Imec CEO Calls for 'Programmable' AI Chips to Address Future AI Challenges

Imec CEO Advocates for Reconfigurable Chip Architectures

Luc Van den hove, CEO of Imec, one of the world's leading semiconductor research and development firms, has called for a significant shift in the chip industry's approach to artificial intelligence (AI) hardware. In a statement ahead of Imec's flagship conference, ITF World, Van den hove emphasized the need for reconfigurable chip architectures to keep pace with the rapidly evolving AI landscape 1

1

2

2

.

Current Challenges in AI Chip Development

The CEO highlighted that the current strategy of developing specific, raw-power-focused chips for AI applications is becoming increasingly problematic. This approach is leading to major drawbacks in energy efficiency, cost-effectiveness, and hardware development speed. Van den hove warned of a "huge inherent risk of stranded assets" as the fast-moving AI software community may outpace hardware development 1

1

2

2

.

The Proposal: Supercells and Network-on-Chip

To address these challenges, Van den hove proposed a new approach for future AI chips:

1. Supercells: All necessary capabilities would be regrouped into building blocks called supercells.
2. Network-on-Chip: A system that would steer and reconfigure these supercells, allowing for quick adaptation to the latest algorithm requirements 1

   1

   2

   2

   .

This innovative approach aims to create more versatile and adaptable AI hardware that can keep up with the rapid pace of AI software development.

The Role of 3D Stacking

Van den hove emphasized that implementing this new chip architecture would require advanced manufacturing techniques, specifically "true three-dimensional stacking." This process involves bonding layers of logic and memory silicon together 1

1

2

2

. Imec has been a significant contributor to the advancement and refinement of 3D stacking technology, which is set to be featured in future chip designs by industry giants TSMC and Intel 1

1

.

Industry Implications and Alternatives

The CEO's statements come at a time when some companies, like OpenAI, are opting to build custom chips to accelerate innovation. However, Van den hove cautioned that this approach is risky and economically unfeasible for most companies 1

1

2

2

.

Imec's influence in the semiconductor industry is substantial, with many of its breakthroughs often adopted by major chipmakers years later. As such, this call for a shift towards reconfigurable architectures could have far-reaching implications for the future of AI hardware development 1

1

2

2

.

Beyond Large Language Models

Van den hove also touched on the evolving landscape of AI applications. He noted that as the industry moves beyond large language models to more advanced forms of AI, such as agentic AI and physical AI for medical or autonomous driving applications, the need for adaptable chip architectures becomes even more critical 1

1

2

2

.

As the AI industry continues to evolve at a rapid pace, Imec's proposal for programmable AI chips represents a potential solution to ensure that hardware development can keep up with the ever-changing demands of AI software innovation.