Computing with Fiber-Optics solitons PERSPECTIVES
Figure 2. Applications of solitons as computational building blocks. Soliton dynamics provide a versatile platform for optical computing. Their robustness and nonlinear interactions enable a range of applications, including extreme learning machines for data classification, optical Ising machines and ultrafast solvers for optimization, optical memory based on circulating solitons, and logic gates constructed from soliton collisions and interactions.
Together, these results highlight a key distinction from electronic machine learning hardware. In electronics, nonlinear mappings are explicitly programmed into layered architectures, and dimensionality is expanded through iterative matrix operations. In contrast, soliton-driven fiber propagation physically generates a high-dimensional feature space in a single ultrafast pass, with dynamics governed by well-understood nonlinear physics rather than algorithmic construction. This makes the fiber itself a tunable kernel generator, inherently parallel, ultrafast, and energy efficient. In this sense, soliton dynamics provide not just an optical analog of electronic accelerators, but a new computational substrate well suited for machine learning.
FUTURE DIRECTIONS Looking ahead, soliton-based computing opens several promising directions. Using multimode fibers would allow spatiotemporal solitons, orbital angular momentum states, and intermodal interactions to provide
REFERENCES
[ 1 ] B. Kibler et al., Photoniques 122, 41( 2023) [ 2 ] M. Hary et al., Nanophotonics 14, 2733( 2025) [ 3 ] A. V. Ermolaev et al., Optics Letters 50, 4166( 2025) [ 4 ] B. Fischer et al., Advanced Science 10, 2303835( 2023) [ 5 ] K. F. Lee, M. E. Fermann, Physical Review A 109, 033521( 2024) [ 6 ] S. Saeed et al., Nanophotonics 14, 2749( 2025) [ 7 ] T. Ignaki et al., Science 354, 603( 2016) additional computational degrees of freedom, effectively expanding the dimensionality. Integration with metasurfaces, silicon photonics, or 2D materials may yield compact, tunable processors that combine the richness of fiber dynamics with the scalability of integrated platforms. Beyond ELMs, soliton physics offers opportunities to implement optical Ising machines [ 7 ], ultrafast optimization solvers, and neuromorphic processors operating at femtosecond timescales( see Figure 2 for a concept illustration). The broadband character of soliton interactions could make them attractive for edge AI in optical communication networks, where computation could be embedded directly into the transmission medium. Furthermore, soliton chaos and noise-resilient states may enable entirely new computing architectures such as stochastic optical processors.
CONCLUSION Solitons have been a cornerstone of nonlinear optics for five decades. Today, they are re-emerging not only as a subject of fundamental study but as a resource for computing. By bridging nonlinear wave physics with artificial intelligence, soliton-based computing opens a path toward ultrafast, energy-efficient, and noise-robust photonic processors. From long-haul communications to learning machines, solitons continue to find new applications in photonics.
Photoniques 134 I www. photoniques. com 67