Scientists have developed a brain-inspired device that detects light, stores information as memory, and allows electronic control over whether that memory strengthens or fades. This could help artificial intelligence (AI) systems run faster while using less electricity.
The device combines sensing, memory, and processing in a single phototransistor, which is a light-responsive electronic switch. Current AI systems usually separate these tasks across multiple components, forcing data to move between them and thereby raising energy use while lowering speed. In conventional systems the separation of functions requires constant data movement that consumes extra power. The new integrated design reduces that movement by handling everything in one place.
How the new device controls memory duration
Light generates electrical charges that become trapped and act as stored memory in the device. A small electrical signal can then adjust the strength of this memory, much like chemical signals in the brain regulate memory persistence. This tunability supports the development of neuromorphic computing, which copies the way biological neural systems handle information. It also advances in-sensor computing, where data processing happens directly at the sensor instead of being sent elsewhere.
The light-sensitive memory with programmable lifetime provides a flexible period for processing visual and other sensor data on the spot. This feature could improve vision systems and sensor-based artificial intelligence technologies.
The device uses two materials with distinct jobs. An oxide semiconductor creates the transistor channel that carries electric current. An organic photosensitive material on top absorbs light and produces charges, some of which get trapped. These trapped charges continue influencing the current after light ends, preserving a memory of past signals. What makes the approach different is the ability to move the charges rather than leaving them fixed. This movement is controlled electronically without needing to rewrite the entire memory each time.
Unlike fixed charges in ordinary memory, these can be repositioned using gate voltage, an electrical control signal. Moving trapped charges nearer the channel strengthens their effect and makes memory last longer. Moving them away weakens the effect and lets memory fade faster.
This research is published in Advanced Functional Materials.