DxPTA: An Architecture Design Space Exploration with Optical Dataflow-guided Strategy for HW/SW Co-Design of Photonic Transformer Accelerators
Researchers have developed DxPTA, a new methodology for designing photonic transformer accelerators (PTAs). This approach uses optical dataflow to guide hardware and software co-design, addressing limitations of previous manual methods that did not consider application constraints. DxPTA significantly reduces design time and finds suitable PTA architectures for various transformer models, achieving notable improvements in area, power, energy, and latency. AI
IMPACT Streamlines the development of energy-efficient hardware for advanced AI models, potentially accelerating AGI research.