The inception of the memristor traces back to 1971 when Leon Chua introduced the concept of a nonlinear resistor with memory. Nearly four decades later, in 2008, Stan Williams and his team at Hewlett Packard Labs established a tangible connection between Chua's theoretical proposition and their experimental findings, heralding the discovery of the fourth fundamental circuit element. Their demonstration featured a relatively straightforward device—a bilayer of titanium dioxide nestled between two electrodes—yet its behavior was remarkably intricate, showcasing a resistance modulated by past current interactions.

Fast forward to 2024, the surge in artificial intelligence (AI) innovation has precipitated a soaring demand for computational power, doubling every two months and surpassing the trajectory of Moore's Law. Consequently, as CMOS transistor scaling approaches its limits, there's a growing impetus to explore alternative technologies, diverging from the conventional von-Neumann architecture that segregates memory and computation. The advent of electrical memristors has kindled interest in in-memory computing, promising to curtail the overheads associated with data transfer. Moreover, these devices have emerged as promising contenders for implementing neuromorphic computing systems, aimed at emulating the cognitive processes of mammalian brains.

At Hewlett Packard Labs, a number of us in the Large-Scale-Integrated-Photonics (LSIP) Lab worked on heterogeneous integration of III-V compounds onto silicon for the purposes of realizing ultra-energy-efficient optical communications. Despite being aware of memristor research next door, it wasn't until later that we realized our heterogeneous material platform inherently harbored memristive properties within the photonic devices themselves in the form of thin dielectrics between III-V and silicon! Slowly, we started applying experimental techniques we learned from our electrical memristor colleagues, and sure enough, we were able to imprint electrical non-volatile memristive behavior onto various photonic devices.

               In this work, we leverage our heterogeneous III-V/Si optical interconnect platform to co-integrate memristors based on semiconductor-insulator-semiconductor capacitors. This platform simultaneously allows seamless integration of quantum dot lasers, modulators, photo-detectors, and optical filters which are all necessary for a fully integrated optical in-memory hardware accelerator. The Mach-Zehnder optical memristor exhibits non-volatile optical phase shifts enabling ~33 dB signal extinction while consuming 0 electrical power consumption. We demonstrate 6 non-volatile states each capable of 4 Gbps modulation. More advanced filters were also demonstrated to exhibit memristive non-volatile passband transformation with full set/reset states. Time duration tests were performed on all devices and indicated non-volatile lifetimes up to 24 hours and beyond. We believe these initial results can pave the way for in-memory optical computing and large-scale non-volatile photonic circuits.