MemComputing is a new computing architecture invented by physicists Dr. Fabio Traversa and Dr. Massimiliano Di Ventra at UC San Diego. Mr. John Beane was introduced to the technology while serving as an Entrepreneur in Residence at the University. He was so impressed with the technology that he worked with Dr. Traversa and Dr. Di Ventra to spin out and form a company (of the same name) that would advance and commercialize the technology. MemComputing, at its core, is a new circuit design that overcomes the performance bottlenecks faced by today’s computers, enabling ultra-fast processing for complex problems. The first product is the cloud-based Virtual MemComputing Machine (VMM) that solves intractable optimization problems faced by Industry and the military. These optimization problems, often found in transportation logistics applications, are extremely complex and take years for current supercomputers to solve at scale. The VMM solves these problems in minutes and has demonstrated $10s of millions in annual savings for each problem set. Essentially, MemComputing provides the power and performance expected of Quantum Computing but is non-quantum and realized in silicon using traditional transistors.
Below is our recent interview with John Beane, the Founder & CEO at MemComputing:
Q: Who is your ideal customer and why?
A: Our earliest adopters come from Transportation Logistics, Oil & Gas, the Air Force, and Space Force. The most common problems we address are related to routing and scheduling applications. Although virtually every company in the Global 2000 is at a scale where they face intractable optimization problems. Some other sectors and related problems include companies in Fin Tech (Portfolio Optimization), Pharmaceutical (Drug Discovery), Genomics (DNA Sequencing), and Smart Cities (Traffic Management). While most of our customers come from the Global 2000, we’ve worked with smaller companies in Transportation Logistics that may only have fleets of 25 – 500 vehicles (e.g., trucks, helicopters, drones, and even supertankers). Generating optimal schedules for these fleets for daily last-mile delivery or predicting schedules over 30 days, 90 days, or even years are already intractable for current computers. These companies are solving these problems today using manual or heuristic approaches, which render approximate, sub-optimal solutions. By providing an optimal solution, we help these companies achieve new levels of operational efficiencies and avoid costs in the $10s of millions annually. This goes directly to their bottom lines.
Q: How about consumers? Will we be using MemComputers in the near future?
A: Unfortunately, MemComputing isn’t going to make your spreadsheets or browsing faster. However, there is a big market for us in gaming. We recently solved an intractable problem for managing a swarm of autonomous drones. The problem is known as Multi-Agent Path Finding or MAPF. As it turns out, multi-player games, especially games where live players are pitted against automated players, which might be zombies or vehicles, continue to struggle with the MAPF problem. We’ve reached out to some game makers and hope to begin working with them this year. The MAPF problem is intractable and one that we solve optimally at scale.
Q: Can you give us more insights into products and services?
A: The cloud-based Virtual MemComputing Machine (VMM) is a general-purpose application builder for intractable optimization problems. It is “programmed” using mathematical formulations following the specification for Integer Linear Programming (ILP), a common method used by industry. Companies can subscribe to the service and run their problems directly through our website. However, we also provide professional services. We will do everything from analyzing your ILP, teach your company how to build good mathematical models that take the most advantage of the VMM, and building the mathematical models and applications for you. Most companies start by having us do a Proof-of-Concept (PoC). This is where they essentially say to us, “you claim you can leap a tall building in a single bound; here is our tall building.” A typical PoC takes 2-3 months, and in the end, we demonstrate how quickly the VMM can generate an optimal solution to your problem and the expected annual cost savings or cost avoidance.
Q: What can we expect from MemComputing in next 12 months? What are your plans?
A: While our cloud-based Virtual MemComputing Machine serves many, in 2023, we will begin developing MemComputing-based Application Specific Integrated Circuits (ASIC), or computer chips. Our ASICs are ultra-high-performance while also being ultra-low-power. For instance, our patent pending MemGPU™ will be at least 1,000 times faster than today’s fastest GPU but will require a fraction of the energy. MemCPU™ and MemGPU ASICs will be ideal for assisted and autonomous driving and motion planning.
Q: What is the best thing about MemComputing that people might not know about?
A: The biggest thing people don’t know about MemComputing, is that it exists! The technology also sounds too good to be true when you first hear about it. However, our website has links to all of our scientific publications. Eighteen of these publications have gone through peer review by experts who validate that the science is valid and original. You have to be a Ph.D. to appreciate these publications. For the mere mortals among us, we also publish Case Studies of actual intractable problems solved using MemComputing. One will find that we are very transparent about our technology.