SK Hynix has sent 12-layer stacked HBM3E samples to Nvidia for product verification testing

High Bandwidth Memory (HBM) products are regarded as one of the cornerstones of artificial intelligence (AI) computing, with the industry witnessing rapid development over the past two years. Under the influence of artificial intelligence (AI) and high-performance computing (HPC), memory manufacturers have seen a surge in revenue growth. As a leading partner of NVIDIA in high-bandwidth memory, SK Hynix currently holds a dominant position in the HBM market, extensively supplying HBM3 for NVIDIA’s range of artificial intelligence chips.

According to ZDNet, SK Hynix dispatched new 12-layer stacked HBM3E samples to NVIDIA last month for product validation tests.

HBM technology necessitates the interconnection of multiple layers of DRAM on a base silicon wafer through silicon vias (TSVs). The initial batch of HBM3E products featured an 8-layer stack with a capacity of 24GB. Micron, SK Hynix, and Samsung each sent samples to NVIDIA in July, August, and October of the previous year, respectively. Reports have indicated that NVIDIA has made advance payments of approximately $540 million to $770 million to SK Hynix and Micron for the supply of the next-generation HBM3E.

Micron, SK Hynix, and Samsung are all advancing the development of 12-layer stacked HBM3E, aiming to increase the capacity to 36GB. This development requires addressing certain device characteristics and reliability verification issues. Last month, Samsung announced the industry’s first 12-layer stacked HBM3E, boasting a bandwidth of up to 1280GB/s, and anticipates commencing mass production in the latter half of this year. However, it appears that SK Hynix is also making rapid progress, rumored to have implemented new manufacturing processes.

Industry insiders have stated that the samples provided by SK Hynix represent early versions, primarily intended to establish new product standards and characteristics. SK Hynix refers to these as “Universal Test Vehicles (UTV),” anticipating that the product validation tests will not consume excessive time. With the rapid growth of the artificial intelligence chip market, competition among HBM products has intensified.