New Chinese AI Chips and their Huge Problems

Anastasi In Tech
16 Feb 202414:13

TLDRThe video discusses the semiconductor technology competition between the US and China, focusing on China's advancements in AI chips despite restrictions. It highlights the challenges in design, manufacturing, and software stack development, with companies like Huawei and Biren making significant strides. The script emphasizes China's efforts to overcome these obstacles, including the development of domestic EDA software and high-bandwidth memory, as well as the potential for these companies to establish a strong presence in the AI hardware market.


  • 🌐 The US and China are in economic competition, with semiconductor technology playing a crucial role.
  • 🚫 Restrictions have been imposed on China's access to critical semiconductor manufacturing equipment and advanced AI chips.
  • πŸ’‘ China's domestic chip development is stimulated by these restrictions, leading to companies like Huawei and Alibaba creating their own AI chips.
  • πŸ› οΈ Chinese companies rely on EDA tools from American companies like Synopsys and Cadence, but Huawei is developing its own EDA software.
  • πŸ”„ Huawei's 910b GPU is designed to compete with Nvidia's A100 GPU, with official specs showing higher performance.
  • 🏭 Manufacturing challenges exist due to limited capacity at SMIC, which is the primary fabricator for Chinese AI chips.
  • πŸ”„ SMIC is exploring alternative lithography techniques to overcome the limitations of their current DUV machines.
  • πŸ€– Chinese AI chip manufacturers are also focusing on developing a software stack to optimize the performance and utilization of their hardware.
  • πŸ’» Companies like Biren and MetaX are working on compatibility with established platforms or creating their own software solutions.
  • 🌟 Despite challenges, there is optimism that China will become self-sufficient in AI chip manufacturing and develop advanced packaging and manufacturing capabilities.

Q & A

  • What are the two major parts of the US restrictions on China's access to semiconductor technology?

    -The US restrictions consist of limiting access to critical semiconductor manufacturing equipment from companies like ASML, LAM Research, KLA, and others, and directly affecting the sales of advanced AI chips to China.

  • How have China's domestic developments in semiconductor technology been stimulated?

    -The restrictions have stimulated domestic developments by forcing Chinese companies and startups, including Huawei, Alibaba, and MetaX, to innovate and develop their own semiconductor technologies and AI chips.

  • What challenges are Chinese companies facing in developing new chips?

    -Chinese companies face challenges related to the design, manufacturing, and software stack for their new chips, including the reliance on American electronic design automation (EDA) tools and difficulties in achieving high manufacturing yields.

  • What is the significance of EDA tools in chip design, and how are Chinese companies managing without complete access to them?

    -EDA tools are critical for placing and routing millions of transistors in chips efficiently. Despite partial restrictions, Chinese companies are still using American EDA tools for designing chips. Huawei is also developing its own in-house EDA software, which is currently capable of handling chip layouts down to 14 nanometers.

  • How does the Huawei 910b GPU compare to Nvidia's offerings in the Chinese AI market?

    -The Huawei 910b GPU, fabricated by SMIC in 7 nm, is theoretically more powerful than Nvidia's H20 GPU, offering 512 teraflops at 8-bit precision compared to the H20's 296 teraflops.

  • What are the implications of Huawei prioritizing the fabrication of AI GPUs over mobile chips?

    -Huawei's decision to prioritize AI GPU fabrication reflects a strategic focus on capturing a significant share of the AI hardware market, even at the expense of its mobile chip production, highlighting the high demand and potential profitability of AI chips.

  • How is SMIC managing to fabricate 7 nm and 5 nm chips without access to EUV machines from ASML?

    -SMIC is using older immersion deep ultraviolet (DUV) machines from ASML and employing multi-patterning techniques to fabricate 7 nm and 5 nm chips, although at a lower yield and higher cost per die.

  • What unique approaches are Chinese companies exploring to overcome the limitations of current lithography techniques?

    -Chinese companies are investigating alternative lithography techniques, including the use of particle accelerators instead of EUV machines, to potentially bypass the limitations of DUV machines and advance beyond 5 nanometers.

  • Why is software stack development critical for the success of new AI chips, and how are Chinese companies addressing this?

    -A robust software stack is essential for efficiently utilizing a chip's architecture and distributing workloads. Chinese companies, including Huawei and startups like MetaX and Biren, are either making their hardware compatible with existing platforms like Nvidia's CUDA or investing in developing their own software stacks.

  • What are the prospects for Chinese semiconductor self-sufficiency and global competitiveness in the next 5 years?

    -Given the advancements and investments in chip design, manufacturing capabilities, and software ecosystem, it's anticipated that China will make significant progress toward semiconductor self-sufficiency and competitiveness, especially in the AI hardware sector, within the next 5 years.



