According to information gathered from five reliable sources, Samsung Electronics is planning to use a chip-making method supported by rival SK Hynix in an effort to compete in the growing market for high-end circuits driving artificial intelligence (AI). This is a calculated move by Samsung, the world’s top memory chip maker, as they work to close the gap in the production of high-bandwidth memory (HBM) chips that are essential for artificial intelligence (AI) applications.
The rise in popularity of generative AI corresponds with the increase in demand for HBM processors. Samsung has noticeably stayed out of the fray until recently, while competitors SK Hynix and Micron Technology have actively entered into collaborations with AI chip leader Nvidia to produce state-of-the-art HBM chips.
Samsung’s production difficulties are attributed by analysts and industry watchers to the company’s dedication to non-conductive film (NCF) chip manufacturing technology. On the other hand, in order to address the limitations of NCF, SK Hynix switched to the mass reflow molded underfill (MR-MUF) technique. Three people with intimate knowledge of the situation have revealed that Samsung is changing its strategy and placing buy orders for chipmaking equipment designed for the MUF technology.
One insider claims that Samsung’s adoption of the MUF process represents a pragmatic change, even if it was made grudgingly since it is in line with the approach that SK Hynix pioneered. As of right now, many experts have confirmed that SK Hynix produces HBM3 chips at yield rates of 60–70%, whilst Samsung’s yields are now between 10 and 20%.
The newest HBM chips, HBM3 and HBM3E, are attractive because they integrate with core CPU processors to analyze large datasets in generative AI with efficiency. Samsung is actively enlisting material producers, such as Japan’s Nagase, to supply MUF materials; but, mass manufacturing using MUF is not expected to occur until next year, pending more testing. Samsung’s activities go beyond the acquisition of equipment.
Remarkably, Samsung plans to use both NCF and MUF processes for its newest HBM chip, demonstrating the adaptability of its manufacturing approach. Samsung reiterates its commitment to the progress of HBM technology by stating that its in-house created NCF technology is still the best option for HBM products and would be prominently featured in its upcoming HBM3E chips.
Nvidia and Nagase declined to respond when asked, citing the sensitive nature of the material and the need to protect secrecy. All of the sources have asked to remain anonymous in order to protect the privacy of the conversations.
Samsung’s adoption of MUF highlights the growing competitiveness in the battle for AI chips, especially in the HBM chip market, which TrendForce, a research group, projects will reach $9 billion in sales this year due to demand spurred by AI.
NCF vs. MUF
Through the use of thermally compressed thin film to reduce the amount of space between stacked chips, chipmakers are now able to stack numerous layers of chips compactly in high-bandwidth memory chipsets thanks to the use of non-conductive film chip production technology. Nevertheless, when production complexity increases with more layers, problems related to adhesive materials emerge. These difficulties are best shown by Samsung’s most recent HBM3E chip, which has 12 chip layers. This has led chipmakers to look at other approaches.
Thanks to its effective switch to the bulk reflow molded underfill process, SK Hynix is now able to supply Nvidia with HBM3 chips before its rivals. According to analyst Jeff Kim of KB Securities, the fact that SK Hynix holds a nearly 80% market share in HBM3 and advanced HBM products for Nvidia highlights the importance of technical breakthroughs in sustaining market supremacy.
Competition is heightened by Micron’s recent entry into the high-bandwidth memory chip market. Nvidia plans to use Micron’s HBM3E chip to power its H200 Tensor processors, which are scheduled for release in the second quarter. Samsung’s HBM3 series, on the other hand, has not yet met Nvidia’s requirements for supply agreements, indicating further difficulties in establishing strategic alliances.
Investors are aware of Samsung’s losses in the race for AI chips; this year, its shares have dropped 7%, trailing the gains of SK Hynix and Micron, which have increased by 17% and 14%, respectively. This indicates how the market views Samsung’s competitiveness in this emerging market.
Samsung has made a calculated decision to strengthen its position in the AI chip race by strategically switching to the MUF technology. This shows that the company recognizes the need to keep up with technical breakthroughs in order to be competitive in a market that is changing quickly. The industry participants competing for dominance in the AI chip market depend heavily on the convergence of innovation and collaboration, given the ongoing spike in demand for AI-driven technology.