BlockBeats News, July 6th. Critini Research analyst Jukan pointed out that Samsung and SK Hynix are reevaluating the timing of adopting hybrid bonding in HBM, and they may even postpone it until HBM5. There are two core reasons:
First, JEDEC is discussing loosening the thickness specification of HBM5 to a maximum of about 1000μm (HBM3E is 720μm, and HBM4 has already been expanded to 775μm). After the standard relaxation, the advantage of hybrid bonding with no bumps for thinning is no longer urgent;
Second, there are simpler alternative solutions to the heat dissipation issue—Samsung has developed the Heat Path Block, and SK Hynix has introduced iHBM (ICE HBM), both of which involve placing independent heat dissipation devices next to HBM, with plans to start implementing them from HBM5. The technical difficulty is lower, and the commercialization is more reliable.
In addition, major customers such as NVIDIA do not currently have a pressing demand for products with more than 16 layers of stacking. 12-layer products are still the mainstream in the HBM4E stage. However, the research and development of hybrid bonding have not stagnated. The I/O count of the current HBM4 has doubled to 2048, and the existing TC compression bonding process is reaching its limit; if the I/O doubles again to 4096 in the future HBM5E phase, the lateral diffusion of bumps will make it difficult for TC bonding to support, and copper-to-copper direct bonding hybrid bonding must be adopted to achieve higher-density connections.
Jukan's assessment: In the short term, due to the simpler solutions for thickness and heat dissipation, hybrid bonding will not be widely deployed. However, in the medium to long term, when the I/O density explodes again, it is still the inevitable direction. This will have a direct impact on the market expectations of Besi, the core supplier of hybrid bonding equipment. The delay in technology transition means that the scale and timing of related equipment orders need to be reassessed.
