Oregon State University Develops Brain-Inspired Memory Component That Integrates Light Sensing and AI Processing for Greater Energy Efficiency

Researchers at Oregon State University have developed a brain-inspired phototransistor that integrates light sensing, memory storage, and signal processing into a single device — a breakthrough that could revolutionize the energy efficiency of future AI vision systems.

Breaking Through the Bottleneck of Conventional Architectures

Traditional AI sensing systems typically distribute sensing, memory, and computation across separate hardware components, requiring frequent data transfers between them. This constant movement of data not only introduces latency but is also a primary source of energy consumption. The Oregon State University research team addressed this fundamental inefficiency by drawing on the design principles of biological neural synapses, consolidating all three functions into a single phototransistor device and dramatically reducing the need for data movement.

Electronically Controllable Optical Memory

The most significant feature of this device is that researchers can electronically and precisely control the duration of "optical memory" — in other words, dynamically adjusting how quickly the device retains or forgets a particular light signal. This flexible memory control mechanism enables the component to mimic the short-term and long-term memory behavior found in biological neural systems, further enhancing the intelligence of signal processing.

Implications for AI Vision Systems and Drone Applications

This technological advance carries considerable significance for devices equipped with AI vision systems, including drone-mounted onboard sensors. When performing autonomous flight, obstacle detection, and environmental monitoring missions, drones require intensive real-time image processing, and high power consumption has long been a key constraint on flight endurance. If this type of brain-inspired phototransistor technology can eventually be applied to edge computing sensing modules, it could substantially reduce the power draw of AI vision systems without sacrificing processing performance — indirectly extending drone flight times and operational capabilities.

Looking Ahead

The research is currently at the fundamental research stage, and commercial mass production remains some way off. Nevertheless, this brain-inspired component — integrating sensing, memory, and processing in one device — represents a new direction in AI hardware architecture design and lays an important technical foundation for the development of next-generation energy-efficient AI sensors.

原文來源： 查看原文