Samsung Electronics, Broadcom Forge AI Chip Alliance Worth KRW 292 Trillion, From HBM to 2 nm
TECHWORLD ·
✦ AI Summary
Samsung Electronics and Broadcom are expanding their strategic cooperation to support next-generation AI infrastructure. The scope of the partnership is widening beyond memory supply to include leading-edge processes and 2.3D and 2.5D advanced packaging, while memory solutions including HBM are also being supplied for Broadcom's next-generation AI accelerators. The two companies signed an MOU at the AI Summit in San Francisco, and the combined memory-and-foundry cooperation is projected to exceed USD 200 billion over the next 5 years through 2030.
Samsung Electronics and Broadcom are expanding their strategic partnership to support next-generation AI infrastructure. According to industry sources on the 2nd, the two companies are broadening their existing collaboration beyond memory supply to include leading-edge process technologies and advanced packaging to meet demand for AI infrastructure.
Samsung Electronics will supply memory solutions, including HBM, for Broadcom's next-generation AI accelerators, and the scope of cooperation covers memory, foundry, and advanced packaging. The partnership will extend to Samsung Electronics' leading-edge process technologies below 2 nm, as well as its 2.3D and 2.5D advanced packaging technologies.
To that end, the two companies signed a strategic memorandum of understanding, or MOU, for next-generation AI infrastructure at the AI Summit in San Francisco, U.S. The core of the partnership is the combination of Samsung Electronics' memory, foundry, and advanced packaging technologies with Broadcom's AI and network semiconductor design capabilities.
The combined memory and foundry cooperation is projected to exceed USD 200 billion over the next 5 years through 2030.
The direction of cooperation between Samsung Electronics and Broadcom is the supply of advanced memory solutions, including HBM, for Broadcom's next-generation AI accelerators. AI accelerators are required to process massive amounts of data in a short time. Their key competitive factors are computing performance, data transfer speed between the processor and memory, and power efficiency.
Against this backdrop, the importance of jointly optimizing HBM bandwidth, capacity, and power efficiency from the AI accelerator design stage is growing. As not only throughput but also memory connectivity performance and power efficiency have become important, the need to align these factors early in the design process is increasing.
Samsung Electronics began mass production shipments of HBM4 in February this year, using 1c DRAM and a 4 nm base die. It then supplied HBM4E samples to customers in May.
Samsung Electronics is pursuing expanded supply of high-performance, high-efficiency memory for Broadcom's next-generation AI accelerators based on HBM4 and HBM4E. It is also working to further define its next-generation HBM roadmap.
At a Sept. 1 presentation at SEMICON Taiwan, Samsung Electronics said HBM5's target performance is twice that of HBM4E. It also presented plans to improve the HBM5 target performance-to-power ratio by 20% and reduce thermal resistance in the HBM5 development direction by 20%. Samsung is reportedly seeking to apply its 2 nm process to the HBM5 base die.
As the possibility of broader cooperation with Broadcom emerges, the partnership is moving beyond HBM4 and HBM4E toward combining Samsung Electronics' next-generation HBM, leading-edge logic, and packaging technologies.
This possibility suggests that memory collaboration could expand into a combination of next-generation HBM with logic and packaging. The structure of the Broadcom partnership could enable integrated cooperation with Samsung Electronics in the logic, memory, and packaging required for AI accelerators and network semiconductors.
Foundry cooperation is also under way. Samsung Electronics' 2 nm and below leading-edge process technologies are being applied to major Broadcom products, and the products targeted for application include Broadcom's wireless broadband communication, or WBC, solutions that support next-generation high-speed data communications.
The two companies aim to realize high-performance, low-power AI and network semiconductors based on the 2 nm process. This collaboration is not limited to a simple foundry production contract.
Samsung Electronics is presenting a one-stop turnkey solution that links semiconductor design and process co-optimization with chip design, advanced packaging, and mass production. Accordingly, the collaboration will proceed in a way that Samsung Electronics provides an integrated package from design optimization to packaging and mass production.
