Non-Hermitian lithography-free programmable nonlinear integrated photonic networks

Photonics serves as the backbone of modern information infrastructure, transmitting and processing data at unparalleled speeds with minimal energy consumption. However, the existing programmable integrated photonic platforms employ discrete, single-function devices, leading to exponential architectural complexity and hindering full programmability. In contrast to the state of the art, we explore programmable integrated photonic platforms driven by the imaginary part of the permittivity in semiconductor-based optical gain materials. An unprecedented lithography-free paradigm for integrated photonic computing is demonstrated in an unpatterened device driven by the imaginary index. This non-Hermitian platform enables field-programmability and dynamic robustness, culminating in a high-fidelity photonic matrix processor capable of real-time error correction and in-situ photonic network training. Furthermore, the capabilities of photonic field-programmability can be pushed into the nonlinear realm by meticulous spatial control of distributed carrier excitations and their dynamics, achieving programmable photonic nonlinear interconnects. Leveraging the polynomial building blocks, in-situ training of integrated photonic nonlinear networks is demonstrated. Our novel non-Hermitian integrated photonic nonlinear networks serve as a pioneering example in the exploration of photonic paradigms tailored for computing and networking with light.