Optical neuromorphic computing based on chaotic frequency combs in nonlinear microresonators
Negar Shaabani Shishavan 1, Egor Manuylovich 1, Morteza Kamalian-Kopae 2, Auro M. Perego 1, · 2025
Read Optical neuromorphic computing based on chaotic frequency combs in nonlinear microresonators by Negar Shaabani Shishavan 1, Egor Manuylovich 1, Morteza Kamalian-Kopae 2, Auro M. Perego 1, (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-17113.txt.
Introduction
Negar Shaabani Shishavan, Egor Manuylovich, Morteza Kamalian-Kopae, and Auro Perego build a light-based computing system in this 2025 paper. Their system works like a reservoir computer, which uses a fixed complex system to process signals. Here the reservoir is a tiny optical ring that forms chaotic combs of frequencies. No delay line is needed.
The ring's high quality factor gives both memory and a rich nonlinear map of inputs. In computer tests, the system forecast about a thousand symbols of a chaotic series with good accuracy. It needed no special tuning for that task. The authors point to on-chip fit, small size, high speed, and low power use. The work opens a path to optical brain-like computing hardware.
Catalog record
| CREATOR | Negar Shaabani Shishavan 1, Egor Manuylovich 1, Morteza Kamalian-Kopae 2, Auro M. Perego 1, |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 39 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-17113
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Optical neuromorphic computing based on chaotic frequency combs in nonlinear microresonators Negar Shaabani Shishavan 1, Egor Manuylovich 1, Morteza Kamalian-Kopae 2, Auro M. Perego 1, Abstract In this work we present a novel implementation of delay line free reservoir computing based on state-of-the-art photonic technologies, which exploits chaotic optical frequency comb formation in optical microresonator as the nonlinear reservoir. Our solution leverages the high resonator Q Q -factor both for memory and for enhancing high dimensional nonlinear mapping of input symbols. We numerically demonstrate the accurate prediction of about one thousand symbols in chaotic time series without the need of dedicated optimisation for specific tasks. Our results will enable design of optical neuromorphic computing architectures combining on-chip integrability, low footprint, high speed and low power consumption. Preprint: APS/123-QED Optical reservoir computing (RC) is a computational approach which leverages potential benefits of photonic platforms, such as low latency, high bandwidth, fast reconfigurability, and low power consumption [ 1 , 2 ] , to perform computation in a neuromorphic fashion going beyond the traditional Von Neumann paradigm. RC is a neuromorphic computing approach within the recurrent neural networks (RNNs) family [ 3 ] . Its operational principle relies on a mapping of the input signal into a higher dimensional space exploiting the nonlinear dynamics of a reservoir. By training only the output layer, without changing the inner layers parameters, RC significantly reduces training time compared to standard RNN, making it an efficient and scalable solution for various computational tasks, including time-series prediction, pattern recognition, and nonlinear signal processing. Photonics solutions for RC based on wave dynamics have been proposed [ 4 , 5 , 6 ] and have been implemented in several platforms including optical fibers [ 7 , 8 ] , optical parametric oscillators [ 9 , 10 ] , saturable absorbers [ 11 ] , media exhibiting quadratic nonlinearity [ 12 ] , disordered scattering media [ 13 ] semiconductor optical amplifiers [ 14 , 15 ] , semiconductor laser arrays [ 16 ] , phase-change materials [ 17 ] , and optical cavities and resonators [ 7 , 18 , 19 , 20 , 21 , 22 , 23 ] . Among photonic hardware for optical RC, microresonators (MR) are promising due to compact dimensions, possible on-chip integrability, CMOS compatibility, and low power consumption through nonlinear response achievable with input power in the mW order, enabled by high-quality ( Q ) factor platforms. In particular, so far, both theoretical and experimental investigation of RC in semiconductor MR have been performed leveraging two photon absorption and free carrier dispersion as a nonlinear tools for achieving classification and prediction tasks [ 19 , 20 , 22 , 21 , 18 , 23 , 24 ] . While several approaches to optical RC use delay lines – for instance in MR or in other settings as well – to enable system memory and symbol interaction [ 25 , 19 , 18 , 24 ] , in order to reduce system complexity and footprint, a particularly appealing