Samsung and TSMC to Co-Develop HBM4 AI Chips for 2025 Mass Production
Samsung Electronics and Taiwan Semiconductor Manufacturing Company (TSMC), traditionally fierce rivals in the semiconductor foundry market, have announced a strategic partnership to jointly develop bufferless High-Bandwidth Memory 4 (HBM4) chips. This collaboration was revealed by Dan Kochpatcharin, TSMC’s head of Ecosystem and Alliance Management, during the Semicon Taiwan 2024 forum. HBM technology is critical for artificial intelligence applications due to its superior processing speeds compared to traditional DRAM. The new bufferless HBM4 architecture is projected to deliver a 40% improvement in power efficiency and a 10% reduction in latency over current models. The two tech giants aim to provide customized chip solutions and services to major clients, including NVIDIA and Google. According to reports from the Korea Economic Daily, mass production of these next-generation AI chips is scheduled to begin in the second half of 2025. This unprecedented alliance highlights the intensifying demand for advanced AI hardware and the industry's shift towards collaborative innovation to meet the rigorous performance requirements of modern data centers and AI systems.
Wire timeline
Samsung and TSMC to Co-Develop HBM4 AI Chips for 2025 Mass Production
Samsung Electronics and Taiwan Semiconductor Manufacturing Company (TSMC), traditionally fierce rivals in the semiconductor foundry market, have announced a strategic partnership to jointly develop bufferless High-Bandwidth Memory 4 (HBM4) chips. This collaboration was revealed by Dan Kochpatcharin, TSMC’s head of Ecosystem and Alliance Management, during the Semicon Taiwan 2024 forum. HBM technology is critical for artificial intelligence applications due to its superior processing speeds compared to traditional DRAM. The new bufferless HBM4 architecture is projected to deliver a 40% improvement in power efficiency and a 10% reduction in latency over current models. The two tech giants aim to provide customized chip solutions and services to major clients, including NVIDIA and Google. According to reports from the Korea Economic Daily, mass production of these next-generation AI chips is scheduled to begin in the second half of 2025. This unprecedented alliance highlights the intensifying demand for advanced AI hardware and the industry's shift towards collaborative innovation to meet the rigorous performance requirements of modern data centers and AI systems.
TechNode