GoKawiilResearchers at Oregon State University have developed a light-sensitive digital memory device that combines sensing, memory, and signal processing inside a single phototransistor, potentially reducing the energy cost of future AI hardware. The device, developed at Oregon State University’s College of Engineering and published in Advanced Functional Materials, is also designed to mimic the brain’s crucial ability to strengthen important memories while allowing less useful information to fade over time.
The new device brings AI processing closer to the sensor, rather than forcing data to travel between separate hardware blocks, so some of the work happens right where the light lands. "Our optoelectronic device introduces a new hardware capability that may enable more efficient processing of information directly at the sensor level," said Larry Cheng, project leader and professor of electrical engineering and computer science.
Today's AI hardware splits sensing, memory, and processing — the key jobs involved in machine perception — across separate components, which means data has to constantly shuttle between them. This shuttling consumes energy and reduces efficiency.
Latest Videos From Watch full video here:
The Oregon State device addresses this challenge by moving some memory and processing functions directly into the light sensor. It does this using a phototransistor made from two different materials. An oxide semiconductor forms the transistor channel, which is the pathway through which current flows. A photosensitive organic layer sits on top, absorbing light and generating electrical charges.
When light hits the device, some of those charges become trapped inside the photosensitive layer. Even after the light disappears, the trapped charges continue to affect the current flowing through the semiconductor channel. In effect, the device retains a memory of the optical signal it previously detected.
The clever part is that this memory is not static. By applying a small electrical gate voltage, the researchers can change where the trapped charges sit relative to the transistor channel. When the charges are moved closer to the channel, their effect becomes stronger, and the memory lasts longer. When they are moved farther away, the effect weakens, and the memory fades more quickly.
That behavior loosely resembles how biological brains regulate memory. In the brain, chemical signals influence whether a memory is reinforced or allowed to fade. In OSU’s device, an electrical signal performs a similar role, giving the hardware a programmable memory lifetime.
Stay On the Cutting Edge: Get the Tom's Hardware Newsletter Get Tom's Hardware's best news and in-depth reviews, straight to your inbox. Contact me with news and offers from other Future brands Receive email from us on behalf of our trusted partners or sponsors
This could be especially useful for neuromorphic computing, a field that tries to build computing systems modeled on biological neural networks. It also fits into the broader push toward in-sensor computing, where data is processed at the point of capture rather than being shuttled off to separate processors and memory banks.
... continue reading