DEEPX today announced that it has signed a tripartite agreement with Samsung Foundry and GAONCHIPS to develop its next-generation generative AI semiconductor, the DX-M2, utilizing Samsung’s cutting-edge 2-nanometer (nm) process technology.

With this agreement, DEEPX will commence full-scale semiconductor fabrication for the DX-M2, an ultra-low-power on-device generative AI inference processor, becoming one of the early commercial customers of Samsung Foundry’s 2nm node. The multi-project wafer (MPW) run for prototype production is scheduled for the first half of next year, with mass production targeted for the following year.

DEEPX previously developed its vision AI-optimized DX-M1 on Samsung’s 5nm process two years ago and is currently focused on yield optimization to enhance manufacturing efficiency. The DX-M1 is deployed across various industries — including robotics, smart cameras, and factory automation — and is being co-developed with global customers for volume production. Building on this success, DEEPX aims to lead the transition toward a hyper-intelligent society by delivering a comprehensive product portfolio spanning vision AI, generative AI, and multimodal AI with its second-generation DX-M2.

The DX-M2 is designed to deliver real-time inference for 20-billion-parameter (20B) generative AI models at 20–30 tokens per second, all within a sub-5W power envelope. This enables deployment of expert-level AI models directly on devices with stringent thermal and power constraints — such as robots, home appliances, and laptops — without reliance on cloud computing.

Leveraging a Mixture of Experts (MoE) architecture, the DX-M2 aims to execute large-scale generative AI models, such as DeepSeek and LLaMA 4 at the 20B scale, while achieving quasi-AGI performance equivalent to 100B-parameter models — all processed in real time entirely on-device.

“The DX-M2 is a key platform ushering in the era of democratized and industrialized generative AI,” said Dr. Lokwon Kim, CEO of DEEPX. “We will continue to lower technological barriers and contribute to creating a world where everyone can benefit from AI.”

Source: https://www.sisajournal-e.com/news/articleView.html?idxno=414350