In the team’s method, the toughest challenge was to ensure that the bonding was secure, removing potential moisture or air bubbles on a minuscule scale. Just three molecules of water between the bonding layers would be enough to break the connection between layers. Removing this hazard, along with air bubbles, while maintaining clean surfaces for the layers that are bonding together — and ensuring that this process could be reliably carried out again and again — were the main challenges the team sought to overcome in their research.
The team’s method is similar to how chip wafers are bonded, which is a standard industry practice at this point — just at a considerably smaller scale — and they’ve been using an ASMPT machine designed for die-to-wafer bonding to carry out their tests. While the team believes that their method would allow chiplets to be built and bonded together at scale, more work needs to be done on researching the tooling of ASMPT’s systems.
The systems of tomorrow are going to require new ways to solve computing problems as they become more complex and varied, and chiplets provide a potential pathway to solving those problems. With hybrid bonding, creating systems that act as if they were part of a single chip, that future comes sharper into view.
Katsuyuki Sakuma, Roy Yu, Nicholas Polomoff, Luke Darling, Promod Chowdhury, Sathya Raghavan, Keodara Seifert, John Knickerbocker, Dale McHerron, Ming Li, Siu Cheung So, So Ying Kwok, Chun Ho Fan, Siu Wing Lau, “Voids-free Die-level Cu/ILD Hybrid bonding,” 2023 IEEE 73rd , Electronic Components and Technology Conference (ECTC), Florida, USA, 2023. ↩