Dr. Ruti Ben‑Shlomi from the Weizmann Institute announced that LightSolver’s Laser Processing Unit (LPU) can now directly map partial differential equations onto a 2‑dimensional grid of coupled lasers, enabling the device to solve canonical problems such as the heat equation and Poisson equation in a single physical iteration that takes only a few nanoseconds – a speed that is up to one hundred times faster than contemporary GPU‑based solvers. The new architecture eliminates the need for external memory and statistical sampling, allowing the LPU to scale to 100,000 variables by 2027 and one million by 2029, thereby positioning the technology as a potential accelerator for scientific discovery, engineering design and computational physics. Dr. Chene Tradonsky, also a physicist from the Weizmann Institute, joined Ben‑Shlomi in founding LightSolver in 2020 and has been instrumental in translating the LPU’s optical computing paradigm into a commercially viable platform.

LightSolver Maps Partial Differential Equations onto Laser Processing Unit Accelerating Scientific Discovery

LightSolver’s latest announcement details a pivotal advance: the Laser Processing Unit (LPU) can now map entire families of partial differential equations (PDEs) directly onto its 2‑dimensional lattice of coupled lasers. By configuring the optical network to emulate the spatial discretisation of the heat equation, Poisson’s equation, and, in preliminary tests, wave‑propagation and Schrödinger dynamics, the company has effectively created a laser PDE solver that operates in physical time rather than in a digital emulation. The mapping is achieved through a proprietary compiler that translates a user’s PDE specification into a set of laser‑control instructions, obviating the need for specialised knowledge of laser physics.

The optical architecture exploits the fact that each laser in the grid behaves as an electromagnetic wave that can interfere with its neighbours. This interference pattern encodes the sparse coupling matrix that underpins most PDE discretisations. Because the computation is performed entirely in the optical domain, each iteration step is executed in a few nanoseconds, a duration that remains constant regardless of problem size. In benchmark tests, the LPU has delivered speed gains of up to one hundred times over contemporary GPU‑based solvers, while consuming far less power and eliminating the memory‑bandwidth bottlenecks that plague digital approaches. Moreover, the system requires no external runtime memory or repeated statistical sampling; the physical evolution of the laser field itself constitutes a single, deterministic trial that converges to the PDE solution.

Looking ahead, LightSolver has outlined a clear scaling trajectory. The current prototype already handles tens of thousands of variables, and the company projects a 100‑kilovariable capacity by 2027, expanding to one million variables by 2029. To support researchers and industry partners, LightSolver will soon open its LPU Lab, which houses an alpha version of the hardware alongside a digital emulator that mimics the optical behaviour. This dual‑platform offering will allow users to prototype and validate their PDE models before deploying them on the full‑scale processor.

The breakthrough has been disseminated through both academic and industrial channels. At ACM Computing Frontiers 2025, LightSolver’s team presented a detailed account of the technology, and the findings were subsequently published in a peer‑reviewed article titled “Solving Partial Differential Equations on an Analog, Optical Platform.” The company has also secured a partnership with Ansys and is collaborating with high‑performance computing centres and national laboratories, positioning the laser PDE solver for immediate real‑world impact in fields ranging from fluid dynamics to structural mechanics.

LightSolver was founded in 2020 by Dr. Ruti Ben‑Shlomi and Dr. Chene Tradonsky, both physicists from the Weizmann Institute of Science. Their vision of an all‑optical supercomputer has attracted investment from TAL Ventures, Entree Capital, IBI Tech Fund, Angular Ventures, Maverick, and Artofin, as well as a €12.5 million grant from the European Innovation Council. With these resources, the company is poised to deliver a room‑temperature, rack‑unit‑sized processor that harnesses the speed of light to accelerate scientific discovery.

Laser Based Analog Computing Eliminates Data Transfer Bottlenecks Delivering Nanosecond Iterations

On 16 September 2024, LightSolver announced that its Laser Processing Unit can now encode partial differential equations directly onto a two‑dimensional lattice of coupled lasers, thereby transforming the device into a laser PDE solver capable of handling the heat, Poisson, wave, and Schrödinger equations. The optical design stores and processes information within the laser field itself, eliminating the need for external memory transfers and thereby avoiding the bandwidth limitations that constrain conventional digital solvers.

Because each computational step is executed entirely in the optical domain, the LPU achieves iteration times of only a few nanoseconds, a duration that remains constant regardless of the number of variables involved. The physical evolution of the laser field represents a single deterministic trial, removing the requirement for repeated statistical sampling that is typical of quantum approaches and reducing overall power consumption compared with GPU‑based solvers.

A compiler supplied with the LPU translates formal PDE expressions into laser control instructions, enabling users without specialised knowledge of laser physics to deploy their models directly on the hardware. LightSolver plans to open its LPU Lab, where researchers will have access to an alpha version of the hardware alongside a digital emulator, and the company has set a roadmap to support 100,000 variables by 2027 and one million by 2029.

This breakthrough positions LightSolver’s laser PDE solver as a compelling alternative to electronic and quantum solvers, offering rapid, scalable, and energy‑efficient solutions for complex physical simulations.

100K Variable Roadmap by 2027 Positions LightSolver at Forefront of Optical Supercomputing

LightSolver, headed by CEO Ruti Ben‑Shlomi, has set a concrete target for its Laser Processing Unit (LPU): the ability to process 100 000 variables by 2027. The milestone, announced from Tel Aviv on 16 September 2024, signals the company’s intention to become a leading force in optical supercomputing, a domain that has until now been dominated by electronic and quantum platforms. By achieving this scale, LightSolver will be able to tackle problems that exceed the reach of conventional high‑performance computing systems.

