Samsung Unveils Next-Gen AI Memory Chips at NVIDIA GTC
三星於 NVIDIA GTC 大會發表新一代 AI 記憶體晶片
At the 2026 NVIDIA GTC event in San Jose, Samsung Electronics showcased its evolution from a memory manufacturer to a comprehensive "Total AI Solution" provider.
在聖荷西舉辦的2026年NVIDIA GTC大會上，三星電子展示了其從記憶體製造商蛻變為全方位「全面人工智慧解決方案」提供商的過程。
A major highlight was the unveiling of HBM4E, the industry’s first seventh-generation High-Bandwidth Memory.
其中一個重大的亮點是推出了業界首款第七代高頻寬記憶體HBM4E。
Boasting 16Gbps per pin and 4TB/s bandwidth, it is set to power NVIDIA’s upcoming Vera Rubin Ultra platform.
它擁有每針腳16Gbps的傳輸速率與4TB/s的頻寬，將為NVIDIA即將推出的Vera Rubin Ultra平台提供動力。
Beyond memory, Samsung introduced Hybrid Copper Bonding, an innovative packaging technique that enhances thermal resistance by 20%, allowing for the stacking of 16 or more layers.
除了記憶體之外，三星還引進了混合銅鍵合技術，這是一種創新的封裝技術，能提升20%的抗熱性，並允許堆疊16層或以上的記憶體。
Jensen Huang, NVIDIA’s CEO, emphasized the strength of this partnership by autographing a Samsung HBM4 wafer.
NVIDIA執行長黃仁勳透過在三星HBM4晶圓上簽名，強調了雙方合作關係的穩固。
Furthermore, Samsung is actively manufacturing the Groq 3 Language Processing Unit (LPU) as a key foundry partner.
此外，三星正作為關鍵晶圓代工夥伴，積極製造Groq 3語言處理單元。
Looking ahead, the company is already developing HBM5 and HBM5E to address critical AI data center bottlenecks like heat management and transmission speed.
展望未來，該公司已在研發HBM5與HBM5E，以解決人工智慧資料中心在散熱管理與傳輸速度上的關鍵瓶頸。
By integrating memory, logic, foundry services, and advanced packaging, Samsung is positioning itself as a cornerstone of the next generation of artificial intelligence hardware.
透過整合記憶體、邏輯晶片、晶圓代工服務與先進封裝技術，三星正將自己定位為下一代人工智慧硬體的核心基石。
