Self-calibrating programmable photonic integrated circuits

• Published in: Nature photonics Vol. 16; no. 8; pp. 595 - 602
• Main Authors: Xu, Xingyuan, Ren, Guanghui, Feleppa, Tim, Liu, Xumeng, Boes, Andreas, Mitchell, Arnan, Lowery, Arthur J.
• Format: Journal Article
• Language: English
• Published: London Nature Publishing Group UK 07-07-2022; Nature Publishing Group
• Subjects: 639/624/1075/1076; 639/624/1075/1079; 639/624/1075/187; Algorithms; Applied and Technical Physics; Calibration; Computers; Crosstalk; Impulse response; Integrated circuits; Machine learning; Photonics; Physics; Physics and Astronomy; Quantum computers; Quantum Physics; Reconfiguration; Self calibration; Signal processing; Temperature gradients
• ISSN: 1749-4885; 1749-4893
• DOI: 10.1038/s41566-022-01020-z

Programmable photonic integrated circuits (PICs) are dense assemblies of tunable elements that provide flexible reconfigurability to enable different functions to be selected; however, due to manufacturing variations and thermal gradients that affect the optical phases of the elements, it is difficult to guarantee a stable correspondence between the electrical commands to the chip, and the function that it provides. Here we demonstrate a self-calibrating programmable PIC with full control over its complex impulse response, in the presence of thermal cross-talk between phase-tuning elements. Self-calibration is achieved by: (1) incorporating an optical reference path into the PIC; (2) using the Kramers–Kronig relationship to recover the phase response from amplitude measurements; and (3) applying a fast-converging self-calibration algorithm. We demonstrate dial-up signal processing functions with complex impulse responses using only 25 training iterations. This approach offers stable and accurate control of large-scale PICs, for demanding applications such as communications network reconfiguration, neuromorphic hardware accelerators and quantum computers.
Researchers demonstrate a self-calibrating programmable photonic integrated circuit. The findings may be useful for the accurate control of large-scale photonic integrated circuits in applications such as light-based machine learning.