As they advance next-generation High Bandwidth Memory (HBM) technology, Samsung Electronics and SK Hynix are facing a critical decision: whether to fully adopt hybrid bonding technology. Despite its significant advantages, the two giants have become increasingly cautious in their rollout, as the immediate demands for HBM "thinning" and "heat dissipation" have eased. However, the industry widely expects that hybrid bonding will eventually be implemented due to the impending explosive growth in HBM input/output (I/O) terminals. In the short term, the industry will rely on a transitional phase using Thermal Compression (TC) bonding combined with independent cooling solutions for HBM4 and the early 12-layer versions of HBM4E.
Technology Rollout Falls Short of Expectations; Mass Production May Be Delayed
According to industry sources cited by South Korean media, the widespread application of hybrid bonding in next-generation HBM may occur later than previously anticipated. While the market had once optimistically predicted that the technology would debut in the sixth-generation HBM (HBM4), this vision was not realized on schedule due to technical challenges and other factors.
Currently, leading manufacturers like Samsung Electronics and SK Hynix continue to advance hybrid bonding R&D. At present, HBM mass production primarily relies on TC bonding, a process that uses tiny bumps and supporting underfill between DRAM chips, which are then connected using heat and pressure.
In contrast, hybrid bonding directly connects the copper wiring of each DRAM chip. By eliminating the need for bumps, this technology not only effectively reduces the overall thickness of the HBM but also significantly improves thermal performance and power efficiency, while enabling higher-density I/O connections. However, based on current progress, both Samsung and SK Hynix have opted for traditional TC bonding for HBM4. Industry forecasts suggest that hybrid bonding will be implemented at the earliest in the 16-layer version of HBM4E (an enhanced iteration of HBM4).
Reduced Demand for Thinning and Heat Dissipation: The Main Reason for the Delay
Why has the industry become more conservative regarding hybrid bonding? The primary reason is that the demand for its two core advantages—reducing HBM thickness and improving heat dissipation—has weakened at this stage.
Regarding thickness, industry standards are gradually being relaxed. The standard thickness for fifth-generation HBM (HBM3E) was 720 micrometers, while for HBM4, this limit has been raised to 775 micrometers, primarily because DRAM stack configurations have increased from 8/12 layers to 12/16 layers. It is reported that JEDEC is currently discussing further relaxing the thickness standards for the next-generation, 20-layer HBM5, potentially allowing it to reach 900 to 1,000 micrometers. This relaxation means manufacturers no longer need to compress DRAM spacing to the extreme, thereby alleviating the stringent requirements on bonding technology.
Additionally, a slowdown in demand from core customers like NVIDIA for ultra-high-stack HBM has become a major variable. According to industry insiders, "Discussions between customers and memory manufacturers regarding 16-layer HBM are currently not very active. It is highly likely that the 12-layer version will remain the mainstream product for HBM4E."
Alternative Approaches: Independent Cooling Solutions Expected to Debut in HBM5
Another major advantage of hybrid bonding is the removal of underfill, which has a low thermal conductivity, thereby improving HBM heat dissipation. However, Samsung and SK Hynix have recently developed alternative solutions by integrating independent cooling components next to the HBM. Samsung refers to its technology as the Heat Path Block (HPB), while SK Hynix calls its solution iHBM (ICE HBM). Both companies are currently testing the integration of these technologies for HBM5.
A packaging industry expert noted, "There are no major technical obstacles to implementing and placing cooling components next to the HBM core die, and commercialization should not be overly difficult. For memory manufacturers, this is a more stable choice."
Looking Ahead: Surging I/O Counts Will Force a Technology Upgrade
Despite short-term headwinds, Samsung and SK Hynix have not halted their hybrid bonding R&D. The industry points out that once the I/O count and density of next-generation HBM increase significantly, hybrid bonding will become a mandatory option.
For example, HBM4 has doubled its I/O count to 2,048 compared to the previous-generation HBM3E. This requires a substantial reduction in internal HBM spacing. Because traditional TC bonding causes bumps to spread laterally during heating, it can no longer meet the manufacturing demands of higher I/O densities.
In the medium to long term, the industry is discussing doubling the I/O count again to 4,096 starting from the HBM5E generation. At that point, the I/O pitch will become extremely narrow, making the introduction of hybrid bonding absolutely necessary.
All Rights Reserved © 2026 MemoryMarket.com
