Computing with Fiber-Optics solitons PERSPECTIVES
COMPUTING WITH FIBER-OPTICS SOLITONS
///////////////////////////////////////////////////////////////////////////////////////////////////
Goëry GENTY 1,*, Mathilde HARY 2, Andrei ERMOLAEV 2, Daniel BRUNNER 2, and John M. DUDLEY 2
1
Laboratory of Photonics, Tampere University, Tampere, Finland
2
Institut Femto-St, Université Marie et Louis Pasteur, Besançon, France * goery. genty @ tuni. fi
https:// doi. org / 10.1051 / photon / 202513465
Optical solitons, discovered in optical fibers fifty years ago, have long been celebrated as robust manifestations of nonlinear wave physics. Initially conceived as candidates for long-haul communications, their role has since evolved toward a broader landscape of nonlinear dynamics and ultrafast photonics. In one of their most recent applications, they have emerged as computational blocks for optical machine learning. Nonlinear fiber dynamics naturally generate high-dimensional feature spaces, robust nonlinear transformations, and femtosecond-scale processing speeds, properties that electronics cannot match.
In this perspective, we focus on soliton-inspired optical computing, with particular emphasis on fiber-based extreme learning machines( ELMs). We highlight recent experimental demonstrations and numerical modeling that establish soliton propagation as a potential powerful computational substrate. We also discuss opportunities and challenges for taking advantage of nonlinear fiber dynamics in machine learning, from multimode soliton processors to hybrid integrated systems, and argue that solitons may enable the next generation of ultrafast, energy-efficient, and noise-resilient computing.
The discovery of optical solitons in fiber optics in the early 1970s marked a milestone in nonlinear science [ 1 ]. Solitons, formed from the balance of Kerr nonlinearity and chromatic dispersion, represent self-organized states of light propagating without distortion over long distances. Fiber solitons were first extensively studied in the context of high-capacity telecommunications systems, driving many advances in the design of modern optical systems. More recently, they have had major impact in the development of broadband supercontinuum light sources, where soliton dynamics play a central role in driving the spectral broadening. Beyond these applications, solitons have influenced diverse areas of physics and served as a versatile testbed for nonlinear wave theory. Today, solitons are again finding a new application, this time as computational primitives for optical computing. As electronics approach fundamental speed and energy limits, nonlinear photonics provides an attractive route to achieve ultrafast, energy-efficient information processing. Solitons, with their robustness, dynamics, and compatibility with ultrafast lasers, provide a natural basis for optical machine learning hardware.
FIBER-OPTICS SOLITONS AS COMPUTATIONAL PLATFORM The benefit of optical computing lies in harnessing the intrinsic properties of light: massive parallelism, ultrahigh bandwidth, and the ability to harness its nonlinear response. Within the different photonic computing approaches, extreme
Photoniques 134 I www. photoniques. com 65