Photonic-Mediated Neuromorphic Computing Enabled by a Copper Oxide Microcrystal Optoelectronic Synapse

The concept of photonic neuromorphic computing offers fast, energy-efficient, and autonomous data processing yet faces challenges in the design of an active material, enabling the desired performance. Here, we demonstrate a copper oxide microcrystal optical synapse, demonstrating efficient, fast, and highly enhanced photonic neuromorphic computing. By optically pumping a single microcrystal with 2.3 eV photons, we observe a history-dependent response of photoexcited electrons (spike), controlled by the pumping repetition rate. This neuromorphic behavior exhibits a 1 ms spike response time, a 102 on/off ratio, exceptional endurance over 13,400 cycles, and allows achieving 95% accuracy in handwritten digit recognition in three training epochs. The reported optical synapse surpasses most existing designs, paving the way for efficient and long-lasting photonic neuromorphic data processing.