As HBM competition heats up, introduction schedule of 'hybrid bonding' receiving attention

Amy Fan, Taipei; Jack Wu, DIGITIMES Asia 0

Credit: Samsung Electronics

With the increased prevalence of generative AI, the demand for High Bandwidth Memory (HBM) is rapidly rising, leading to heightened competition in stacking. For HBMs, a higher stack means the capability to process more data. Currently, major HBM products feature 8-layer stacking, with 12-layer stacking expected to enter mass production soon.

According to a report by South Korea's The Elec, the current technologies used in the HBM stacking process are primarily Thermal Compression (TC) bonding and high-efficiency Mass Reflow (MR). Samsung Electronics uses TC with non-conductive film (NCF), while SK Hynix utilizes MR with molded underfill (MUF). As with the hybrid bonding technology, it is expected to be officially introduced with 12-layer stacking.

Observations from the South Korean industry suggest that Samsung and SK Hynix's current HBM stacking technology will reach its limit at 12 layers. In the next generation of HBM products, which could feature 12 or even 16 layers, the adoption of hybrid bonding technology is expected to speed up.

In fact, sources from SK Hynix previously stated that MR will be applied to 12-layer HBM and that they are currently developing new technology to apply hybrid bonding technology to the next generation of high-capacity, high-stack HBM.

Hybrid bonding involves connecting chips copper-to-copper (Cu-Cu), as opposed to the traditional solder bump and ball method. Compared to existing techniques, hybrid bonding can significantly increase input and output (I/O) speeds.

However, the mass production of HBM with 12 layers or more still needs to address issues such as warpage and HBM height. The chips used in HBM stacking are extremely thin, and warping may occur during mass production. Samsung has previously pointed out that TC bonding has an advantage in solving the warping in HBMs. Additionally, the need to reduce HBM height is due to consideration for packaging and other factors.

South Korean media reports emphasize that both Samsung and SK Hynix are not willing to comment on a specific application timeline of hybrid bonding. Industry insiders also suggested that hybrid bonding research is not limited to the HBM field; it is also being explored in the realm of 3D stacking.

Because hybrid bonding is expected to maximize I/O speeds upon adoption, sources believe that the adoption timing of this technology in high-performance HBM products will arrive sooner than expected.