◎ 일시 : 2025.12.2. (화) 16:00

◎ 장소 : 연세대학교 제1공학관 A685호

◎ 제목 : Unstoppable Rise of On-device AI and Physical AI

◎ 연사 :  김녹원 대표 / ㈜딥엑스 

◎ 초청 : 시스템반도체공학과 정재용 교수

◎ 내용 :

 As generative AI scales, the limits of cloud-centric computing are becoming structural—power, cost, latency, and infrastructure capacity. This shift is accelerating the rise of on-device AI, where intelligence is executed directly at the endpoint. In this talk, I will discuss the technical foundations of on-device AI, including memory architecture, compute sparsity, model compression, and low-power acceleration.

 Especially, I will present DEEPX’s DX-M2 architecture as a case study: a sub-5W AI processor capable of running 10–30B models with real-time throughput, enabling AI PCs, robots, cameras, and industrial systems to operate without cloud dependency. Beyond architecture, I will examine the emerging industry landscape—why global semiconductor leaders are pivoting toward Physical AI, how on-device compute can reduce data-center load by over 80%, and what opportunities Korea has to lead with next-generation 2nm low-power AI chips.