IEEE Spectrum breed · Technology

New IEEE Program Addresses AI Chip Design Challenges

The increasing complexity of AI hardware, driven by larger models, presents challenges. A new IEEE program aims to equip professionals with skills to navigate these complexities.

Modern AI models are computationally demanding. This leads to AI chips optimized for specific tasks, as data movement is a bottleneck. This "AI memory wall" requires domain-specific accelerators and co-optimization.

The "AI Processor Architecture, Design Principles, and Performance" program offers five courses on AI processor technologies. It covers design principles, advanced architectures, neural processing units, trends, and edge/IoT design.

The program targets AI hardware professionals, including chip designers and embedded systems developers. It bridges theory with cross-disciplinary engineering reasoning.

The curriculum covers AI processor conception, structure, and optimization, including compute units, memory, and dataflows. Participants will evaluate processor behavior.

A dialogue-driven approach uses AI avatars representing engineers discussing scenarios. This makes complex concepts approachable and interactive.

Avatar-based instruction can increase learner confidence and improve retention. The program models expert reasoning in AI processor development.

Learners actively participate in resolving trade-offs, simulating an apprenticeship. This program merges technical content with an innovative delivery model.

AI-samenvatting op basis van de bron.