Material-Defined Photonic Computing: A Full-Stack Co-Design Approach from Thin-Film Lithium Niobate Physics to Operating System Runtimes
The relentless scaling of AIoT applications is severely constrained by the von Neumann bottleneck, where continuous data movement between discrete sensors, memories, and processors dominates system energy and latency. To overcome this, this research proposes a full-stack co-design framework for material-defined photonic computing, collapsing sensing, non-volatile state retention, and in-situ execution into a unified Thin-Film Lithium Niobate (LNOI) substrate. By bridging low-level electro-optic device physics with a state-centric Instruction Set Architecture (ISA), a hardware-aware compiler, and a Photonic Operating System (PhOS) that manages predictive thermal calibration, Wavelength-Division Multiplexing (WDM) virtualization, and lazy conversion policies, this work demonstrates how a unified hardware-software stack can eliminate redundant electronic conversions and deliver orders-of-magnitude improvements in end-to-end energy efficiency and latency for real-time edge workloads.
