VyperCore: ‘Our chip technology has the potential to turn the industry on its head’

VyperCore, a fabless processor seed-stage start-up based at 9 Hills Road, is looking to double its headcount in Cambridge and Bristol.

The incoming software and hardware engineers will be developing a 5nm chip and card to accelerate server-class applications by a factor of five.

The ingenuity of the technology is to shift the complexity of memory allocation management from software to hardware. Without changing a line of code, the technology eliminates memory safety issues such as buffer overflows and use-after-free (a memory corruption bug). It achieves this by replacing software-based memory allocation management – a common feature of all managed-language runtimes – with dedicated hardware inside the CPU, which eliminates up to 80 per cent of the processor cycles typically required for handling memory allocation.

VyperCore was founded in 2022 by Russell Haggar and Ed Nutting Russell has been directly involved with more than 30 tech start-ups, and now helps the UK government deliver its national semiconductor strategy. Ed originated VyperCore’s technology in his nine years of study and research at the University of Bristol. Last year the company raised £4m to develop its first generation of accelerated compute silicon.

Currently a team of 17 in Cambridge and Bristol, VyperCore uses extended RISC-V processor architecture to address the most prominent memory safety issues, which make up around 70 per cent of all technical security vulnerabilities according to reports by Microsoft, Google, Apple and others.

“Our technology can make a substantial difference to secure existing and new software, without having to rewrite all the world’s existing software and with a performance increase,” explains Russell.

“The technology we’ve developed lets you take existing software, not change a single line of code, move it to our software and it’ll run five times faster,” Russell adds in a video call. “We allow customers to redeploy their code.

“When writing in Java or Python a vast amount of processing is wasted. Up to 80 per cent of what a CPU is doing is internal; it’s supporting the task.”

The first test chip that Russell and his team are working on is expected to be ready next year.

“It’s a processor, so it’s got cores,” Russell explains. “It’ll be called the Booth Core.”

Any particular reason, I ask Russell?

“It’s from Katherine Booth, who invented the world’s first programmable computer.”

Prof Booth (1922-2002) was a British computer scientist and mathematician who wrote the first assembly language and designed the assembler and autocode for the first computer systems at Birkbeck College, University of London.

Russell adds: “She recently passed away in Canada aged 100. She was married to Arnold Booth, who also worked on computers, while she wrote code. She did her work in Welwyn Garden City – I live in nearby Welwyn. Then she retired to raise her children. Given how poorly women are represented in this industry it seemed an opportunity to make something of it.

“The Booth Core has 100-plus cores, it redeploys software from the current CPU and motherboard into our CPU.

“The Booth chip is the main part of the accelerator card and is where all the computation and acceleration will take place. The accelerator card itself is how we are able to be placed into the server environment via a common connectivity interface.

While initially the new technology is geared for data servers, in the longer term “it could be inside every CPU from a toaster to a server”.

The new-style processor also has implications for generative AI – and Arm and Nvidia are likely to be watching its progress closely.

“So where there’s applications running flat out all the time, such as Zoom who probably have a large number of servers, with their customers probably going through a cloud data centre with thousands of servers at peak times – it’s five times less power usage.

“It works for operations in the cloud such as Amazon or Google, and also private data centres such as those operated by banks – it’s for big large-scale industrial server farms.

“Our customer is the person who owns the data centre, or rents it out to a third party. We put in a new card: there’s a card for each server, so if a client owns 10,000 data centres, that’s 10,000 cards, and the cards are then owned by the data centre.”

The venture is likely to require further funding after the initial £4m tranche has run its course.

“We’re having conversations across the industry,” Russell says. “We’re talking with potential customers for when the demo goes live in December, and other processor companies are interested. Our technology works inside 3Arm, Intel’s X86 and RISC-V – equally well in all of them.

“The money we raised is to last two years. The £4m was pre-seed: the next round is a late seed round. We’re out now raising finance for the next round which will last until the chip is tested. There’s a lot of interest and investors are very positive. This is a genuine game-changer which has the possibility to turn the industry on its head.

“One interested party said this is geopolitically significant. The technology itself has the potential to apply to the entire scope of the industry from a washing machine to a laptop, but we can’t go after all those at once so, as a go-to-market, data servers makes sense to us.

“But this technology should end up in every general-purpose processor on the planet and it has the potential to achieve that.”