On September 9th, according to the &#34;Korean Economic Daily&#34;, Dan Kochpatcharin, head of the Taiwan Semiconductor ecosystem and alliance management, stated at the Semicon Taiwan 2024 forum last week that Samsung and Taiwan Semiconductor are collaborating to produce bufferless high-bandwidth memory (HBM).
The report mentioned that the collaboration between Samsung and Taiwan Semiconductor will provide &#34;customized chips and services&#34; requested by customers such as Nvidia and Google.
Compared to existing models, the energy efficiency of bufferless HBM4 will increase by 40%, and latency can be reduced by 10%. HBM has been widely used in AI computing.
Although Samsung is a competitor to Taiwan Semiconductor in logic process foundry, Taiwan Semiconductor does not produce DRAM. However, the adoption of multi-chip packaging or so-called &#34;Chiplet&#34; advanced packaging in AI processor collaboration is increasing, including HBM.
Moreover, when HBM transitions to the new generation HBM4, it will use logic process-based base chips, allowing customers to incorporate their own IP for customization and enhance the efficiency of HBM. Both Samsung and SK Hynix will allow customers to design the base chips for this logic process and choose external logic process wafer foundries for production.
Sources also indicated that while Samsung can provide comprehensive HBM4 manufacturing services, including memory production, logic process base chip foundry, and advanced packaging, it hopes to gain more customers by leveraging Taiwan Semiconductor's technology.
This move by Samsung may be an attempt to counter competitor SK Hynix, which leads as the top HBM supplier with 53% market share, while Samsung holds only 35%.
It is worth noting that SK Hynix, Samsung, and Micron are all introducing HBM3E DRAM and plan to launch HBM4 format in 2025. SK Hynix recently announced its intention to develop the next generation of HBM, with performance 20-30 times that of current products and offering customized products to customers.
Article by: Chip Intelligence, Source: Chip Intelligence, Original Title: &#34;Samsung is Collaborating with Taiwan Semiconductor to Develop HBM4&#34;

GOOGL

SSNGY

GOOG

Samsung and Taiwan Semiconductor are collaborating to develop High Bandwidth Memory 4 (HBM4) without buffer, in order to meet the demand from customers such as NVIDIA and Alphabet for customized chips and services. The energy efficiency of HBM4 will be improved by 40% compared to existing models, with a 10% reduction in latency. Despite competing in the logic foundry sector, Samsung hopes to attract more customers by leveraging TSMC's technology, aiming to counter SK Hynix's market leadership position

- Samsung and TSMC are collaborating on producing High Bandwidth Memory (HBM), aiming to enhance efficiency by 40% and reduce latency by 10% with the new HBM4 model, catering to Nvidia and Google's demand for customized chips. The shift towards HBM4 involves advanced packaging methods like Chiplet technology, allowing for increased customization and efficiency. Samsung seeks to leverage TSMC's expertise to expand its customer base in the HBM market, potentially challenging SK Hynix, the current leading HBM supplier. SK Hynix, Samsung, and Micron are all gearing up to launch HBM4 in 2025, with SK Hynix planning a significant performance boost for customized products.

Samsung is collaborating with Taiwan Semiconductor to develop HBM4