Optical Tech Would Update a Robot’s AI on the Fly

2026-07-26 · Source: IEEE Spectrum · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Emerging Technologies & Innovation · Depth: Advanced, quick

Summary

Cornell Tech researchers Yifan He and Jae-sun Seo have developed a novel optical receiver design, presented at the IEEE/JSAP Symposium on VLSI Technology & Circuits, that directly alters its own static random-access memory (SRAM) using photocurrents from beamed, QR code-like light arrays. This innovation enables fully digital optical communication, allowing AI chips to update model parameters with significantly less energy than traditional electrical connections between dynamic random-access memory (DRAM) and processors. The system addresses a major bottleneck in scaling AI systems by eliminating power-hungry analog circuits. While the current proof of concept uses a static 14x14-bit matrix, the team aims for gigabit-per-second transmission rates. Although current photosensitive bit cells are larger than conventional SRAM, efforts are underway to shrink them using CMOS scaling, targeting applications in edge AI, such as robotics and microrobots, to save time and energy in model updates.

Key takeaway

For AI Hardware Engineers designing next-generation edge AI systems, this optical memory update technology offers a path to significantly reduce energy consumption and overcome DRAM-processor bottlenecks. You should evaluate its potential for on-the-fly AI model updates in robotics and autonomous vehicles, despite current memory density trade-offs. Monitor ongoing research into CMOS scaling for bit cell size reduction, as this could enable more efficient, memory-rich optical integration in your future designs.

Key insights

A new optical receiver directly programs AI model parameters into SRAM using light, bypassing electrical bottlenecks.

Principles

• Optical data links reduce energy loss.
• Direct light-to-memory updates avoid analog circuits.
• SRAM cells can integrate photodiodes for optical input.

Method

A transmitter beams light matrices to modified SRAM cells with photodiodes. Light generates current, flipping binary values to update AI model parameters, with calibration for alignment.

In practice

• Update AI robot models in factories.
• Enhance memory-constrained microrobots.
• Lower energy for edge AI applications.

Topics

• Optical Communication
• Edge AI
• AI Model Parameters
• SRAM Memory
• VLSI Technology
• Robotics

Best for: AI Scientist, AI Hardware Engineer, AI Architect, Research Scientist

Related on AIssential

• Photonics: A Foundational Scaling Layer for AI-Era Computing — Photonics is becoming the foundational layer for scaling AI computing by addressing system-level data movement and memory bottlenecks.
• From Vision-Language Models to Physical AI: Embedded Intelligence Enters a New Phase — Embedded AI is evolving towards practical, multimodal intelligence at the edge, requiring innovation in models and deployment.
• Brain-Inspired Light Sensor Could Speed AI Image Processing — A novel phototransistor integrates light detection, data storage, and tunable forgetting directly on-sensor, mimicking brain function.
• This Breakthrough Could Kill the GPU Race — Optical computing with active metasurfaces offers a path to dramatically reduce AI's energy footprint by performing computation directly with light.
• Google's "Frozen v2" chip reportedly bakes Gemini's architecture directly into silicon for efficiency gains — Google's "Frozen v2" chip hardcodes Gemini's architecture for 6-10x AI inference efficiency, deploying 2028.
• 😺 🎙️ Watch: Is Brain-like Computing What's Next? — Brain-inspired hardware using light and integrated memory-compute can overcome traditional AI bottlenecks.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by IEEE Spectrum.