A Kerr soliton Ising machine for combinatorial optimization problems

Yan Jin, Nitesh Chauhan, Jizhao Zang, Brian Edwards, Pratik Chaudhari, Firooz Aflatouni, Scott B. Papp · 2025

Read A Kerr soliton Ising machine for combinatorial optimization problems by Yan Jin, Nitesh Chauhan, Jizhao Zang, Brian Edwards, Pratik Chaudhari, Firooz Aflatouni, Scott B. Papp (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-2508-00810.txt.

Introduction

Yan Jin and six co-authors build a machine that solves hard choice tasks by the play of hundreds of light pulses in a tiny ring. Their paper is from two thousand twenty-five.

Such tasks grow too fast for plain digital machines. Light is a swift carrier of facts, but its parts barely touch, which makes direct work hard. Fresh non-linear photon gear gives strong play among pulse states. The team traps hundreds of pulses in one ring and ties them with a feedback web that can be set at will. By a tweak to self-feedback, each pulse takes on a spin-like split. The full set then acts as an Ising machine, a form built for choice tasks.

Catalog record

CREATORYan Jin, Nitesh Chauhan, Jizhao Zang, Brian Edwards, Pratik Chaudhari, Firooz Aflatouni, Scott B. Papp
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE37 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-00810

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

A Kerr soliton Ising machine for combinatorial optimization problems Yan Jin, Nitesh Chauhan, Jizhao Zang, Brian Edwards, Pratik Chaudhari, Firooz Aflatouni, Scott B. Papp Abstract The growing challenges of scaling digital computing motivate new approaches, especially through the dynamical evolution of physical systems that mimic neural networks and combinatorial optimization problems [ 1 , 2 , 3 ] . While light is a hyper efficient information carrier, intrinsically weak light interactions make direct information processing difficult to implement [ 4 ] . Recently, specialized nonlinear photonics have opened new controls over light fields with extraordinary bandwidth, coherence, and the emergence of strong interactions among nonlinear eigenstates like solitons [ 5 ] . We harness an ensemble of hundreds of Kerr-nonlinear microresonator solitons and implement an analog feedback network to create an Ising machine with fully programmable all-to-all interactions. By increasing the feedback for self, on-diagonal interactions, each soliton exhibits a universal spin-like bifurcation. Using this palette of interactions amongst the entire soliton ensemble, we encode the Ising machine to solve the benchmark Boolean satisfiability problem (SAT). The combination of uniform soliton interactions and the compatibility of our Ising machine with high-speed data interconnects enables rapid and precise solutions of complex SAT problems. Indeed, the soliton properties bound the tradeoff of optical power and time use by the machine at approximately 10 mW and 1 \mu s for a single feedback step. We performed >10,000 trials on more than 100 randomly generated SAT instances to evaluate the Ising machine, demonstrating the potential to exceed the performance of benchmark digital SAT solvers. Our work highlights the convergence of optical nonlinearity, ultralow loss photonics, and optoelectronic circuits, which can be combined for a wide range of computation-acceleration tasks. The end of rapid scaling in digital-computer technology and the rise of machine optimization and artificial intelligence has opened interest in controllable and scalable physical systems to solve combinatorial optimization problems, which are widely used in schedule optimization, finance, chemical discovery, and deep neural network architectures [ 1 , 6 , 7 , 8 ] . In particular, physical approaches leverage quantum-inspired behaviors like discrete spin interactions, featuring sharp state transitions and the potential for scalable spin networks with low operation power. Since combinatorial optimization can fall into the NP-complete category, they are challenging to solve with digital computers. Boolean satisfiability (SAT) is the canonical NP-complete example, meaning that every problem in NP can be efficiently reduced to SAT, and thus SAT serves as a central benchmark. Therefore, numerous non-von-Neumann architectures have been explored to solve these problems by mapping them to an Ising model. An extraordinary range of physical systems can be arranged to create an Ising machine, notably parametric oscillators [ 9 , 10 , 11 ] , opto-electronic networks [ 12 , 13 ] , Kerr nonlinear optics [ 14 ] , electronic circuits [ 8 , 15 ] , and CMOS circuits with static random-access memory (SRAM) [ 16 ] . Despite their connection with interacting quantum-spin