One thing to understand is that quantum computing may not be the only game in town when one is looking for faster alternatives to classical computers. Recently, two companies have made introductions within the past week of physics based computational devices targeted at specific applications they they contend will be even more effect at solving problems than either classical or quantum computing.

The first is a company based in Israel called LightSolver that has just introduced its LPU100 system that uses 100 lasers to solve tough optimization problems. They are now providing early access to a cloud based system accessible via a Python software package that can solve NP-hard optimization problems with as many as one million variables. The system operates at room temperature and is the size of a traditional desktop computer. It can process operations such as a vector matrix multiplication as fast as 10 nanoseconds, which would be much faster than could be performed classical. The technology is based upon translating the problem variables into laser bits which have a relative phase. The computation is performed by the interactions between the lasers.

The company believes their device has many potential applications in the areas of logistics, manufacturing, aerospace, and finance. The company recent published a benchmark for a hypothetical problem of field service technician scheduling that compared their solution versus a classical implementation using Gurobi, one of the leading classical optimization solvers in widespread use today and reported a significant improvement with their solution as shown in the chart below.

The company was founded by physicists from the Weizmann Institute in 2020 and its investors include TAL Ventures, Entree Capital, IBI Tech Fund, and Angular Ventures. A press release announcing this new LPU100 system can be seen here.

The second company is called Extropic and uses a completely different principle of physics and completely different applications. They are taking advantage of thermodynamics and matter’s natural fluctuations to solve problems in the Generative AI domain. The thermodynamic approach is particularly useful for processing probabilistic AI algorithms and the company believes their approach is orders of magnitude faster than anything using digital logic. They are developing a full-stack platform that is based upon chips. Their initial designs are being based upon superconducting neuron designs, but they have plans to also implement their approach using room temperature semiconductors. A major drawback of today’s AI supercomputers is the massive amount of power that is needed for the data centers. So besides a potential advantage in performance, an alternative approach like there could provide a huge benefit in terms of power consumption and operating cost.

Early Superconducting Based Prototype Device. Credit: Extropic

The Extropic team has a wide range of experience AI and physics from several well-known companies including Alphabet X, Google Quantum AI, AWS, IBM, Meta, NVIDIA, Xanadu, and leading academic institutions. They closed a $14.1 million seed round from a number of venture capital firms and angel investors. To learn more about Extropic’s technical approach, you can read a “Litepaper” posted on their website here.

March 19, 2024