According to Luc Van den hove rapid AI algorithm innovation was fast outpacing the current strategy of developing specific chips, which was in itself having an impact on energy, cost and hardware development speed.
"By the time the AI hardware is finally ready, the fast-moving AI software community may have taken a different turn," said Van den hove.
Van den hove pointed to companies that were building custom chips to speed up innovation but warned that approach was risky and for most companies uneconomical.
According to Van den hove future chips will see their capabilities merged into building blocks called supercells, with a network-on-chip steering and reconfiguring these supercells so they can be quickly adapted.
This approach will require true three-dimensional stacking, a manufacturing technique where layers of logic and memory silicon are bonded together, Van den hove said.
He warned that without this approach hardware is going to struggle to handle diverse workloads and consequently we need to, “prevent bottlenecks from slowing down next-gen AI” and “we must reinvent the way we do hardware innovation.”
In conclusion brute force compute is unlikely to be the right solution going forward especially when dealing with agentic AI and physical AI, which will require a host of different models.
“The root of the problem is that AI is often running on a suboptimal compute architecture, consisting of suboptimal hardware components for the specific workload the algorithms need. Adding new, challenging workloads to the mix will cause AI-related energy use to rise exponentially. Making it even more challenging is the fact that AI workloads could change overnight, instigated by a new algorithm,” Van den hove warned in his keynote address.
The answer, at least for Van den hove, is innovative silicon hardware that is programmable and as ‘codable’ as software is.
A reconfigurable approach will give more companies the ability to design their own hardware for specific AI workloads.
As Van den hove concluded, “AI’s future hinges on hardware innovations. And given the vast impact of AI on all societal domains, it is probably not exaggerated to state that our very future hinges on it.”