systems, Ising machines are implemented with a range of architectures in which memory, connectivity, interaction speed, and scalability are key. A notional digital/analog electronic circuit can implement an Ising machine in which pristine spin states are stored in a digital SRAM array. Connectivity is realized through multiplexing of analog current signals derived from the SRAM. Nonlinearity is introduced via comparators that project the analog interaction sum back into discrete digital spin states, closing the feedback loop. However, the power consumption of digital-to-analog and analog-to-digital converters, latency in SRAM access, and signal propagation delay in comparators impose fundamental limits on update throughput and ultimately constrain the overall spin capacity and performance of such hybrid systems. These ubiquitous CMOS constraints motivate the exploration of alternative physical substrates like photonics that promise lower energy per operation and far greater parallelism. Figure 1: Kerr soliton Ising machine. (a) The resonator with soliton spins (circle) and the feedback network with the pump and the feedback circuit (oval). Pump pulses (orange) and solitons (blue) are described by |f(\theta_{i})|^{2} and |a(\theta_{i})|^{2} in the spatial domain, respectively, and \xi refers to the feedback circuit. (b) Schematic of the Ising machine, including the feedback circuit (green box), the soliton resonator and photodetection (blue box), and the pump (red box). (c) Structure of a 3-SAT problem, where V_{i} ( i\in\{1,2,...n\} ) and C_{i} ( i\in\{1,2,...m\} ) are variables and clauses, respectively. The solid lines connect variables, and dashed lines connect the negation of variables. (d) A 3-SAT problem with 64 variables and 192 clauses. (e) Evolution of the Ising machine versus gradient-descent step for the 3-SAT problem in (d). In the upper panel, the soliton spins bifurcate to 2 states. In the lower panel, the number of unsatisfied clauses \sum C_{E} and the Hamiltonian H evolve to the global minimum. Kerr-microresonator solitons are discrete and localized electromagnetic excitations of an intraresonator field, exhibiting dissipation, gain, and the balance of nonlinearity and phase matching [ 17 , 18 , 19 ] . Such solitons exhibit modelocking and they form an outcoupled, ultrafast optical pulse train from the resonator that we associate with an optical frequency comb. Hence, we refer to this system as a soliton microcomb, and it exhibits extraordinary optical and microwave coherence with fractional optical frequency accuracy and precision reaching 10^{-20} [ 20 ] , which supports applications ranging from optical frequency metrology [ 20 ] to optical data interconnects to universal electronic synthesis [ 21 ] . From the earliest experiments with Kerr solitons, their discreteness was recognized, and ensembles of solitons feature a high degree of indistinguishability and long-range interactions [ 5 ] . Moreover, soliton microcombs can be implemented with increasingly advanced photonic integration technologies [ 22 ] and foundry manufacturing [ 23 ] . Here, we introduce a Kerr-soliton Ising machine and use it to solve combinatorial optimization problems. We implement the Ising machine with a large ensemble of Kerr solitons in a resonator, and we map fully programmable all-to-all interactions amongst the ensemble with an opto-electronic feedback network. The emergent collective state of the discrete solitons represents the spins of the Ising model, and we program the hybrid opto-electronic feedback network to directly couple SAT instances, multiplexing the solitons in time and amongst one electrical path for each optimization problem mapped to the Ising model. The soliton spin bifurcation arises directly from Kerr nonlinearity because there are only two, distinct, quantum-based eigenstates of the resonator [ 24 ] , the empty state and the soliton state. To evolve the Ising machine, we photodetect the soliton spin states and use stepped gradient descent to update the feedback circuit. Each step fundamentally requires 3.5 \mu s of time in accord with the soliton nonlinear dynamics. Moreover, each soliton only consumes around 2.3 mW of pump power in the present apparatus. Hence, we can solve complex, arbitrary SAT problems more rapidly and with two orders of magnitude less optical energy than benchmark digital computer algorithms. Furthermore, the Kerr soliton Ising machine shows low latency, rapid execution, energy efficiency, and good stability and scalability. The concept of our Kerr soliton Ising machine is shown in Fig. 1 (a), including a high finesse Kerr resonator (circle) to generate solitons and a low finesse

Read the complete text