South Korean AI Chipmakers Turn to TSMC for Next-Gen Processors Amid Samsung Foundry Competition
Several South Korean AI processor developers are moving their production from Samsung Foundry to Taiwan Semiconductor Manufacturing Company (TSMC) for their next-generation chips. Companies like DeepX, FuriosaAI, and Mobilint are opting for TSMC’s advanced nodes to enhance performance and mitigate risks. DeepX, known for its DX-M1 AI accelerator and DX-V1 chip made using Samsung’s technologies, plans to use TSMC’s 12nm node for its upcoming DX-V3 chip.
FuriosaAI, which utilized Samsung’s 14nm process for its first-generation Warboy chip, is now manufacturing its Renegade processor, designed for large language models (LLMs), using TSMC’s 5nm node. This next-gen processor boasts significant improvements in computing power, utilizing TSMC’s advanced CoWoS packaging with high-bandwidth memory (HBM3). The company also plans to continue using TSMC for its refined RenegadeS chip.
Mobilint, while still producing its Aries NPU with Samsung’s 14nm technology, is shifting its ultra-low-power Regulus chip to TSMC’s 12nm node. The decision allows the company to achieve improved power efficiency while maintaining competitive performance in low-power applications.
The shift to TSMC reflects a strategic move to diversify supply chains and leverage better yields, pricing, and performance for AI-driven processors. Experts recommend that Samsung Foundry bolster its support for smaller fabless companies to close the market share gap with TSMC, emphasizing the need for improved design assets and services to remain competitive.
source: ZDNet