The breakthrough rests on a novel mapping of partial differential equations onto a two‑dimensional array of coupled lasers. In this scheme, the equations are encoded directly into the laser field, allowing the LPU to execute each iteration in a few nanoseconds regardless of the number of variables. The constant‑time behaviour translates into speed gains of up to one hundredfold over contemporary GPU‑based solvers, while the absence of external memory transfers removes the bandwidth bottlenecks that plague digital architectures.

Each convergence produced by the LPU represents a deterministic physical trial. This property eliminates the need for repeated statistical sampling that is typical of quantum approaches and reduces overall power consumption relative to GPU‑based solutions. The result is a laser PDE solver that delivers energy‑efficient, scalable solutions for a wide range of scientific and engineering problems, from heat transfer to quantum wave dynamics.

Designed to fit within a single rack unit and to operate at ambient temperature, the LPU is compatible with existing data‑centre infrastructure. Its compact, room‑temperature design removes the cryogenic and space constraints that often limit the deployment of high‑performance computing hardware, making it a practical addition to current facilities.

Looking ahead, LightSolver plans to open an LPU Lab where researchers will gain access to an alpha prototype and a digital emulator. The company has already secured a partnership with Ansys and collaborations with high‑performance computing centres and national laboratories, paving the way for real‑world deployment. By 2029, LightSolver aims to support one million variables, further expanding its capacity to address large‑scale physics simulations and cementing its role as a transformative platform for optical supercomputing.

Partnership with Ansys and National Labs Signals Commercial Readiness for Engineering Design

LightSolver’s announcement on 16 September 2024 of a formal partnership with Ansys, the global leader in engineering simulation software, marks the first concrete step toward bringing its laser PDE solver into mainstream engineering design workflows. The collaboration brings together LightSolver’s all‑optical Laser Processing Unit (LPU) and Ansys’ extensive suite of finite‑element and computational fluid‑dynamics tools, creating a pathway for designers to embed optical supercomputing directly into their simulation pipelines.

The alliance is significant because Ansys’ software is the de‑facto standard in automotive, aerospace, civil engineering and energy sectors. By integrating LightSolver’s laser PDE solver, which can map equations such as the heat equation, Poisson equation, wave dynamics and Schrödinger’s equation onto a two‑dimensional lattice of coupled lasers, Ansys customers will gain access to a solver that delivers constant‑time iteration steps on the order of nanoseconds. This capability promises up to a hundred‑fold speed improvement over contemporary GPU‑based solvers, while simultaneously reducing energy consumption through the elimination of external memory transfers and statistical sampling.

In addition to the Ansys partnership, LightSolver has secured active collaborations with high‑performance computing centres and national laboratories across several countries. These relationships provide a rigorous testing ground for the LPU, exposing it to large‑scale, real‑world engineering problems that require thousands to millions of variables. The labs’ involvement also serves as a validation of the LPU’s reliability and scalability, reinforcing confidence among potential commercial adopters.

The technical foundation of the laser PDE solver—its ability to encode sparse systems directly into the phase and amplitude of laser fields—means that each convergence represents a deterministic physical trial. This deterministic nature eliminates the need for repeated statistical sampling that is typical of quantum approaches, and it aligns neatly with the deterministic expectations of engineering simulation workflows. The partnership will therefore enable engineers to perform rapid, repeatable analyses of heat transfer, structural mechanics, electromagnetism and wave propagation without the overhead of conventional digital architectures.

Looking ahead, the collaboration with Ansys and the national‑lab network is expected to accelerate LightSolver’s roadmap to 100 000 variables by 2027 and one million variables by 2029. By embedding the laser PDE solver within Ansys’ commercial platform, LightSolver positions itself to deliver a fully commercial, rack‑unit‑sized optical supercomputer that can be deployed in existing data‑centre infrastructure. The forthcoming LPU Lab, slated for launch in the coming weeks, will provide industry partners with early access to both the alpha hardware and a digital emulator, further smoothing the transition from prototype to production.

Peer Reviewed Validation at ACM Computing Frontiers 2025 Confirms Physical Modeling Advantage over Quantum and GPU Solvers

LightSolver’s laser PDE solver was formally validated at the ACM Computing Frontiers 2025 conference, where a team of LightSolver researchers presented a peer‑reviewed study titled Solving Partial Differential Equations on an Analog, Optical Platform. The paper demonstrates that the Laser Processing Unit (LPU) can encode and solve canonical PDEs—such as the heat equation, Poisson equation, wave dynamics and Schrödinger’s equation—directly onto a two‑dimensional lattice of coupled lasers. In contrast to conventional digital approaches, the LPU achieves iteration steps that complete in a few nanoseconds, regardless of the number of variables, yielding speed improvements of up to one hundred times over contemporary GPU‑based solvers.

The study details the physical‑modeling methodology that underpins the LPU’s performance. By mapping the coefficients of a sparse PDE system into the phase and amplitude of the laser fields, the device eliminates the need for external memory transfers and the statistical sampling that characterises quantum solvers. Each convergence of the optical network represents a deterministic physical trial, so the solver requires only a single run to reach a solution. The paper contrasts this with quantum computers, which, even with 1,000 logical qubits, are limited by input/output bottlenecks, memory bandwidth constraints and the necessity of repeated statistical experiments to achieve reliable results.

The peer‑reviewed findings reinforce LightSolver’s roadmap to commercial‑grade optical supercomputing. The authors report that the LPU can scale to 100,000 variables by 2027 and one million variables by 2029, a trajectory that aligns with the company’s announced target for a rack‑unit‑sized optical processor. By providing independent, scholarly confirmation of the laser PDE solver’s physical‑modeling advantage, the ACM Computing Frontiers 2025 paper strengthens the case for deploying LightSolver’s technology in high‑performance computing environments and accelerates its transition from prototype to production.