The Memory Bottleneck in Photonic Neural Network Accelerators
Russell L. T. Schwartz, Belal Jahannia, Nicola Peserico, Hamed Dalir, Volker J. Sorger · 2024
Photonic Tensor Cores are a competitive accelerator for Neural Networks, offering high throughput, but requiring large bandwidths to operate at their maximum efficiency. Here we offer an analysis of the memory bottleneck for PTC.