Nanoscale isolation layer design for non-volatile integrated photonics targeting high-energy efficiency programming and computing.

Abstract

The revolutionary advancements in artificial intelligence have precipitated an urgent need for low-latency and energy-efficient processing systems. Photonic in-memory computing stands out as a promising solution. Here, we introduce a novel, to the best of our knowledge, design for non-volatile integrated photonic devices that significantly reduces both programming and computing energy consumption. Through introducing a nanoscale isolation layer, the programming energy is reduced by over 80%. The device exhibits a programming precision of 6 bits and maintains consistent performance after 1000 switching cycles. Furthermore, the isolation layer mitigates the impact of non-ideal modulation of the device, reducing the insertion loss to 0.5 dB. This improvement is expected to decrease the computing energy consumption by more than 35%. This study provides new insights into high-energy efficiency integrated photonic computing and contributes to the large-scale integration of photonic computing systems.