Heekyung Yang and Fanny Potkin
SEOUL/SINGAPORE (Reuters) – Samsung Electronics’ latest high-bandwidth memory (HBM) chips have not yet passed Nvidia’s (NASDAQ:) tests for use in the U.S. firm’s artificial intelligence processors due to problems with heating and power consumption, three people said. they said.
The problems affect Samsung’s (KS:) HBM3 chips, which are the fourth-generation HBM standard currently most widely used in graphics processing units (GPUs) for artificial intelligence, as well as the fifth-generation HBM3E chips that the South Korean tech giant and its According to him, its competitors will bring it to market this year.
The reasons for the failure of Samsung tests on Nvidia tests have been reported for the first time.
In a statement to Reuters, Samsung said HBM is a specialized memory product that requires “optimization processes in tandem with customer needs,” adding that the company is in the process of optimizing its products through close collaboration with customers. The company declined to comment on specific clients.
In separate statements after Reuters first published the report, Samsung said that “claims of failures due to heat and power consumption are not true” and that testing was “proceeding smoothly and as planned.”
NVIDIA declined to comment.
HBM, a type of dynamic random access memory or DRAM standard first released in 2013 in which chips are arranged vertically to save space and reduce power consumption, helps process the massive amounts of data generated by complex artificial intelligence applications. As demand for sophisticated GPUs has skyrocketed amid the generative AI boom, so has the demand for HBM.
Meeting the needs of Nvidia, which controls about 80% of the global market for GPUs for artificial intelligence applications, is seen as key to the future growth of HBM manufacturers – both in terms of reputation and profit dynamics.
Samsung has been trying to pass Nvidia’s HBM3 and HBM3E benchmarks since last year, three sources said. According to two sources, the results of a recent failed test of Samsung’s 8-layer and 12-layer HBM3E chips were received in April.
It’s unclear whether the problems can be easily resolved, but three sources said Nvidia’s failure to comply has fueled fears in the industry and among investors that Samsung could fall further behind rivals SK Hynix and Micron Technology (NASDAQ:). in HBM.
The sources, two of whom were briefed on the matter by Samsung officials, spoke on condition of anonymity because the information was confidential.
Samsung shares fell 2% in early trading on Friday, slightly lower than the broader market.
NEW CHIP BLOCK HEAD
Unlike Samsung, domestic rival SK Hynix is Nvidia’s main supplier of HBM chips and has been shipping HBM3 since June 2022. At the end of March, the company also began shipping HBM3E to a buyer it declined to identify. According to sources, supplies went to Nvidia.
Micron, the only major HBM manufacturer, also said it will supply Nvidia with HBM3E.
In a move that analysts say underscores Samsung’s concerns about its lagging position in HBM, Samsung this week replaced the head of its semiconductor division, saying it needed a new person at the top to handle that , which she called a “crisis” affecting the industry.
Market expectations were high that Samsung, as the world’s largest memory chip maker, would quickly pass Nvidia’s tests, but it was also natural that dedicated products like HBM would take some time to meet customer performance scores, said Jeff Kim, head of research . at KB Securities.
While Samsung has not yet become an HBM3 supplier to Nvidia, it supplies customers such as Advanced Micro Devices (NASDAQ:) and plans to begin mass production of HBM3E chips in the second quarter. In a statement to Reuters, Samsung said its product launch schedule was on track.
Analysts also say that SK Hynix, which developed the first HBM chip in 2013, has spent far more time and resources on HBM research and development than Samsung over the past decade, which explains its technological advantage.
In a statement, Samsung said it developed the first commercial HBM solution for high-performance computing in 2015 and has continued to invest in HBM since then.
The sources also said GPU makers such as Nvidia and AMD are keen for Samsung to improve its HBM chips to give it more options to suppliers and weaken SK Hynix’s pricing power.
At the Nvidia AI conference in March, Nvidia CEO Jensen Huang signed a sign at the Samsung booth declaring that “Jensen has approved” Samsung’s 12-layer HBM3E, underscoring the company’s enthusiasm for the prospect of Samsung supplying these chips.
HBM3E chips are likely to become the main HBM product in the market this year, with shipments concentrated in the second half of 2024, according to research firm Trendforce.
SK Hynix estimates that demand for HBM memory chips overall could increase by 82% per year until 2027.
Samsung’s relatively weak position in HBM has been noticed by investors, analysts said. The company’s shares are down 2% year to date, while SK Hynix shares are up 42% and Micron shares are up 48%.
($1 = 1,362.8600 won)