According to BlockBeats, on July 6th, Jukan, an analyst at Critini Research, pointed out that Samsung and SK Hynix are reassessing the timing of adopting hybrid bonding in HBM, and may not adopt it even by HBM5. There are two main reasons for this:
First, JEDEC is discussing relaxing the thickness standard for HBM5 to a maximum of about 1000μm (HBM3E is 720μm, and HBM4 has been relaxed to 775μm). With the standard relaxed, the advantage of bumpless thinning of hybrid bonding is no longer urgent.
Secondly, there are simpler alternatives to the heat dissipation problem—Samsung developed Heat Path Block, and SK Hynix launched iHBM (ICE HBM), both of which place independent heat dissipation devices next to HBM and are planned to be applied starting with HBM5. The technology is less difficult and commercialization is more reliable.
Furthermore, major customers like NVIDIA do not currently have an urgent need for high-stack products with 16 or more layers, and 12-layer products may still be the mainstream in the HBM4E stage. However, research and development of hybrid bonding has not stopped. The number of I/Os in HBM4 has now doubled to 2048, and the existing TC hot-press bonding process is nearing its limit; if the number of I/Os further doubles to 4096 in the future HBM5E stage, the lateral diffusion of bumps will make TC bonding difficult to support, at which point hybrid bonding with direct copper bonding will be necessary to achieve higher density connections.
Jukan predicts that hybrid bonding will not see large-scale deployment in the short term due to simpler solutions for thickness and heat dissipation; however, it remains an inevitable trend in the medium to long term when I/O density explodes again. This will directly impact the market expectations of Besi, a core supplier of hybrid bonding equipment. The delay in this technological shift means that the timeline for scaling up orders for related equipment needs to be re-evaluated.