Written by Yang Hee-kyung
SEOUL (Reuters) – Samsung Electronics, the world's top memory chip maker, aims to catch up in the race to produce high-end chips that power artificial intelligence, using chip-making technology backed by rival SK Hynix. Five people involved revealed that the plan is to do so.
With the growing popularity of generative AI, the demand for high-bandwidth memory (HBM) chips has skyrocketed. However, unlike peers SK Hynix and Micron Technology, Samsung is conspicuously not involved in any deal with AI chip leader Nvidia to supply its latest HBM chips.
One of the reasons Samsung fell behind was its decision to stick with a chip manufacturing technology called non-conductive film (NCF), which caused production problems, while Hynix decided to stick with a chip manufacturing technology called non-conductive film (NCF) that addressed NCF's weaknesses. The company switched to the mass reflow mold underfill (MR-MUF) method. According to analysts and industry watchers.
However, Samsung recently placed orders for chip manufacturing equipment designed to support MUF technology, three people with direct knowledge of the matter said.
“Samsung had to do something to increase HBM (production) yields…The adoption of MUF technology is a bit of a sacrifice of pride for Samsung, because Samsung “This is because we decided to follow the technology that was first used,” one of the people said.
According to multiple analysts, the production yield of Samsung's HBM3 chips is about 10-20%, while SK Hynix has secured a production yield of about 60-70% for HBM3.
The latest versions of HBM chips, HBM3 and HBM3E, are in high demand. These are bundled with a core microprocessor chip that helps process vast amounts of data in generative AI.
According to a source familiar with the matter, Samsung is also negotiating with material manufacturers including Japan's Nagase Sangyo to procure MUF materials, but mass production of high-end chips using MUF requires Samsung to operate. More tests, he added, likely won't be ready until next year at the earliest.
The three sources also said Samsung plans to use both NCF and MUF technologies in its latest HBM chips.
Samsung said its self-developed NCF technology is the “optimal solution” for HBM products and will be used in the new HBM3E chip. “We are executing our HBM3E product business as planned,” Samsung said in his statement.
Nvidia and Nagase declined to comment.
All officials spoke on condition of anonymity because the information is private.
Samsung's MUF adoption plan highlights the growing pressure Samsung faces in the AI chip race, according to research firm TrendForce. is expected to more than double to nearly $9 billion.
NCF vs MUF
Non-conductive film chip manufacturing technology allows stacking multiple layers of chips into compact high-bandwidth memory chipsets by using thermo-pressed thin films to minimize the space between stacked chips. Widely used by chip manufacturers to
However, as more layers increase manufacturing complexity, problems related to adhesive materials often arise. According to Samsung, the latest HBM3E chip has 12 chip layers. Chipmakers have been looking for alternatives that address these weaknesses.
SK Hynix was the first to successfully switch to high-volume reflow underfill technology, becoming the first vendor to supply HBM3 chips to Nvidia.
SK Hynix's market share in HBM3 and more advanced HBM products for NVIDIA is estimated to be more than 80% this year, according to KB Securities analyst Jeff Kim.
Micron entered the high-bandwidth memory chip race last month, announcing that its latest HBM3E chip will be chosen to power NVIDIA's H200 Tensor chips, which start shipping in the second quarter.
Samsung's HBM3 series has not yet qualified for Nvidia's supply deal, according to one of the four sources and another person familiar with the discussions.
The company's setbacks in the AI chip race have also been noted by investors, with its stock price down 7% this year, lagging behind SK Hynix and Micron, which have risen 17% and 14%, respectively.
(Reporting by Heekyung Yang and Joyce Lee in Seoul, Fanny Potkin in Singapore, Sam Nassey in Tokyo and Max Charney in San Francisco; Editing by Miyoung Kim and Himani Sarkar)
