Dating and Marrying: An AI Chiplets Perspective

Learn about the challenges and potential solutions in AI chiplet development through this 18-minute conference talk by IBM Research Principal Researcher Arvind Kumar. Explore how disaggregation into chiplets represents a promising future direction for IBM's AI accelerator SoC, with particular focus on how AI chiplets could combine with other chiplets in an open ecosystem to create powerful domain-specific architectures. Examine key barriers facing AI chiplet implementation, including the absence of architectural modeling tools, interoperability challenges between chiplets, validation complexities, packaging sourcing issues, and testing methodologies. Gain insights into potential approaches for overcoming these technical hurdles as the technology progresses toward enabling more modular and specialized AI hardware solutions.