三星2nm工艺迎来新客户

Core Viewpoint - DEEPX has signed a development agreement with Samsung Foundry and Gaonchips to create the next-generation generative AI chip, DX-M2, utilizing advanced 2nm technology, marking DEEPX as the first commercial client for this process in the AI chip sector [2][3]. Group 1: Development and Production Timeline - The multi-project wafer (MPW) prototype production is scheduled for the first half of 2026, with mass production targeted for 2027, aiming to enhance South Korea's high-value semiconductor industry and accelerate the global 2nm AI semiconductor ecosystem [2]. - The early commitment to Samsung's 2nm GAA process is crucial for DEEPX, as achieving practical device-level generative AI with existing technology nodes is nearly impossible due to high computational demands and strict power constraints [2][3]. Group 2: Performance and Efficiency - The 2nm process offers a power-to-performance ratio that is approximately twice as efficient as DEEPX's current 5nm-based DX-M1 chip, which is essential for overcoming technological barriers [3]. - The DX-M2 chip is designed to provide ultra-low power, device-level AI solutions, significantly outperforming data center capabilities in terms of energy efficiency, with a performance ratio tens of times higher than traditional data centers [3][5]. - The chip can perform real-time inference on models with up to 20 billion parameters at a speed of 20 to 30 tokens per second, consuming only 5 watts or less, which is a significant advancement compared to competitors running smaller models [3][4]. Group 3: Strategic Importance - DX-M2 is positioned to be the world's first practical device-level solution capable of real-time processing of advanced generative AI models, bridging the gap between high-performance AI and energy efficiency [4][5]. - DEEPX's CEO emphasizes that DX-M2 will lead the large-scale application and industrialization of generative AI, aiming to eliminate current cost and energy barriers associated with AI technology [5].