Dickinson, Cornell University
Abstract: High-granularity pixel detectors at the LHC and beyond must operate in extreme collision environments and satisfy stringent constraints on power consumption and read-out bandwidth. Current-generation pixel detectors produce too much data to read out for every collision, and the future experiments will face even bigger challenges due to smaller pixel pitch and more complex beam backgrounds. Edge machine learning (ML) algorithms implemented in the pixel front-end readout electronics can reduce the amount of data that must be read out, and sufficient reduction would enable full-collision analysis of every event. Smart pixels is a coordinated effort to co-design ASICs and ML algorithms suited to this task. I will present the status of ongoing work in smart pixels, including an overview of the algorithms designed and the testing of the first 28nm ASIC implementation.
Host: Yuri Gerstein