Neuromorphic Photonic Processors for Ultra-Low-Latency Edge Inference

Neuromorphic photonic processors are promising direction with regard to realising ultra-low-latency inference at the edge, through the parallelism of neuromorphic architectures with the speed of photonic interconnects. A performance evaluation of these processors based on latency, throughput and energy efficiency is presented in this paper. Compared to traditional electronic accelerators, comparative results demonstrate the benefits of neuromorphic systems based on photonics to support nextgeneration edge AI.

Hali tarjima qilinmagan