Something that gets lost in all the noise and hype associated with quantum computing is the fact that innovations in classical computing keep on coming. As we indicated in our previous articles, Innovations in Classical Computing Architectures Make Quantum Advantage a Moving Target and Quantum Computing is Already Creating Commercial Value – By Creating Competition, advances continue in classical computing algorithms and architectures that increase performance and capabilities and raise the bar for quantum computing researchers working to develop something better.
We’ve seen two recent examples that show how innovations in classical computing algorithms may be able provide competitive solutions for hard combinatorial optimization problems. These types of problems are common in industrial settings and have applications in finance, logistics, forecasting, routing, machine learning and other problems that can be expressed as an optimization.
The first is a recent announcement from Toshiba called a Simulated Bifurcation Machine (SBM). This algorithm derives from an observation made by a Toshiba scientist that the qualities of some complex systems can suddenly change with additional input, a phenomenon they describe as bifurcation. This property makes the problem much more amenable for solution on arrays of parallel computing as it allows for parallel updates of variables and can provide much better performance that other well-known classical algorithms such as simulated annealing. For more details about Toshiba’s SBM, you can view their web page that describes it here and a more detailed technical article published in Science Advances here.
The second development is from a company in San Diego, California called MemComputing Inc. Using research originally developed at the University of California, San Diego they have a technology they call the MemCPU™ Coprocessor technology. A key element in their approach is a device called a Self Organizing Logic Gate. These gates are arranged in a network and each one can accept an input from any terminal. All the gates then work together collectively to achieve an optimal state. MemComputing has published several case studies and white papers on the MemComputing web site and is currently offering their technology on the cloud as SaaS (Software as a Service). They indicate that even with their current approach of simulating their technology using CPU or GPU resources in a classical cloud data center they are already achieving very good performance. They are also planning in the future to develop an ASIC that will implement their algorithm in hardware and further improve their performance by several orders of magnitude. One notable paper posted on arXiv titled Stress-testing memcomputing on hard combinatorial optimization problems describes how they were able to handle a problem with 64×106 variables using a computer with 128GB of memory.
So for those of you who believe that Moore’s Law is starting to end and will cause continued improvements in classical computing to end, we suggest you think again. Much of classical computing is still based upon the von Neumann architecture developed in 1945 and a lot of innovation is still possible as researchers explore alternate new ways of processing classical data.
Although many of these new innovations may not handle as wide a range of problems as a large scale error corrected quantum computer would, these large scale error corrected quantum computers are still far in the future. For any particular computational problems you may have, we do recommend testing of multiple alternatives, both classical and quantum, to see which one is best. The best approach may differ significantly depending upon the specific problem, so it is best not to make any assumptions without first trying it out.
January 24, 2020