New Technology Increases Semiconductor Density by 4 Times Over HBM, Boosting AI Chip Performance

A new technology has been developed that allows for stacking semiconductors with a density approximately four times higher than High Bandwidth Memory (HBM), which is currently gaining global attention.
Pohang University of Science and Technology (POSTECH) announced on Tuesday (June 30) that a research team led by Professor Kim Suk of the Department of Mechanical Engineering, integrated master's and doctoral student Kim Woo-hyun, and Dr. Keum Ho-hyun of the Korea Institute of Industrial Technology has developed a technique to stably stack ultra-thin semiconductor chips.
HBM, which currently determines the performance of artificial intelligence (AI) semiconductors, is structured by stacking multiple memory chips, making the ability to stably stack a large number of chips a critical challenge.
The problem is that as chips become thinner, they become more difficult to handle, making chips thinner than a human hair—specifically those under several dozen micrometers (㎛)—prone to bending or breaking.
In response, the research team developed a technology that integrates "transfer printing," which precisely moves and attaches chips to desired locations, with "real-time bonding," which completes metal bonding the moment the chip is transferred.
Using this method, the team succeeded in stably stacking 14㎛-thick ultra-thin chips more than 10 layers high under low-temperature and low-pressure conditions.
The research team explained that even after continuous stacking, the interlayer alignment error was small, and warping was minimized.
The "integration density," which represents the number of stacked layers relative to the total thickness, reached a level approximately four times higher than that of existing HBM.
Professor Kim Suk stated, "We expect this to be utilized as a core foundational technology for the development of high-performance AI semiconductors and next-generation memory systems."
(Photo: Provided by POSTECH, Yonhap News)
※ Please note: This article was translated by AI and may contain errors.