This announcement is a strategic partnership in the form of an MOU, not a supply contract with specific terms finalized. It is not a finalized supply deal, and the size of orders by product, the timing of mass production, and the process to be applied remain undecided.
The two companies plan to expand the scope of cooperation to 2.3D and 2.5D advanced packaging technologies, in addition to foundry.
AI semiconductors require the integration of a large logic chip and multiple HBM stacks into a single package. Packaging technology is increasingly seen as a key factor that determines system performance.
In AI accelerators, it is necessary to reduce the distance of data movement between the logic die and HBM, increase bandwidth, and minimize power consumption.
Leading-edge processes alone have limits in improving performance. Accordingly, the approach of optimizing HBM, logic, and packaging simultaneously is becoming more important.
Samsung Electronics is pushing a strategy of providing integrated solutions to Broadcom by combining its capabilities in memory, logic, foundry, and advanced packaging. It plans to support the realization of high-performance, high-efficiency AI semiconductors through advanced packaging based on 2 nm.
As reflected in the headline, there is a shift in which Samsung's 2 nm opportunity expands amid TSMC bottlenecks. This collaboration is linked to shortages in leading-edge foundry and advanced packaging supply in the AI semiconductor market.
According to domestic reports on Aug. 25, Broadcom was said to be seeking to raise up to USD 100 billion to expand AI semiconductor supply. The report analyzed that the need to expand the supply chain had grown because of limited leading-edge process capacity at TSMC, its existing main production partner.
Amid this trend, there is speculation that a USD 200 billion-scale partnership with Samsung Electronics could serve as the basis for establishing a separate long-term supply line. Accordingly, the need for Broadcom to diversify its supply chain is being linked to growing opportunities for Samsung Electronics.
Broadcom's actual expansion of 2 nm production would be significant for Samsung Electronics' foundry business in securing a large leading-edge process customer. This collaboration with Broadcom is also being viewed as an important opportunity to expand Samsung Foundry's customer portfolio.
Broadcom is a representative fabless company that designs AI accelerators and network semiconductors in-house. The importance of Broadcom's custom AI semiconductor business is also growing. Rising demand from hyperscalers for their own AI ASICs is also supporting this trend.
Demand for leading-edge processes is increasing, centered on AI accelerators and network semiconductors. As a result, the supply and demand environment for leading-edge foundry services is changing rapidly. Foreign media have reported that Samsung Electronics raised some new foundry order prices by up to 15% on the back of rising AI semiconductor demand.
Amid these changes, this collaboration aligns with Samsung Electronics' broader strategic expansion. Samsung Electronics is now fully launching an AI turnkey strategy that spans memory, foundry, and packaging.
The biggest significance of this collaboration lies in laying the groundwork for Samsung Electronics to expand from an HBM supplier into a comprehensive AI semiconductor solutions provider. Broadcom's possible expansion of 2 nm production is linked to Samsung Electronics' opportunity to secure a major leading-edge process customer and broaden its customer base, while also serving as a foundation for extending from a memory-centered model to an integrated AI supply system covering foundry and packaging as well.
Samsung Electronics' strategy is to combine HBM, leading-edge logic, foundry, and advanced packaging with Broadcom's design capabilities. Potential structures mentioned include Broadcom's AI accelerator and network semiconductor design, Samsung Electronics' HBM supply, Samsung Electronics' production of logic using processes below 2 nm, and single-system integration through advanced packaging.
A semiconductor industry expert said that, while citing TSMC's foundry strength and SK Hynix's HBM strength, Samsung Electronics appears to be trying to differentiate itself by using its memory and foundry capabilities at the same time. The expert also said that Samsung Electronics and Broadcom are likely to combine memory, foundry technologies, AI, and connectivity technologies, which is expected to improve the performance and power efficiency of next-generation AI infrastructure.
Source: TECHWORLD · Park Gyu-chan
Original: https://www.epnc.co.kr/news/articleView.html?idxno=406372
References
This article was produced with the help of an automated content generation algorithm.
Source: TECHWORLD
View originalThis article was summarized and organized by BizCrush based on the original article from TECHWORLD. For exact quotations and full details, please refer to the original article.