Samsung Electronics’ latest high bandwidth memory (HBM) chips have not yet passed Nvidia’s tests due to issues with heat and power consumption, Reuters news report said.
These issues affect Samsung’s HBM3 chips, the current fourth-generation standard widely used in GPUs for AI, as well as the upcoming fifth-generation HBM3E chips set to hit the market this year, the report said.
In response to Samsung’s failure in Nvidia’s tests, Samsung emphasized that HBM is a customized product requiring optimization in line with customer needs and is actively working on this through close collaboration.
HBM features vertically stacked chips to save space and reduce power consumption, crucial for processing the massive data volumes generated by complex AI applications. As demand for advanced GPUs has surged alongside the AI boom, the need for HBM has similarly increased.
Meeting Nvidia’s standards is critical for HBM manufacturers due to Nvidia’s dominant position in the global GPU market for AI applications. Samsung has been attempting to pass Nvidia’s tests for its HBM3 and HBM3E chips since last year.
Recent test results for Samsung’s 8-layer and 12-layer HBM3E chips in April revealed ongoing issues, increasing industry and investor concerns that Samsung might lag further behind rivals SK Hynix and Micron Technology in the HBM market.
SK Hynix, Nvidia’s primary HBM supplier, has been providing HBM3 since June 2022 and began supplying HBM3E in late March.
Micron has also announced it will supply Nvidia with HBM3E.
In contrast, Samsung has not yet secured a position as an HBM3 supplier to Nvidia but does supply other customers, including Advanced Micro Devices (AMD). Samsung plans to start mass production of HBM3E chips in the second quarter of this year.
Reflecting its concerns, Samsung recently replaced the head of its semiconductor unit, highlighting the crisis in the industry. Jeff Kim, head of research at KB Securities, noted that while market expectations for Samsung were high, it is not unusual for specialized products like HBM to require time to meet performance evaluations.
Despite its challenges, Samsung continues to invest in HBM technology, having developed the first commercial HBM solution for high-performance computing in 2015. Analysts attribute SK Hynix’s technological edge to its longer and more resource-intensive focus on HBM R&D over the past decade.
GPU manufacturers like Nvidia and AMD are keen for Samsung to perfect its HBM chips to provide more vendor options and reduce SK Hynix’s pricing power. At a recent Nvidia AI conference, Nvidia CEO Jensen Huang endorsed Samsung’s 12-layer HBM3E, signaling interest in Samsung’s progress.
Trendforce predicts that HBM3E chips will dominate the market this year, with shipments increasing in the latter half of 2024. SK Hynix estimates an annual growth rate of 82 percent for HBM demand through 2027.
