Three-in-One Diode Integrates Sensing, Memory and Processing for Smart Cameras

Why It Matters

Integrating sensing, memory and compute eliminates costly data shuttling, enabling ultra‑compact, energy‑efficient smart cameras for edge AI applications.

Key Takeaways

• Diode merges sensing, storage, compute in two terminals
• Aluminum gallium nitride layer provides electron reservoir for memory
• Eight distinct electrical states enable multilevel photomemory
• Array achieves over 95% accuracy on noisy image classification
• Removes separate readout circuitry, cutting power and area

Pulse Analysis

Modern imaging pipelines still rely on discrete sensor, memory and processor blocks, a structure that forces large volumes of raw pixel data to travel across chip interconnects. This data movement consumes a disproportionate share of power and latency, especially in battery‑powered edge devices such as autonomous drones, wearables and smart surveillance cameras. As AI workloads shift toward the edge, manufacturers are scrambling for architectures that can compress the sensing‑to‑inference chain without sacrificing performance.

The breakthrough from USTC hinges on band‑structure engineering of a GaN‑based p‑n diode. By growing a thin AlGaN layer with a wider bandgap between p‑type and n‑type gallium nitride, the team created an intrinsic electron reservoir that can be charged, held, and released on demand. This enables three functional modes: a self‑powered photodetector at zero bias, a voltage‑biased photosynapse that gradually releases trapped carriers for real‑time noise filtering, and programmable voltage pulses that write and erase up to eight distinct memory states. The entire stack is fabricated via molecular‑beam epitaxy on silicon, preserving compatibility with existing semiconductor manufacturing lines.

The integrated diode array demonstrates that neuromorphic image processing can be performed directly at the pixel level, delivering >95% classification accuracy on noisy clothing images without external denoising hardware. By collapsing sensor, memory and compute into a two‑terminal component, system designers can dramatically reduce board space, power budgets and bill‑of‑materials costs. This paradigm is poised to accelerate the rollout of smart cameras in robotics, AR/VR headsets and IoT edge nodes, while opening new research avenues for ultra‑dense optoelectronic neuromorphic networks.