(C13-6) Scalable Physical AI Inference Solution with NPU Chiplets

The rapid expansion of Physical AI is creating diverse computing requirements. A single fixed system architecture cannot efficiently address these requirements. Chiplet-based solutions provide a more adaptable path. By combining specialized compute and functional dies within a common platform, system developers can tailor solutions for different applications and scale AI inference capacity as requirements evolve. By utilizing NPU chiplets as modular building blocks, a common software-compatible architecture can support products spanning multiple performance tiers. This session explores how NPU chiplets enable flexible and scalable inference for Physical AI systems, balancing performance, system integration, and software portability while supporting a broad range of deployment requirements.