Key Findings
A research team led by Associate Professor Kohei Nakajima of the University of Tokyo and researchers from NTT has developed an optical computing framework that reduces AI processing requirements while simplifying model structure. Verified through simulations and physical experiments, the architecture matched or outperformed complex neural network baselines in certain tests while requiring substantially less computation.
Technical Details
The framework overcomes the technical demands of standard AI algorithms—such as extensive measurement and large data storage—through two primary innovations:
- SS-DFA Learning: The “state-skipping direct feedback alignment” method avoids recording most intermediate system states. Instead, it uses the final system error and structural system dynamics to guide learning updates.
- Single-Map Network: The team replaced complex internal connection topologies with a simplified single-map framework, where processing units follow unified general governing rules.
The physical platform processes information by converting data into optical signals, circulating them through a 200-meter optical fiber loop, and converting them back into electrical signals using electronic components for analysis.
Background & Context
Conventional AI relies on digital chips that transmit electrical signals through transistor networks, a process that consumes significant energy and generates heat. While optical computing offers high-speed parallel transmission and lower energy potential, standard learning algorithms are designed for digital computers and cannot operate effectively on physical substrates without redesign.
Significance & Outlook
The team has currently integrated the single-map architecture as modular blocks into baseline neural network designs. Future research will focus on scalability to larger multimodal models and diverse data types. If successfully miniaturized, this modular architecture could reduce energy and water usage in data centers by utilizing low-heat hardware, or be integrated into smartphones, cameras, and onboard vehicle systems.
Get our weekly technology intelligence — free
Receive an infographic that lets you judge at a glance whether each field’s analysis report is worth reading.
Subscribe Free — Weekly Tech Intelligence
By subscribing, you’ll receive Troy-Technical’s weekly technology intelligence newsletter.
- Your email and selected fields are used only to deliver the newsletter.
- We never share your information with third parties.
- You can unsubscribe anytime via the link in each email.
See our Privacy Policy for details.
Takes about a minute · Unsubscribe anytime



