Magneto-Optically Reconfigurable Photonic Neuromorphic Networks

The growing demand for AI computing power is critically unsustainable with current energy-hungry computational hardware. Neuromorphic computing, leveraging complex dynamics of physical systems to perform AI, has emerged as a hugely promising approach to tackling this issue. However, existing neuromorphic schemes employ a single physical system, and fall short when computational tasks are not well-matched to the specific system dynamics. For the field to progress and provide real answers to global AI energy consumption, there is urgent need for innovative solutions capable of high efficiency and versatility across diverse tasks.In MORPHON, I propose a new paradigm: a ‘hybrid' neuromorphic approach that leverages the complementary strengths of multiple physical systems, delivering a new class of energy-efficient, task-adaptive neuromorphic AI hardware. I will show that photonic semiconductor networks which host many nonlinearly-coupled lasing modes can be sensitively controlled by optically-switched nanomagnetic arrays, with strong preliminary data confirming the promise & feasibility of my approach. Leveraging my deep expertise in nanomagnetism, photonics & neuromorphic physics, I will demonstrate that hybrid magneto-photonic networks hold the key to unprecedented neuromorphic performance: vastly-reconfigurable magnetic arrays provide memory and task-adaptivity, and photonic lasing dynamics grant exemplary nonlinear vision processing.My hybrid systems will perform a wide range of bio-inspired AI tasks, including challenging image classification & medical diagnosis, and neuromorphic video processing - a key unmet need in the field. MORPHON will ignite the next-generation of neuromorphic AI hardware, unlocking the potential of hybrid systems that far exceed the capabilities of each system alone. This project will radically advance physical neuromorphic computing while addressing the crucial global need for sustainable AI.