Beyond ZFLOPS: What's Next?
Architecting the Future of AI Systems
Recent advances in AI have accelerated the transition from generative AI to physical AI. To support this paradigm shift, AI systems have primarily focused on enhancing chips – including CPU, GPU, HBM, Advanced PKG – to maximize FLOPs performance.
However, the realization of next-generation FLOPs performance using a single chip faces fundamental limitations, therefore a new approach beyond conventional architectures is required.Conventional single-chip based AI architecture have been constrained by key bottlenecks, including power, latency, bandwidth, and thermal management. Overcoming these limitations requires a holistic AI system architecture in which design, logic, memory and packaging are simultaneously optimized. Samsung is creating differentiated value through the development of Memory Aware Logic, Logic Aware Memory, and Advanced PKG technologies that go beyond simple interconnects. These capabilities enable the delivery of customer-specific solutions, including custom ASICs and custom memory, while also expanding continuous development in advanced process nodes and differentiated product portfolios.
With comprehensive in-house capabilities across memory, foundry, packaging and design, Samsung is uniquely positioned on a global scale to deliver fully customized and optimally integrated AI solutions aligned with customer requirements.
