Samsung’s upcoming fifth technology HBM3e reminiscence product has been codenamed “Shinebolt,” with the advertising and marketing and growth of Samsung’s new HBM3e reminiscence catching as much as the nice work SK hynix has been doing with its personal next-gen reminiscence.
2VIEW GALLERY – 2 IMAGES
In line with sources at Enterprise Korea, Samsung Electronics is presently transport out HBM3e prototypes to a few of its shoppers for high quality approval (QA) testing. Samsung is reportedly transport out 24 gigabit (Gb) chips in 8 layers (8-Hello), however will quickly have a 36GB HBM3e product with 12 layers (12-Hello).
Samsung’s new HBM3e “Shinebolt” reminiscence could have a 50% efficiency increase over HBM3, with an enormous 1.228TB/sec (12280MB/sec) reminiscence bandwidth. The long run is AI; everyone knows it, with NVIDIA’s crazy-fast AI GPUs needing extra and far quicker VRAM. That is the place Samsung’s new HBM3e “Shinebolt” reminiscence shines — pun not meant, but it surely simply works so effectively right here — capability and bandwidth.
When it comes to manufacturing, the bonding course of enterprise is continuous to develop and is a much-needed requirement of recent HBM reminiscence applied sciences. Samsung has been utilizing the thermal compression-non-conductive movie (TC-NCF) technique from the very first few days of HBM manufacturing. Its competitor — SK hynix — has been utilizing the superior mass reflow-molded underfill (MR-MUF) course of, which SK hynix solely began utilizing when its new HBM3 reminiscence went into manufacturing.
Samsung has not too long ago put extra vitality into its HBM enterprise, re-strategizing with a purpose to higher compete with SK hynix. In line with sources, Samsung is gearing up the event of HBM’s “blended connection” course of which might be “altering the principles of the sport”.
Uhhh, convey that on, Samsung. I need to see it. The world needs to see the principles of the HBM enterprise being modified. Particularly in a world targeted so onerous on AI that NVIDIA is bought out of its AI GPUs proper by way of deep into 2024. At this fee, a 2024 AI GPU launch will hit palms in 2026, and a 2025 AI GPU shall be delivered in 2028. That is quite a lot of high-speed VRAM required, and Samsung is aware of that, and it is aware of it must be severe to fulfill NVIDIA’s strict necessities for its high-end, very costly AI GPUs.
Lee Jung-bae, president of Samsung Electronics’ reminiscence enterprise, stated in a current article titled “Unleashing the Infinite Potentialities of Samsung Reminiscence” posted on the corporate’s newsroom: “We’re presently in manufacturing of HBM3 and are easily creating the next-generation product, HBM3E. We are going to additional broaden to provide custom-made HBM for our shoppers“.
