Arm’s former unofficial tech futurologist joins an obscure startup that wants to solve AI’s sustainability problem – by developing computers that use almost no power
Artificial intelligence systems, especially those using machine learning and deep learning, require an enormous amount of energy due to the computational intensity of the tasks they perform. These processes require powerful hardware, which in turn consumes a lot of electricity. Microsoft is even considering nuclear powering its data centers to tackle the problem.
As AI technology continues to develop and its use becomes more widespread, associated energy consumption is expected to increase, raising concerns about its impact on the environment. Andrew Sloss, a seasoned engineer who recently ended his 25-year tenure at Arm as senior principal research engineer, has taken on a new role at a new UK startup, Variety Computingto tackle this problem from a different angle.
Vaire, which has joined incubator organization Silicon Catalyst UK to advance its plans, has the ambitious goal of tackling ‘reversible computing’, which could lead to the development of logic circuits that consume almost no power.
Physics-aware architectures
In his role at Arm, Sloss focused on future technology, and Vaire’s work seems to be the perfect next step for him. Talk about moving him LinkedInSloss says: “As we all know, the semiconductor industry is under increasing pressure, especially as we move towards the physical limits of Moore’s Law (EoML), i.e. a silicon atom is ~2 Angstroms in size, thermal problems are increasing, etc. . , i.e., ‘You cannot change the laws of physics.’ Therefore, it forces us to go back to the decisions we made in the past and rethink them. We face new and emerging pressures that affect scalability in compute and power constraints (e.g. ML, LLMs and DL).”
He further explains that Vaire Computing wants to “rewrite the laws of George Boole’s thought and the von Neumann architecture by creating physics-aware architectures (with behavioral support for existing software). It is a matter of making logic aware of physics (from transistors upwards). This concept is not the standard 1-2-3 game. Physics offers us a huge amount of tricks that we have not yet used in the architectural world.”
This suggests that Vaire could explore a similar avenue Normal computer use, a company staffed by former members of the Google Brain Team and X Engineers who built generative AI production systems for Alphabet.