Your browser does not support JavaScript!

China's Race for HBM Self-Sufficiency

General Report November 4, 2024
goover

TABLE OF CONTENTS

  1. Summary
  2. Current Landscape of HBM Production
  3. Chinese Consortium for HBM Production
  4. Technological Challenges and Development Goals
  5. Market Implications and Future Prospects
  6. Conclusion

1. Summary

  • China is determined to establish domestic high bandwidth memory (HBM) production by 2026, spearheaded by Huawei Technologies and a consortium of local semiconductor firms. This initiative is driven by US sanctions that have restricted China's access to advanced semiconductor technologies, compelling the nation to accelerate its self-sufficiency efforts. The current global HBM market is led by South Korean giants SK hynix and Samsung, alongside Micron Electronics from the US. While Huawei and other Chinese firms face significant technological hurdles, government support and collaborations among domestic companies are paving the way for advancement in HBM production. The report discusses the potential impact of these developments on the AI chip market and China's semiconductor industry.

2. Current Landscape of HBM Production

  • 2-1. Overview of Global HBM Market

  • The global High Bandwidth Memory (HBM) market is currently dominated by South Korean companies SK hynix and Samsung, along with significant contributions from the U.S. firm Micron Electronics. These companies are ahead in HBM production technology with ongoing developments aiming for the fifth-generation HBM, referred to as HBM3E, expected to be available to customers by 2024. China's presence in the HBM market is notably minimal, with domestic firms like Huawei striving to establish their footprint after years of being technologically disadvantaged.

  • 2-2. Leading HBM Manufacturers and Their Technologies

  • SK hynix and Samsung are the leading manufacturers of HBM, with notable technological advancements that set industry standards. Micron Electronics also plays a crucial role in the market. In China, CXMT and Wuhan Xinxin are emerging players, with CXMT reportedly having developed sample HBM chips and Wuhan Xinxin building a factory for production, anticipated to produce 3,000 12-inch HBM wafers monthly. Huawei is aiming to enter the HBM sector by collaborating with other domestic semiconductor firms and focusing on developing HBM technologies to enhance China’s capabilities in the AI chip market.

  • 2-3. Impact of US Sanctions on HBM Production

  • US sanctions have significantly impacted the ability of Chinese chip manufacturers to advance technologically in the HBM sector. Restrictions on the export of advanced chipsets to China have stalled progress, compelling domestic firms to accelerate their efforts towards self-sufficiency in HBM production. As a result of these challenges, collaborations among Chinese companies and between private firms and international semiconductor equipment suppliers are becoming increasingly vital for overcoming technological hurdles.

3. Chinese Consortium for HBM Production

  • 3-1. Formation of the Huawei-led Consortium

  • A consortium comprising several Chinese semiconductor firms, led by Huawei, is currently working towards establishing domestic production of high bandwidth memory (HBM) by 2026. The initiative is part of China's broader strategy to create home-grown alternatives to foreign technologies, particularly those from Nvidia in the AI chip sector. The consortium's formation is also a response to the challenges posed by US sanctions which have affected several of its key members.

  • 3-2. Government Support and Funding for HBM Development

  • The Chinese government has shown significant support for the consortium's endeavors, providing funding and logistical assistance as part of its efforts to bolster the domestic semiconductor industry. The government aims to reduce China's reliance on foreign technology by enhancing local capabilities in high-end semiconductor production, which includes HBM chips critical for AI applications.

  • 3-3. Collaboration with Local Semiconductor Firms

  • The consortium works closely with local semiconductor producers and packaging technology developers. Notable participants include Fujian Jinhua Integrated Circuit, which, despite facing US sanctions, is a key player in the project along with Huawei. The collaboration aims to tailor HBM chips specifically for Huawei-designed AI processors, demonstrating an effort to align local chip manufacturing with the requirements of the domestic AI landscape.

4. Technological Challenges and Development Goals

  • 4-1. Targeting HBM2 Production by 2026

  • Chinese companies, led by Huawei, are targeting the production of HBM2 chips by 2026. This endeavor is a part of China's larger strategy to establish domestic semiconductor manufacturing capabilities in response to US sanctions. The consortium includes Huawei and Fujian Jinhua Integrated Circuit, focusing on producing high-bandwidth memory chips that are essential for advanced graphics processing units.

