Fractile received a £5M award from the UK's Advanced Research + Invention Agency

The Advanced Research + Invention Agency (ARIA), a research agency sponsored by the UK's Department for Science, Innovation, and Technology, was created by a Parliament Act to take on bold projects that promise long-term technological and scientific projects. The agency does this by supporting different research programs, one of them being ARIA's Scaling Compute program, which aims to support projects with the potential to redefine the current computing paradigm.

Citing the potential economic, geopolitical, and societal implications of the current limitations of AI hardware as motivation, and armed with close to £50 million in funding, ARIA's Scaling Compute program backs twelve different research projects that have shown potential to develop alternative avenues of research in computing, contributing to lowering the cost of AI hardware by more than 1000x.

Among the awarded projects, Fractile's "Better Analogue in Memory Matrix-vector Multiplication" will enable the startup to test whether its novel approach, based on analog in-memory compute, can be used to "run frontier models orders of magnitude faster than current state of the art." Along with Fractile, ARIA's Scaling Compute program also supports research projects from other startups working on optimizing AI hardware, including Alphawave Semi, Rain AI, and Normal Computing.

Fractile's award follows the startup's seed funding raise in July, when Fractile raised $15 million in a round co-led by Kindred Capital, the NATO innovation fund, and Oxford Sciences Enterprises, with participation from Inovia Capital and Cocoa.

The award news also follows a partnership between Fractile and Andes Technology. As part of this partnership, Fractile will leverage Andes Technology's high-performance RISC-V vector processor and integrate it within its innovative in-memory compute architecture via ACE. This partnership represents another milestone in Fractile's journey towards delivering a fast and cost-efficient AI inference system which overcomes the limitations of current AI hardware.