Integrated Photonic Platform for the Edge: Accelerated Computing and Low-Latency Authentication

Overview

Privacy-sensitive edge systems increasingly require efficient local intelligence and low-latency secure operation under strict constraints in footprint, energy consumption, and response time. This dissertation develops integrated photonics as a unified hardware platform for parallel computing and chaos-driven dynamic authentication.
A photonic computing unit (PCU) based on a microring resonator crossbar is designed and fabricated, and its neural-network inference capability is experimentally validated. To improve reliability and deployability, a feedback-control-based strategy is developed for precise weight programming and long-term weight locking, with 11.3-bit benchmark precision under direct monitoring and about 6.25-bit practical precision under non-intrusive through-port monitoring. Hardware-aware model optimization and an in-situ programmable optical activation unit are also introduced to improve platform robustness and add reconfigurable nonlinear functionality.
The validated PCU architecture is then extended to a realistic medical Internet of Things scenario for networked multi-patient cardiac health monitoring. A wearable tag enables multimodal sensing and wireless transmission, while the PCU performs ward-level edge inference. The system also uses multimodal gating to suppress false positives and improve real-time monitoring reliability. In proof-of-concept hardware inference using 100 representative heartbeat segments, the photonic edge pipeline correctly classifies all tested normal and abnormal samples.

To move the PCU architecture toward more complete integration, heterogeneously integrated multi-channel quantum-dot lasers are introduced as on-chip light sources. Experimental source-to-core validation and numerical device-link-system modeling show that QD-laser integration can support a 1.7× projected scalability improvement, >40% computational-density enhancement, and ~30% energy-efficiency gain in a future system-in-package implementation.

In parallel, a chaos-based dynamic authentication framework is developed for low-latency hardware security. D-shaped vertical-cavity surface-emitting lasers serve as chaotic photonic entropy sources for physical unclonable functions, achieving >500 Gbps response/entropy rate and <1 pJ/bit energy consumption. A deep-learning-assisted key-verification method is presented to achieve a near-zero false positive rate, and a protected key-transmission strategy is further introduced to reduce key exposure during communication.

Together, these results show that integrated photonics can serve as a multifunctional edge platform for accelerated computing and low-latency authentication, providing a coherent path toward future privacy-sensitive edge systems.