  • 4-2. Technological Hurdles in HBM Manufacturing

  • The manufacturing of high-bandwidth memory (HBM) poses several technological hurdles. Key challenges include complex processes involved in wafer-level packaging, testing technology, and design compatibility. The integration of HBM in AI chips, particularly in technologies such as CoWoS (Chip on Wafer on Substrate), requires advanced processes including Through-Silicon Via (TSV) and Redistribution Layer (RDL) techniques. Notably, TSV represents approximately 30% of the overall 3D packaging cost for HBM.

  • 4-3. Comparison of Chinese and Global HBM Technologies

  • Currently, Chinese firms face significant challenges in comparing their HBM technologies to those of established global players. Companies in South Korea, such as SK Hynix and Samsung Electronics, have been at the forefront of HBM development for nearly a decade, having developed HBM2 technology as early as 2016. Observers suggest that despite significant governmental investment, Chinese firms may struggle to close the technological gap, which has accumulated over several years of R&D by their global counterparts.

5. Market Implications and Future Prospects

  • 5-1. Potential for Domestic AI Chip Competitiveness

  • This section investigates the potential for domestic AI chip competitiveness as Chinese firms, led by Huawei, seek to produce high bandwidth memory (HBM) semiconductors. The collaborative effort with various Chinese chip companies and government support aims to create home-grown alternatives to Nvidia's AI chips. Notably, the project includes Fujian Jinhua Integrated Circuit, a firm already under US sanctions, highlighting the strategic importance of this initiative as part of China's broader efforts to overcome technological reliance on foreign suppliers.

  • 5-2. Analysis of Supply Chain Dynamics

  • The supply chain dynamics in the production of HBM are becoming increasingly critical as Chinese chipmakers, particularly CXMT and Wuhan Xinxin, navigate challenges posed by US sanctions. CXMT has developed sample HBM chips and is collaborating with Tongfu Microelectronics to advance packaging and testing capabilities. Wuhan Xinxin is establishing a factory capable of producing 3,000 12-inch HBM wafers per month, reflecting significant investment in enhancing local manufacturing capacity. These developments indicate a shift in supply chain reliance towards domestic resources and partnerships, particularly with semiconductor equipment firms in South Korea and Japan.

  • 5-3. Forecast of HBM Market Development

  • The HBM market is poised for evolution as Chinese firms target production capabilities by 2026. Currently, the global HBM market is dominated by SK Hynix, Samsung, and Micron Electronics, who are planning to deliver fifth-generation HBM, or HBM3E, by 2024. In contrast, China's focus remains on HBM2 technology, demonstrating a decade of technological lag. Despite this, the advancements being made by CXMT are viewed as promising, providing a potential pathway for China to elevate its position within the memory chip sector, particularly for AI applications.

Conclusion

  • Huawei Technologies’ leadership, alongside several Chinese semiconductor firms, signals a strong commitment to overcoming the significant challenges posed by US sanctions and the technological gap in high bandwidth memory (HBM) production. Current global leaders in the HBM market, like SK hynix, Samsung, and Micron Electronics, remain ahead with advanced technologies, while China’s focus on HBM2 reflects its ongoing struggle to catch up. However, the strategic collaboration within the domestic industry, bolstered by governmental support, is crucial for China’s vision of reducing dependency on foreign technology for AI chip manufacturing. This progress, despite the hurdles, could redefine the AI chip competitive terrain in China, offering a strategic advantage in addressing growing technological demands amid global tensions. Yet, the report acknowledges the significant hurdles still present, given the existing technological gap and suggests ongoing government investment and collaboration as pivotal to achieving these ambitious goals. The future of Chinese competitiveness in the global semiconductor market hinges on successfully meeting the 2026 production targets and further developing HBM technology for AI applications.

Glossary

  • Huawei Technologies [Company]: Huawei is a leading global technology company that plays a significant role in the development of telecommunications and semiconductor technologies. In this report, Huawei's initiatives to spearhead HBM production in China are critical as they represent a strategic shift towards self-sufficiency in advanced semiconductor manufacturing, especially in the context of increasing AI demands and geopolitical constraints.
  • High Bandwidth Memory (HBM) [Technology]: HBM is a high-performance memory technology used in advanced computing applications, particularly in AI and graphics processing. Its significance lies in its ability to provide faster data transfer rates compared to traditional memory types, making it essential for AI processors. The development of HBM technology in China is crucial for enhancing the competitiveness of local AI solutions against global leaders.

Source Documents