Samsung (SSNLF.US) has broken through the 80% yield threshold for HBM4, officially entering the "golden yield" stage, with revenue in the third quarter expected to triple.

According to reports, Samsung Electronics (SSNLF.US) has significantly improved the yield rate of its sixth-generation high-bandwidth memory (HBM4) from less than 60% during the early stages of mass production to around 80%, reaching the company's previously set target for the end of the year ahead of schedule. The industry refers to an 80% yield rate as the "golden yield," and this breakthrough indicates that Samsung has taken a crucial step in profitability and large-scale supply of HBM4. In February of this year, Samsung was the first in the world to launch mass production of HBM4, but the initial yield was less than 60%. After about six months of process optimization, the yield has stabilized quickly. Reports citing industry insiders indicate that Samsung originally planned to achieve an 80% yield rate by the end of 2026, but the progress is now clearly ahead of schedule. Regarding the reasons for the yield improvement, reports point out that improvements in the thermal compress non-conductive film (TC-NCF) process are an important support for the yield increase; other analysts have emphasized that the improvement in the yield of the 1c DRAM base chip used in HBM4 has also played a significant role. With the realization of the "golden yield," Samsung's capacity expansion and profit improvement expectations have significantly strengthened. The company has made it clear that HBM4 revenue in the third quarter will increase by more than three times quarter-on-quarter, and in the second half of the year, HBM4 will account for more than 60% of Samsung's overall HBM revenue. In terms of market share, Samsung has set its year-end target for HBM market share at around 38%, in line with its share in the DRAM sector. Analysts believe that the expansion of HBM4 output will not only strongly support the production of NVIDIA Corporation's (NVDA.US) next-generation AI accelerator Vera Rubin but will also help major customers like NVIDIA Corporation diversify their supply chains and reduce excessive dependence on a single supplier. Lee Jong-hwan, a professor of system semiconductor engineering at Sangmyung University in South Korea, stated that from the perspective of major customers like NVIDIA Corporation, diversifying supply sources is clearly more advantageous than relying on a single supplier. If Samsung can fully leverage the opportunity of expanding HBM4 supply, it is expected to rapidly consolidate its market leadership. It is noteworthy that Samsung's competitor SK Hynix (SKHY.US) is also making swift progress in HBM4. Industry insiders estimate that SK Hynix's HBM4 yield rate has also reached around 80%. With its mass production experience accumulated over several generations of HBM products, along with its unique advanced quality backfill molding (MR-MUF) technology, the industry generally expects SK Hynix to smoothly ramp up production capacity in the second half of the year.