Beyond ZFLOS: What’s next? Architecting the future of AI systems

AI technology is rapidly evolving beyond generative models toward the era of Physical AI. To date, AI system development has been dominated by the pursuit of higher FLOPs, achieved primarily through performance enhancements in chips such as CPUs, GPUs, and high-bandwidth memory (HBM). However, the emerging demands of next-generation AI systems have outgrown the capabilities of a single chip solution, calling for a fundamentally new paradigm.

To address the critical bottlenecks of power consumption, bandwidth, latency, and thermal management, it is no longer sufficient to develop logic, memory, and packaging in isolation. Instead, a holistic system-level approach is required - one that enables simultaneous and co-optimized designs of these elements. This necessitates not only advanced architectural design capabilities, but also semiconductor process technologies capable of realizing such deep integration.

In this presentation, we will present Samsung’s future technology development strategy and share case studies on AI system architecture that generate tangible customer value and deliver AI solutions aligned with evolving market needs. Furthermore, we will introduce our approach for accelerating R&D through AI, detailing the current progress and implementation status of this initiative.