Former Arm executives join ‘Tsetlin machine’ startup

News | February 20, 2024

By Peter Clarke

Artificial Intelligence AI Analog Tsetlin machines

Noel Hurley, a former vice president at processor IP licensor Arm, has joined AI startup Literal Labs as its CEO. At the same time Jem Davies, who led the graphics and AI business groups at Arm, has taken a position as a non-executive director.

Literal Labs (Newcastle-upon-Tyne, England) has also gone through a name change. The company was founded as Mignon Technologies Ltd. in March 2023.

The company was created by Professor Alex Yakovlev, who serves as chief scientific officer, and Rishad Shafik, who serves as chief technology officer. Both are at the University of Newcastle where they have been active in research on Tsetlin machines.

Literal Labs claims that Tsetlin machines can provide an approach to artificial intelligence (AI) that is up to 10,000 times more energy efficient than artificial neural networks. The company has now recruited two industry veterans to help guide its growth and market entry.

Hurley spent eleven years with Arm from 1994 to 2005 before leaving and co-founding Xmos Semiconductor Ltd. He returned to Arm in 2011 as a vice president in the processor division and spent another 11 years there, during which he held a number of senior posts including general manager of the CPU group.

Davies spent 18 years at Arm during which time he rose to be an Arm Fellow, vice president and general manager of the media processing group and the machine learning group.

A Tsetlin machine?

Literal Labs technology is based on developing and training Tsetlin machine models specific to customer datasets. This results in an optimised machine model that can then be deployed on to target hardware. From Literal Labs the output is a Tsetlin machine model that can be deployed as software only – to run on industry-standard hardware – or can be accelerated using Literal Labs processors.

“Our benchmarking shows we can achieve 250x faster inferencing than XG Boost using software only, and up to 1,000x faster and up to 10,000x less energy consumption when using hardware acceleration,” the company is quoted saying on the Cambridge Future Tech website. Cambridge Future Tech is an incubator and investment organization based in Cambridge, England.

For now the business model has not been disclosed. It could be to license software and hardware IP, in a manner similar to Arm, or Literal Labs could aspire to sell its own chiplets or packaged ICs as a fabless chip company. In either case success will depend on prooving the claimed efficiency of Tsetlin machines to perform various tasks.

Do the maths

A Tsetlin machine is a form of pattern learning automaton based on propositional logic. The method was created and given its name by Professor Ole-Christoffer Granmo, who named it after Soviet mathematician Michael Lvovitch Tsetlin.

A Tsetlin machine is described as being able to perform tasks such as classification, regression, and convolution, and produce interpretable and explainable results. It is reported in the literature that compared to neural networks, a well-designed Tsetlin Machine has lower computational complexity and resource usage, and does not suffer from local optima or vanishing gradients.

Tsetlin (born 1924, died 1966) was a mathematician and physicist who introduced the Gelfand-Tsetlin basis for finite dimensional representation of classical groups. Tsetlin is recognised as the founder of the learning automaton approach to machine learning.

Granmo is professor and director at the Centre for Artificial Intelligence Research (CAIR) at the University of Agder in Norway. Granmo’s seminal paper on Tsetlin machines in which he named them is: The Tsetlin Machine – A Game Theoretic Bandit Driven Approach to Optimal Pattern Recognition with Propositional Logic, published in 2018.