emerging trend is the delay line free reservoir computer [ 19 , 26 ] , which does not require the use of auxiliary delay lines. In this work we present a novel method for optical RC without delay line, which exploits four-wave mixing assisted chaotic optical frequency combs (OFCs) in high Q -factor Kerr MR. Our approach (i) proposes for the first time to exploit modulation instability (MI) gain induced OFC generation to perform optical computing; (ii) it leverages intrinsic high- Q MR memory capabilities bypassing the need of adding extra delay lines, which enhances computational speed; (iii) it enables nonlinear interaction between optical modes of the systems achievable for few tens of mW optical power injection to operate higher dimensionality transformation of input symbols necessary for computation, hence offering low power consumption. Optical frequency combs are ultraprecise optical rulers consisting of equally spaced coherent frequency laser lines that can usually be generated by several platforms, including mode-locked lasers [ 27 ] , electro-optic modulators [ 28 ] and optical microresonators [ 29 ] . OFCs possess a variety of applications ranging from optical clocks to microscopy, from sensing to space exploration and telecommunications to mention just a few [ 30 ] . While for the vast majority of applications it is required that high stability and a fixed phase relation exist between the comb modes, in this work we leverage the potential technological application of less explored but easier to generate OFCs exhibiting random amplitude and phase fluctuations in their spectral lines. Figure 1: Overview of the proposed RC model. The driving field encodes time-series information into the light wave which drives the MR. The input light then enters the MR. The MR high-dimensional output state consists of power spectral densities corresponding to a chaotic frequency comb. Linear regression is then applied to combine generated spectral features, then the next symbols in the chaotic time series are predicted. The proposed RC scheme operates as follows: a symbols time series generated from a chaotic dynamical system is encoded, after suitable rescaling, into input pump power values at which an optical high-Q factor Kerr MR is driven. The symbol input (driving pump power) to the resonator is updated at a rate of 1 GSa/s. The nonlinear dynamics of light inside the resonator, due to the MI process, induces a dramatic spectral broadening leading to the generation of a chaotic frequency comb corresponding to a large number of oscillating equally spaced cavity resonances with amplitudes and phases which randomly change in time (see Supplemental Material for more details). The readout layer consists of the power spectral densities of the MR individual resonances as they can be detected by an optical spectrum analyzer. After training the reservoir, we applied linear regression to combine the output features and predict the next symbol in the time series. In high- Q resonators, the photon lifetime, given by \tau_{p}=Q/\omega_{0} being \omega_{0} the carrier frequency, can significantly exceed the inverse of the input symbol rate. This extended photon lifetime ensures that the spectral features generated by one input symbol persist in the cavity long enough to overlap with subsequent symbols enabling interaction of different symbols inside the cavity. This is further enhanced by the bistable response of Kerr cavities which is associated to an hysteresis cycle of the intracavity power as a function of the injection. Figure 1 provides an overview of the proposed RC model, illustrating each step: from data encoding till time series prediction. We modeled the dynamics of light inside the MR using the established mean field Lugiato-Lefever equation (LLE) in the notation of [ 31 ] , which describes the evolution over time t of the electric field slowly varying envelope E(t,\tau) : \frac{\partial E}{\partial t}=-\frac{(\alpha+i\delta_{0})}{t_{R}}E-i\frac{L}{t_{R}}\frac{\beta_{2}}{2}\frac{\partial^{2}E}{\partial\tau^{2}}+i\frac{\gamma L}{t_{R}}|E|^{2}E+\frac{\sqrt{\theta}}{t_{R}}E_{in}. (1) Here, t_{R} represents the round-trip time of the cavity, \delta_{0} is the detuning of the driving field from the nearest linear resonance, L is the cavity length, \beta_{2} the group velocity dispersion coefficient, \gamma the nonlinearity coefficient, E_{in} is the continuous wave pump injection amplitude related to the injected power by P_{in}=|E_{in}|^{2} . \alpha and \theta characterize cavity losses and coupling strength as a function of the resonator Q -factor, which is