FIONA: Fine-grained Incoherent Optical DNN Accelerator Search Towards Superior Efficiency and Robustness
DescriptionIncoherent optical DNN accelerators (OAs) are booming thanks to unparalleled performance-per-watt and scalability. State-of-the-art work revolutionarily proposed automatic OA search to greatly boost development speed. However, due to the defined coarse-grained search space, the neglect of the inter-tile data transfer network and the impact of variations on OAs, the robustness and the performance of generated OAs are below expectation. To this end, we develop FIONA, a novel framework for Fine-grained Incoherent Optical DNN Accelerator search towards both superior acceleration efficiency and inference robustness. Extensive experiments and ablation studies validate the effectiveness of FIONA and the superior performance of generated OAs.
TimeTuesday, July 11th3:40pm - 3:55pm PDT
Location3003, 3rd Floor
AI/ML Architecture Design