MIT Engineers Develop Modular AI Chips With Light-Based Communication
MIT Engineers have developed a modular AI chip system which uses light to communicate between chips rather than wires for more easily performed reconfiguration.
Created on June 15|Last edited on June 15
Comment
A team of engineers at the Massachusetts Institute of Technology have had their research on a new chip built for AI processing published recently. The chips are LEGO-like in that they can be easily and endlessly reconfigured and connected to each other, allowing for optimal conditions depending on the kind of AI model you're running.
The thing that makes these chips stand out from other modular designs is that all chip-to-chip communication is facilitated through LEDs - these chips communicate with light. Not only is the connection faster than that of physically wired designs, but reconfiguration is much easier without the need for rewiring.
Each individual chip can add a different integral piece to the system - sensors, computing layers, etc. The researchers stacked a couple of chips on top of each other, an eye layer for vision and a classification layer for computing, to classify 5x5 pixel images of the letter M, I, and T to good results.
Find out more
Add a comment
Tags: ML News
Iterate on AI agents and models faster. Try Weights & Biases today.