Autonomous on-chip interferometry for reconfigurable optical waveform generation

Abstract

The generation of user-defined optical temporal waveforms with picosecond resolution is an essential task for many applications, ranging from telecommunications to laser engineering. Realizing this functionality in an on-chip reconfigurable platform remains a significant challenge. Towards this goal, autonomous optimization methods are fundamental to counter fabrication imperfections and environmental variations, as well as to enable a wider range of accessible waveform shapes and durations. In this work, we introduce and demonstrate a self-adjusting on-chip optical pulse-shaper based on the concept of temporal coherence synthesis. The scheme enables on-the-fly reconfigurability of output optical waveforms by using an all-optical sampling technique in combination with an evolutionary optimization algorithm. We further show that particle-swarm optimization can outperform more commonly used algorithms in terms of convergence time. Hence, our system combines all key ingredients for realizing fully on-chip smart optical waveform generators for next-generation applications in telecommunications, laser engineering, and nonlinear optics.

Scalable, Autonomous On-Chip Picosecond Pulse-Shaping Enabled by Smart Optimization

“…In our implementation, the chip is based on a silicon-oxy-nitride glass that offers exceptionally low linear and nonlinear losses, and is coupled to standard single-mode fibers for ease of use [5,8]. The platform consists of an on-chip chain of interferometers with bit-wise increasing delays with a resoltuion of 1 ps.…”

“…However, to date, the narrow linewidth of sub-nanosecond sources has inhibited any demonstration of an efficient and adaptable scheme, which combines user-friendly, reconfigurable picosecond pulse-shaping with an efficient monitoring solution for on-the-fly flexibility and scalability. In our work [8], we demonstrate that the combination of integrated photonic platforms, all-optical sampling, and smart-optimization algorithms allows for the robust generation and control of user-defined waveforms, based on temporal coherence synthesis. We further demonstrate the scalability of the approach, covering the picosecond range from 3 ps to over 150 ps, and compare the impact of different optimization algorithms on performance.…”

Optimizing supercontinuum spectro-temporal properties by leveraging machine learning towards multi-photon microscopy

“…For the initial processing of the input pulse prior to HNLF injection (so that to generate versatile temporal patterns of multiple femtosecond pulses), we also consider the propagation of the initial laser pulses into our on-chip programmable delay line: the considered PDL structure comprises 8 cascaded unbalanced interferometers (with increasing delays) to split an initial single pulse into up to 256 individual pulses with 1 ps separation between two adjacent ones, as reported in (Wetzel et al, 2018) and (Fischer et al, 2021). Each interferometer is an unbalanced interferometric structure (made of two integrated optical waveguides, two 50:50 optical couplers and a phase shifter) to constitute a balanced MZI followed by a pair of unbalanced waveguides.…”

“…Frontiers in Photonics frontiersin.org optimization techniques respectively based on a GA and particle swarm optimization (PSO), each possessing its own advantages (Jiang, et al, 2010;Fischer et al, 2021): GA optimization includes single-objective GA to maximize 2PA and 3PA signals, but also multi-objective GA functions for conjointly optimizing two individual MPA processes: 2PA-2PA or 3PA-3PA for two different fluorophores, as well as 2PA-3PA for a selected fluorophore. In the case of multiobjective GA, the optimization yields a Pareto front (also called Pareto frontier) that represents a trade-off between two MPA signal optimization.…”

“…(Weiner 2011) Among those, an innovative approach based on integrated photonics has been demonstrated over the last years, by using cascaded on-chip Mach-Zehnder interferometers (MZIs) in combination with machine learning algorithms. This strategy was for instance instrumental for the demonstration of picosecond waveform laser pulses being autonomously reconfigured (Fischer et al, 2021). In a similar framework, a genetic algorithm (GA) optimization was recently used for the formation of reconfigurable pulse patterns so that to maximize the intensity of selected SC wavelengths after nonlinear fiber propagation (Wetzel et al, 2018).…”