AI chips evolve as imec lab pursues programmable designs

The rapid evolution of artificial intelligence algorithms is creating an unexpected challenge for chipmakers, with traditional design approaches potentially becoming a bottleneck for future AI development. According to Imec CEO Luc Van den hove, the semiconductor industry must pivot toward more flexible, reconfigurable architectures to keep pace with AI’s swift advancement, or risk creating costly hardware that becomes obsolete before it’s even deployed.

The big picture: One of the world’s leading semiconductor R&D organizations warns that current chip design strategies focusing on raw computing power are creating unsustainable energy consumption, rising costs, and slow hardware development cycles.

• “There is a huge inherent risk of stranded assets because by the time the AI hardware is finally ready, the fast-moving AI software community may have taken a different turn,” Van den hove stated.
• Belgium-based imec has significant influence in the industry, pioneering semiconductor breakthroughs that are later adopted by major manufacturers like TSMC and Intel.

The proposed solution: Van den hove envisions future chips utilizing “supercells” – building blocks that combine various capabilities and can be quickly reconfigured as AI algorithms evolve.

• “A network-on-chip will steer and reconfigure these supercells so they can be quickly adapted to the latest algorithm requirements,” he explained.
• This approach would require true three-dimensional stacking, a manufacturing technique where layers of logic and memory silicon are bonded together.

Behind the technology: The three-dimensional stacking technology championed by imec is already set to appear in upcoming advanced chip designs.

• imec has been a significant contributor to the advancement of 3D stacking, which will be featured in TSMC’s A14 and Intel’s 18A-PT future node technologies.
• This technology would be particularly crucial as AI evolves beyond large language models to agentic AI and physical AI applications in fields like medicine and autonomous driving.

Why it matters: While some organizations like OpenAI have pursued custom chip development to accelerate innovation, Van den hove cautions this approach is risky and economically unfeasible for most companies.

• The mismatch between hardware development cycles and software innovation threatens to create expensive technological dead-ends.
• The semiconductor industry’s ability to address this challenge will directly impact the pace of future AI advancement across multiple industries.