UT Austin / ASU (SimPhony Team)

Type: Research Institution (Electronic-Photonic Design Automation) Institutions: University of Texas at Austin; Arizona State University Key authors: Hanqing Zhu, Shupeng Ning, Hongjian Zhou, Ziang Yin, Ray T. Chen, Jiaqi Gu, David Z. Pan

The SimPhony team at UT Austin and Arizona State University produced the most rigorous system-level benchmarking framework for photonic AI computing as of April 2026. Their paper "Harnessing Photonics for Machine Intelligence" (arXiv 2604.10841) shifts the field's unit of analysis from device metrics to end-to-end system efficiency — modeling the full datapath: optical cores, DAC/ADC converters, memory traffic, and laser power together.

The findings challenge the optimism embedded in most photonic compute benchmarks. When peripheral overheads are properly accounted, the DAC/ADC conversion dominates the energy budget — not the optical computation. MZI mesh architectures, the most widely studied approach, prove "fundamentally ill-suited" for Transformer workloads. The time-multiplexed crossbar architecture emerges as the most competitive option on the density-efficiency Pareto frontier.

The team also produces an Electronic-Photonic Design Automation (EPDA) roadmap covering AI-assisted device simulation, cross-layer co-simulation, architecture modeling, inverse design, and layout automation — positioning full-stack photonic design automation as a necessary infrastructure for the field's maturation.

Key Contributions

• SimPhony: cross-layer benchmarking tool modeling heterogeneous electronic-photonic systems (Harnessing Photonics)
• DAC/ADC dominates energy budget at system level — not optical compute (Harnessing Photonics)
• MZI mesh architectures fundamentally ill-suited for Transformer workloads (Harnessing Photonics)
• Time-multiplexed crossbar: competitive with A100, surpasses B200 on energy efficiency (Harnessing Photonics)
• EPDA roadmap for full-stack photonic design automation (Harnessing Photonics)

Mentioned In

• Photonic Computing Limitations — Primary source for system-level limitation analysis
• Photonic Neural Networks — Architecture taxonomy and benchmarking
• Photonic Accelerators — System-level efficiency benchmarks

Related Entities

• University of Sydney — Hardware demonstrations that SimPhony framework evaluates