A Closer Look at LightSolver’s Laser-based Processing Unit (LPU)
Compared to supercomputers and quantum computers, LightSolver's all-laser processing unit demonstrated exceptional potential during its initial tests.
Emerging computing technologies continue to push the boundaries of speed, efficiency, and problem-solving abilities. Among these advancements, laser-based processing units (LPUs) have emerged as promising candidates for accelerating computation in optimization and simulation tasks. By harnessing the unique properties of light, LPUs offer an alternative to traditional processors, paving the way for innovations in various industries.
With their optical architecture, LPUs tap into coherence, interference, and parallelism to perform computations, allowing them to process various computational elements simultaneously. As a result, LPUs have the potential for remarkable acceleration and efficiency.
LPU startup LightSolver was founded by Dr. Chene Tradonsky (left) and Dr. Ruti Ben Shlomi (right), Ph.D. physicists from the Weizman Institute. Image courtesy of LightSolver
In this article, we delve into the development of the so-called first pure laser-based processing unit (LPU) by Israeli startup LightSolver. We also share insights from our interview with Chene Tradonsky, CTO and co-founder at LightSolver. The company believes its processors hold immense potential for tackling complex optimization problems across diverse domains, from logistics and finance to energy and manufacturing.
Tradonsky says that his doctoral on coupled laser arrays inspired the LightSolver’s optical system, which uses similar physical principles. But there’s more to it than that. “LightSolver also draws on other paradigms of optical computing and advanced mathematical abstractions,” he says. “This enables us to apply our optical device to real world problems and solve them."
Operating Principles of Laser-based Processing Units
At the core of LPUs is their optical architecture, consisting of components such as lasers, beam splitters, modulators, and photodetectors. Together, these components manipulate and control laser beams for computational purposes. They also rely on optical interconnects (high-speed optical channels) to facilitate efficient communication and data transfer within the processing unit.
LPUs leverage various properties of lasers for efficiency and high performance. One of those properties is coherence, the property of light waves in synchronization. The lasers used in LPUs demonstrate coherence, which enables them to perform multiple operations simultaneously. Another crucial property is interference, which occurs when lights interact. By carefully controlling the interference patterns, LPUs can perform computations efficiently.
LightSolver's LPU. Image courtesy of LightSolver
Photonic memory is another essential part of the LPU. It provides high-speed access to information. With their fast and reliable data retrieval, LPUs can quickly access and manipulate large datasets.
In most cases, LPUs draw inspiration from quantum computing techniques, such as quantum annealing and the Ising model. Quantum annealing involves gradually transitioning a physical system, represented by qubits, to a low-energy state that corresponds to the optimal solution of a problem. LPUs conduct the same process with optical components to efficiently search for near-optimal results among a vast set of possibilities.
LightSolver's All-laser Processing Unit
LightSolver recently introduced a pure LPU—claimed to outperform classical computers as well as quantum and supercomputers. The company's quantum-inspired solution uses all-optical coupled lasers and does not require any electronics for computation. This solution is specifically designed for businesses that require solutions for complex multivariable challenges.
The new laser-based processing approach. Image courtesy of LightSolver
Solving complex optimization problems is quite challenging and requires significant computational power. Although supercomputers and quantum computers have historically been the preferred choice for these types of applications, supercomputers are reaching performance limits, and quantum computers are not scalable and practical.
LightSolver's LPU works by first converting a problem into a physical logic, which is then mapped as obstacles within the optical path. Due to the properties of lasers, like coherence and interference, the beams converge in a desired solution. After finding and measuring the optimized solution, it is translated into a suitable language for the user.
Interestingly, Tradonsky says that how he positions the working principles of the LPU depends on who the expert is using the technology. “For complex systems experts, we can generate an array of coupled oscillators with any type of connectivity to simulate the behavior of any other complex systems and find optimal configurations,” he says.
“For lasers and optics experts, LightSolver uses what we call “laser bits,” turning optimization problem constraints into the lasers' relative phases, which interact by diffracting light from each laser to all others in a controllable manner,” says Tradonsky. “This means we can generate a programmable large array of coupled lasers fully connected via integrated electro optical elements.”
How LightSolver's LPUs Performed in Early Trials
In three recent trials testing the performance and accuracy of the new LPU, the device showed promise against its supercomputing and quantum computing counterparts in the following ways:
- Achieved 2X–1,000X faster time-to-solution (TTS) for solving Max-2-SAT problems in a head-to-head challenge with a deep learning solver
- Solved the "3-Regular 3-XORSAT Challenge" in polynomial time, while other classical and quantum computers solved it in exponential time
- Developed a quantum-inspired algorithm for sparse coding that outperformed classical approximation methods in accuracy
According to Tradonsky, the Tel-Aviv-based company's technology can solve optimization problems by converting business challenges into mathematical formulations, such as Ising models. However, it is not limited to binary models and can implement other models. He adds that, unlike quantum technology, LightSolver's device is portable, operates at room temperature, and is not affected by environmental factors or error correction protocols. Scaling is also a big differentiator.
”The scaling ability of the LPU is far superior to alternatives. Unlike in quantum computers, where each logical qubit is typically represented by several physical qubits, the LPU represents each variable with a single laser spin."
”This characteristic of the LPU facilitates scalability, as the number of variables can be increased without the need for a proportional increase in physical resources,” says Tradonsky. “When it comes to supercomputers, solutions are poor in quality, and the time-to-solution increases exponentially with the problem size. In contrast, thanks to LightSolver’s use of continuous laser technology, optimization problems can be solved orders of magnitude faster than other current techniques.”
"The LPU is highly specialized and can perform significantly better on a specific class of problems known as QUBO (Quadratic Unconstrained Binary Optimization) problems. The LPU can also operate as a complementary solution to classical and quantum computers positioned before (pre-processing) or after (post-processing) these computers to enhance their performance and maximize efficiency."