As early as 2013 to 2015, SK Hynix and AMD were at the forefront of the memory industry with the first generation of high-bandwidth memory (HBM), and SK Hynix still leads in terms of market share. In order to maintain and develop its position, SK hynix must adapt to customer needs, especially in the field of artificial intelligence. To this end, it is considering how to create “differentiated” HBM products for large customers.
“Developing customer-specific AI memory requires a new approach as the flexibility and scalability of the technology become critical,” said Hoyoung Son, Vice President, Head of Advanced Packaging Development at SK Hynix.
In terms of performance, HBM memory with a 1024-bit interface developed quite quickly: in 2014-2015 the data transfer rate was 1 GT/s, and later it reached 9.2 GT/s – 10 GT/s. The recently launched HBM3E memory device. With HBM4, memory will transition to a 2048-bit interface, which will ensure a steady increase in bandwidth over HBM3E.
But the vice president said some customers may benefit from differentiated (or semi-custom) solutions based on HBM.
“In order to realize diverse AI, the characteristics of AI memory also need to become more diverse,” Hoyoung Son said in an interview with BusinessKorea. “Our goal is to have a variety of advanced packaging technologies that can cope with these changes. We aim to have a variety of advanced packaging technologies that can cope with these changes. Plans to provide differentiated solutions capable of meeting any customer need.”
For the 2048-bit interface, many (if not most) HBM4 solutions will likely be custom or at least semi-custom based on what we know from official and unofficial information about the upcoming standard. Some customers may want to continue using interposers (but this time they will become very expensive), while others prefer to mount the HBM4 module directly on the logic chip using direct bonding technology, which is also expensive .
Manufacturing differentiated HBM products requires complex packaging technologies, including (but of course not limited to) SK Hynix’s advanced reflow molded underfill (MR-RUF) technology. Given the company’s extensive experience with HBM, it is likely that it will come up with other options, especially differentiated products.
“In order to achieve different types of AI, the characteristics of AI memory also need to be more diverse,” the vice president said. “Our goal is to have a range of advanced packaging technologies to respond to the changing technology landscape. Going forward, we plan to provide differentiated solutions to meet the needs of all customers.”
Source: BusinessKorea, SK Hynix