Speaker
Description
Neural networks (NNs) are commonly implemented on CPU or GPU, which often struggle to meet the strict latency and throughput demands of real-time data processing.
FPGAs offer an alternative: deterministic, low-latency inference directly in the data path.
We present the opportunities FPGA-based NNs offer, how to implement NNs on FPGAs, and how to identify possible use cases.
As our case study, we implemented an NN on the FPGA of a COTS digitizer hardware for real-time determination of particle arrival time in particle counters such as scintillator or semiconductor detectors used at GSI.
When multiple particles arrive at almost the same time, their signals superimpose and the detector response can become non-linear. The presented NN still allows accurate and reliable arrival time detection under these conditions, enabling particle counters to operate reliably at much higher rates.
| I have read and accept the Privacy Policy Statement | Yes |
|---|