Samsung’s compliance with Nvidia’s HBM3 and HBM3E memory chips has been making headlines, with new data suggesting that the first memory type has already received customer approval, and the second should do so within the next four months. However, there is a risk that this long-awaited event will happen as early as 2025.
Image source: Samsung Electronics
The agency reports with renewed vigor on the misadventures of Samsung Electronics in this area Bloomberg. According to informed sources, Samsung has already received approval from Nvidia to use its HBM3 chips in the developer’s products for graphics processors, and the first product with such a combination of components could be the H20 computing accelerator, created specifically for the Chinese market given the current US sanctions.
As for the more modern HBM3E memory chips by Samsung, the company hopes to certify them according to Nvidia requirements either in two or four months, depending on the performance discipline of the employees involved and luck. By the end of the year, Samsung would like to establish supplies of HBM3E chips for Nvidia’s needs. If everything does not go according to plan, then the blessed moment may be postponed to 2025.
As is known, the smaller company SK hynix is not only ahead of Samsung in terms of the pace of introducing new types of HBM memory to the market, but also controls more than half of the world market for such chips. Morgan Stanley analysts believe that if Samsung is successful in certifying its memory to Nvidia’s requirements, it will take at least 10% of the world market next year, which will allow it to significantly increase its revenue. Experts expect that the capacity of the world HBM market will grow from last year’s $4 billion to $71 billion by 2027. Even if Samsung is slightly behind its competitors now, the production capacities at its disposal will allow it to quickly catch up and find a worthy place in the market.
Some sources believe that Samsung has been let down by its earlier decision to use the TC-NCF method with thermal compression of a non-conductive film to connect memory layers in a stack. SK hynix uses a different method, and its share of the HBM market leaves no doubt about its effectiveness. At the same time, Samsung has no doubt about the demand for the TC-NCF method in the market and does not demonstrate a willingness to modify the technology for combining memory layers in a stack. And yet, sources claim that in order to obtain approval from Nvidia, the South Korean giant has gone for a modification of the HBM3 production technology. The company has been producing this type of memory since the second half of last year.
Nvidia first started using HBM3E this year, and currently only gets the chips from SK hynix. Next year, HBM3E will be in much higher demand, and SK hynix simply won’t be able to meet the market demand quickly, even if it keeps its promise to quadruple HBM3E production. Micron Technology has also managed to get approval for its HBM3E chips in Nvidia products, so only Samsung is left in line. The company is working to complete certification of both 8-layer and 12-layer chips of this type according to Nvidia’s requirements.
If you notice an error, select it with your mouse and press CTRL+ENTER.
