Human Brain-Inspired Digital Model Could Improve AI’s Energy Efficiency

Current AI hardware isn’t exactly as energy-efficient as engineers want it to be. Researchers at Oregon State University are working on changing that with a device that mimics the human brain’s ability to handle memory. The light-sensitive technology could make AI systems run faster with less electricity.

AI Mimics the Brain

Right now, AI hardware spreads out light sensing, memory, signal processing, and other functions. Information must travel between these separate parts, which increases energy demands and lowers efficiency. To address this, project leader Larry Cheng and his team combined all those functions into a single piece of hardware called a phototransistor.

“Our optoelectronic device introduces a new hardware capability that may enable more efficient processing of information directly at the sensor level,” said Cheng.

While most tech is built to hold onto data, this device controls how memories fade over time, similar to how chemical signals work in our own brains.

“Unlike conventional memory that is designed to preserve information, our device can electronically control how memories strengthen or decay,” Cheng said.

Moving Charges Around

The device blends a bottom layer that acts as a channel to carry electrical current, while a top layer absorbs light. When light hits the device, it creates trapped electrical charges that act as a memory of that light.

“What makes this work unique is that the stored charges are not fixed in place,” Cheng said.

By adding a small electrical voltage, researchers can actually move those trapped charges. Pushing them closer to the current pathway makes their influence stronger, meaning the memory lasts longer. Pulling them further away makes the memory fade faster.

This brain-inspired approach is a big step toward neuromorphic computing. It helps process dynamic information efficiently right where it is collected.

“This light-sensitive memory with a programmable memory lifetime creates a tunable time window for processing visual and other sensor signals directly where they are detected, a capability that could enable more efficient vision systems and other sensor-based AI technologies,” Cheng added.