Samsung’s HBM3E Hits Snag in Nvidia Certification Process
Previous reports indicated that Micron, SK Hynix, and Samsung had successively provided Nvidia with samples of 8-layer vertically stacked HBM3E (24GB) in late July, mid-August, and early October of last year. Among them, Micron and SK Hynix’s HBM3E passed Nvidia’s validation earlier this year and received orders. According to DigiTimes, Samsung’s HBM3E has yet to pass Nvidia’s tests and requires further validation.
It is understood that Samsung’s inability to pass Nvidia’s validation is primarily due to the approval process at TSMC. As the manufacturer and packager of Nvidia’s data center GPUs, TSMC is also a critical participant in Nvidia’s validation process. Rumors suggest that the testing standards are based on SK Hynix’s HBM3E products, whereas Samsung’s HBM3E products have slight differences in manufacturing processes, such as SK Hynix using MR-RUF technology and Samsung using TC-NCF technology, which could impact certain parameters.
In its first-quarter 2024 financial report released last month, Samsung stated that 8-layer vertically stacked HBM3E had entered mass production in April, with plans to mass-produce 12-layer vertically stacked HBM3E in the second quarter, ahead of the originally planned second half of the year. According to Samsung, this acceleration is to better meet the growing demand for generative AI.
Industry insiders revealed that if the testing standards are adjusted accordingly, Samsung’s HBM3E will likely pass Nvidia’s validation without issue.
