Fractile raises $15m for AI in-memory compute
Cette publication existe aussi en Français
Chip startup Fractile has raised $15m for its in-memory AI compute that it says can run the latest AI models at least 100x faster and 10x cheaper.
The funding round was co-led by Kindred Capital, NATO Innovation Fund, and Oxford Science Enterprises, alongside leading angel investors including Herman Hauser of ASM, Stan Boland of Icera and Five.ai and Amar Shah of Wayve.
Founded in 2022 in London.UK, the company had previously raised $2,5m before emerging from stealth mode and has recruited staff from ARM, Imagination Technologies and Nvidia.
It has filed patents protecting key circuits and its approach to in-memory compute and is in discussions with potential partners and expects to sign partnerships ahead of production of its first commercial AI accelerator hardware.
NATO’s first equity deal backs UK startups Fractile, Space Forge
“In today’s AI race, the limitations of existing hardware – nearly all of which is provided by a single company – represent the biggest barrier to better performance, reduced cost, and wider adoption. This is more than just a speed-up – changing the performance point for inference allows us to explore completely new ways to use today’s leading AI models to solve the world’s most complex problems,” said Dr. Walter Goodwin, CEO and Founder of Fractile
“There’s no question that, in Fractile, Walter is building one of the world’s future superstar companies. He’s a brilliant AI practitioner but he’s also listening intently to the market so he can be certain of building truly compelling products that other experts will want to use at scale. To achieve this, he’s already starting to build one of the world’s best teams of semiconductor, software and tools experts with track records of flawless execution. I’ve no doubt Fractile will become the most trusted partner to major AI model providers in short order,” said Stan Boland, angel investor.
There are two paths available to a company attempting to build better hardware for AI inference, says Fractile. The first is specialisation: honing in on very specific workloads and building chips that are uniquely suited to those specific requirements. Because model architectures evolve rapidly in the world of AI, whilst designing, verifying, fabricating and testing chips takes considerable time, companies pursuing this approach face the problem of shooting for a moving target whose exact direction is uncertain.
Instead Fractile is using in-memory compute and is targeting 20x the TOPS/W of any other system available today for data centre AI inference. This allows for more users to be served in parallel per inference system, with – in the case of LLMs for example – more words per second returned to those users, thereby making it possible to serve many more users for the same cost.
Currently, to get output from the largest models that matches human reading speed, AI companies tend to deploy systems which use purely ‘next token prediction’. Faster speeds would allow cost-effective recursive queries, chain of thought prompting and tree search to improve the quality of answers.
The additional performance can also accelerate AI’s ability to solve scientific and computationally heavy problems, from drug discovery to climate modelling to video generation.
Other companies such as Axelera and Femtosense and even GraphCore have developed in-memory compute architectures for embedded and edge AI applications.
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
