Search for: All records

Electromagnetic coupling is ubiquitous in photonic systems and transfers optical signals from one device to the other, creating crosstalk between devices. While this allows the functionality of some photonic components such as couplers, it limits the integration density of photonic chips, and many approaches have been proposed to reduce the crosstalk. However, due to the wave nature of light, complete elimination of crosstalk between closely spaced, identical waveguides is believed to be impossible and has not been observed experimentally. Here we show an exceptional coupling that can completely suppresses the crosstalk utilizing highly anisotropic photonic metamaterials. The anisotropic dielectric perturbations in the metamaterial mutually cancel the couplings from different field components, resulting in an infinitely long coupling length. We demonstrate the extreme suppression of crosstalk via exceptional coupling on a silicon-on-insulator platform, which is compatible with a complementary metal-oxide-semiconductor process. The idea of exceptional coupling with anisotropic metamaterials can be applied to many other electromagnetic devices, and it could drastically increase the integration density of photonic chips.

The incorporation of high‐performance optoelectronic devices into photonic neuromorphic processors can substantially accelerate computationally intensive matrix multiplication operations in machine learning (ML) algorithms. However, the conventional designs of individual devices and system are largely disconnected, and the system optimization is limited to the manual exploration of a small design space. Here, a device‐system end‐to‐end design methodology is reported to optimize a free‐space optical general matrix multiplication (GEMM) hardware accelerator by engineering a spatially reconfigurable array made from chalcogenide phase change materials. With a highly parallelized integrated hardware emulator with experimental information, the design of unit device to directly optimize GEMM calculation accuracy is achieved by exploring a large parameter space through reinforcement learning algorithms, including deep Q‐learning neural network, Bayesian optimization, and their cascaded approach. The algorithm‐generated physical quantities show a clear correlation between system performance metrics and device specifications. Furthermore, physics‐aware training approaches are employed to deploy optimized hardware to the tasks of image classification, materials discovery, and a closed‐loop design of optical ML accelerators. The demonstrated framework offers insights into the end‐to‐end and co‐design of optoelectronic devices and systems with reduced human supervision and domain knowledge barriers.