en.Wedoany.com Reported - Researchers at Oregon State University have developed a photosensitive digital memory device that integrates sensing, storage, and signal processing into a single phototransistor, potentially reducing the energy costs of future AI hardware.
The device, developed by the College of Engineering at Oregon State University, was published in the journal Advanced Functional Materials. Its design mimics the brain's key ability to strengthen important memories while allowing secondary information to fade over time. The new device moves AI processing closer to the sensor, rather than forcing data to be transmitted between separate hardware modules, thus completing part of the work directly at the point where light strikes.
Project lead Larry Cheng, a professor of electrical engineering and computer science, stated: "Our optoelectronic device introduces a new hardware capability that may enable more efficient information processing directly at the sensor level." Current AI hardware distributes the sensing, storage, and processing involved in machine perception across different components, requiring data to frequently shuttle between them, which consumes energy and reduces efficiency.
Oregon State University's device addresses this challenge by integrating some storage and processing functions directly into the optical sensor. It employs a phototransistor made from two different materials: an oxide semiconductor forms the transistor channel (the path through which current flows), and a photosensitive organic layer sits on top to absorb light and generate charge. When light strikes the device, some charge becomes trapped within the photosensitive layer. Even after the light disappears, the trapped charge continues to influence the current flowing through the semiconductor channel, allowing the device to retain a memory of previously detected light signals.
The memory is not static. By applying a small electrical gate voltage, researchers can alter the position of the trapped charge relative to the transistor channel. When the charge moves closer to the channel, its effect strengthens, and the memory lasts longer; when it moves away, the effect weakens, and the memory fades more quickly. This behavior resembles how biological brains regulate memory: in the brain, chemical signals determine whether a memory is reinforced or forgotten; in the device, electrical signals play a similar role, endowing the hardware with a programmable memory lifetime.
This is particularly useful for the field of neuromorphic computing, which seeks to build computer systems that mimic biological neural networks. It also aligns with the broader trend toward in-sensor computing, where data is processed at the point of capture rather than being transmitted to separate processors and storage repositories. For AI vision systems, this means hardware can filter, weight, and temporarily retain visual information before it reaches traditional processors. Robots, drones, security cameras, or autonomous systems may not need to permanently store every visual signal; some information requires only brief attention, some needs to be retained longer, and others should disappear almost immediately.
Larry Cheng said: "This photosensitive memory with programmable memory lifetime creates a tunable time window for processing visual and other sensor signals directly at the detection point. This capability may enable more efficient vision systems and other sensor-based AI technologies." The research remains at the device level and cannot directly replace current AI accelerators or image sensors. However, this direction points to a hardware development path that could make future AI systems less reliant on the constant movement of data between sensors, memory, and processors. If successfully scaled, AI devices could become faster, more compact, and more energy-efficient, especially in edge systems where energy efficiency is critical.
This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com
