Hardware Implementation of Neural Network Algorithm for Ultra Low Latency Signal Classification

The CookieBox is an angular streaking detector developed for LCLS-II which extracts intrinsic characteristics from an X-ray shot. Those characteristics can then be used to classify and determine if the X-ray shot respects the pre-established experimental framework. However, to be of any value to the rest of the experiment, this classification must be done quickly, ideally less than 100 µs after the shot. Preliminary work demonstrated that our encoding techniques enable the use of very small neural network which shows great potential for fast and accurate classification. The proposed research project is to implement those already existing and functioning algorithms on dedicated hardware like FPGA.

Faculty Supervisor:

Audrey Corbeil Therrien

Student:

Partner:

Stanford University

Discipline:

Engineering

Sector:

Education

University:

Université de Sherbrooke

Program: