TLDR: DEEPX has signed a 2nm process development agreement with Samsung Foundry and its design partner Gaonchips for its next-generation generative AI chip, the ‘DX-M2’. This collaboration aims to produce the world’s first on-device generative AI chip using Samsung’s advanced 2nm Gate-All-Around (GAA) process, with mass production targeted for 2027. The DX-M2 is designed for ultra-low-power generative AI inference on devices, promising to significantly reduce the cost and energy barriers of current AI technologies.
Seoul, South Korea – DEEPX, a leading on-device AI semiconductor company, has announced a landmark 2nm process development agreement with Samsung Foundry and its design partner, Gaonchips. This strategic partnership is set to revolutionize the generative AI landscape with the development and production of DEEPX’s next-generation chip, the ‘DX-M2’. The DX-M2 is poised to become the world’s first on-device generative AI chip to utilize Samsung Foundry’s cutting-edge 2nm Gate-All-Around (GAA) process, marking a significant leap in semiconductor technology for artificial intelligence applications.
According to Lokwon Kim, CEO of DEEPX, ‘DX-M2 is the core platform that will usher in the mass adoption and industrialization of generative AI.’ He further elaborated, ‘By combining 2nm process efficiency with our AI architecture, we aim to eliminate the cost and energy barriers of today’s AI and create a world where anyone can benefit freely and sustainably from this technology.’ This initiative underscores DEEPX’s commitment to making advanced AI accessible and sustainable across various devices.
Development of the DX-M2 began with architecture design in early 2024, leading to a demonstration-ready prototype by early 2025. The ambitious timeline includes Multi-Project Wafer (MPW) prototype production in the first half of 2026, with mass production slated for 2027. The chip is specifically engineered for ultra-low-power generative AI inference on devices, promising to deliver unparalleled computational performance, intelligence, and power efficiency.
The DX-M2 is designed to handle generative AI models with up to 20 billion parameters, performing real-time inference at a speed of 20 to 30 words per second while consuming less than 5W of power. This represents a substantial improvement in power efficiency, approximately doubling that of the previous DX-M1 chip, which utilized a 5nm process. This enhanced efficiency is crucial for enabling expert-level AI capabilities directly on devices such as robots, home appliances, and laptops, thereby reducing reliance on cloud-based servers and mitigating thermal and power constraints.
Also Read:
- Generative AI Propels Robotics Market Towards Explosive Growth, Led by NVIDIA and OpenAI
- Korea District Heating Corporation Joins Energy Sector Collaboration for AI-Driven Innovation
Industry experts anticipate that this project will significantly contribute to the early establishment of the 2nm system semiconductor ecosystem and further advance the domestic fabless industry. The DX-M2 is a cornerstone of DEEPX’s strategy to expand its portfolio into generative and multimodal AI, driving the transition towards a hyper-intelligent, AI-driven society.
