With the rise of AI, energy and heat efficiency have once again become pressing concerns for companies that use and build chips. The skyrocketing demand for hardware to run AI models is dragging up energy bills, as these servers require vast numbers of chips and enormous cooling setups.
Vaire Computing, based in London and Seattle, is betting that reversible computing will be the way forward. It has now raised $4 million in a seed round to work on building silicon chips that would consume negligible amounts of energy and generate little heat, if any. The round was led by deep-tech fund 7percent Ventures and Jude Gomila, the co-founder of Heyzap. The company had previously raised $500,000, so this round brings its total funding to $4.5 million.
In reversible computing, instead of running a calculation in only one direction (inputs followed by outputs) and then feeding the output to a new calculation and running it again, the computing can be done in both directions (known as “time-reversible” computing). Effectively, energy is retained inside the chip instead of being released as heat. The theory is that this method would generate negligible amounts of heat, vastly reducing energy consumption. (A better explanation of its potential lies in this essay by Azeem Azhar and David Galbraith.)
Vaire Computing was founded by serial entrepreneur Rodolfo Rosini and Hannah Earley, a researcher at the University of Cambridge who works on “unconventional computing” such as reversible and molecular computing.
Over a call, Rosini told me: “Close to 100% of the energy in a chip ends up being dissipated as heat. So you’re basically wasting it. But in a reversible chip, you actually never dissipate this energy. You don’t allow the energy to become heat, and you recycle it internally. This means that two things happen: One, the chip doesn’t get hot, and two, you only need a tiny amount of energy to make it work. So, it uses almost no energy, other than the same amount of energy that it has just recycled.”
The concept of reversible computing is not new, and there are a lot of challenges before Vaire’s chips can become a reality, but Rosini thinks the shift to this new approach to computing would not be too dissimilar from how we switched from filament bulbs to LEDs. “The similarity is between an old light bulb based on incandescent filaments and LEDs,” he said. “LEDs are colder and more efficient, and there’s a cluster of them… This is virtually identical to reversible computing. You don’t have a single core that is super fast, you have a lot of smaller cores where each one is super efficient.”
He says a big advantage of chips that can do reversible computing would be their ability to be used on generic applications, just as normal CPUs are used today. “Other kinds of chips are domain-specific, but with computing, you can do anything… We could also build a CPU or GPU, and it would look like any other chip.”
When asked why the funding in the space is so low if the tech is as revolutionary as it sounds, Rosini said: “Because the amount of money that went into reversible computing and alternative chip architecture is almost nothing,” he said, pointing to the billions spent on quantum computing, photonics and GPUs.
“If you go outside these well-trodden areas and talk about building a brand new architecture, there’s absolutely nobody who will fund it. Secondly, we don’t really need a lot of money to make the first chip and prove the technology… Once we prove that, we’ll need a much larger round to actually build a chip,” he added.
For her part, Earley believes reversible computing could be used to make the most powerful computers. “I got involved in this area during my PhD in 2016,” she said. “Serendipitously, my PhD supervisor sent me the thesis of my friend who was at the University of Florida group that was looking into reversible computing. It got me interested in how I could apply it to my research field at the time, which was molecular programming. I started to think that reversible computing is interesting in its own right, particularly as it could make the most powerful form of computers possible. After completing my PhD, I was introduced to Rudolfo and we realized that we had the same vision.”
“Vaire Computing is different because its technology is innovative at a foundational level, positioning the company extraordinarily well to capture a huge chunk of the future AI chip, and ultimately, computer processor market,” Andrew J Scott, founding partner at 7percent Ventures, said in a statement.
The round also saw participation from Seedcamp, Clim8, Tom Knight (an inventor of modern reversible computing) and Jared Kopf, founder of Ramble.ai.
Additionally, Vaire has hired Mike Frank, a noted researcher in reversible computing, as the company’s senior scientist.
Vaire recently became one of only 10 companies named to the second U.K. cohort of Intel Ignite, Intel’s global startup accelerator program for early-stage deep tech startups.
Comment