🌐 US-China Tech Rivalry and Domestic AI Chip Development

This paragraph discusses the impact of US-China economic competition on semiconductor technology, particularly in the context of restrictions imposed on China's access to critical tech. It highlights the two main areas of restriction: semiconductor manufacturing equipment and sales of advanced AI chips to China. The narrative then shifts to focus on China's domestic advancements in AI chip technology, mentioning companies like Huawei and Alibaba, and the challenges they face in design, manufacturing, and software stack development. The paragraph also touches on Huawei's efforts to develop its own EDA software and the competitive AI GPU, Huawei 910b, which is said to outperform Nvidia's H20 GPU based on official specs.


🏭 Huawei's Entry into AI Hardware and Manufacturing Challenges

This paragraph delves into Huawei's strategic move into the AI hardware market, emphasizing their prioritization of AI GPU fabrication over mobile chips. It outlines the technical aspects of Huawei's 7-nanometer AI chip fabrication by SMIC, contrasting it with TSMC's high-yield EUV machine fabrication process. The paragraph also addresses the manufacturing bottleneck faced by SMIC and their efforts to scale up production, as well as exploring alternative lithography techniques. Additionally, it mentions the importance of high-bandwidth memory in AI chip manufacturing and China's efforts to develop this technology.


πŸš€ Chinese AI Chip Market: Innovations, Challenges, and Future Prospects

The final paragraph provides an overview of the competitive landscape in China's AI chip market, highlighting the efforts of companies like Biren and MetaX in developing their hardware and software stacks. It discusses Biren's BR 100 GPU, which was built on TSMC's 7-nanometer process and utilized advanced packaging technology. The paragraph also touches on the impact of US export regulations on Biren's manufacturing and the need for domestic alternatives. It mentions other Chinese companies like Moretech and Hygen Technology, which are developing GPUs for AI acceleration, and their compatibility with Nvidia's CUDA platform. The summary concludes with a forward-looking perspective on China's potential to develop a self-sufficient AI chip manufacturing industry.



πŸ’‘Semiconductor Technology

Semiconductor technology is foundational to modern electronics, including computers, smartphones, and advanced AI chips. The video highlights its significance not just in technological innovation but also in the economic competition between the US and China, emphasizing the strategic importance of semiconductors in global trade and national security. Restrictions on China's access to this technology have spurred domestic developments, illustrating the critical role semiconductors play in both technological and geopolitical landscapes.

πŸ’‘AI Chips

AI chips, designed specifically to efficiently process artificial intelligence tasks, are central to the video's discussion. The script mentions China's advancements in developing AI chips that rival Nvidia's GPUs, highlighting the competitive landscape in the AI hardware market. AI chips' relevance is underscored by their application in various domains, including large language model training, showcasing their importance in advancing AI technologies.

πŸ’‘EDA Tools

Electronic Design Automation (EDA) tools are software platforms used to design electronic systems such as integrated circuits and printed circuit boards. The video explains how Chinese companies use American EDA tools for chip design, despite geopolitical tensions. EDA tools' critical role in chip design, enabling the placement and routing of millions of transistors, underscores the complexity and the high-tech artistry involved in semiconductor manufacturing.

πŸ’‘Huawei 910b GPU

The Huawei 910b GPU is presented as a competitive AI GPU developed in China, comparable to Nvidia's A100 GPU but fabricated domestically by SMIC at 7 nm. This GPU's development reflects China's efforts to build domestic capabilities in semiconductor manufacturing and AI technologies. The video's comparison of the 910b GPU's performance with Nvidia's offerings highlights the competitive nature of the global AI hardware market.


SMIC (Semiconductor Manufacturing International Corporation) is China's leading semiconductor foundry, mentioned in the video as fabricating Huawei's 910b GPU. The video discusses SMIC's challenges and innovations in producing advanced chips with older equipment due to restrictions on accessing newer technologies. SMIC's role illustrates the resilience and adaptability of China's semiconductor industry in the face of technological embargoes.

πŸ’‘High Bandwidth Memory (HBM)

High Bandwidth Memory (HBM) is a type of memory architecture used in high-performance computing to allow faster data transfer between the GPU and memory. The video mentions China's efforts to develop HBM production capabilities as crucial for the performance of AI chips. HBM's importance in enhancing GPU performance, particularly for AI applications, is highlighted as a key factor in the competitive development of semiconductor technologies.

πŸ’‘Software Stack

The software stack for AI hardware refers to the layers of software that enable the efficient use of AI chips' architecture, including operating systems, drivers, and development tools like CUDA for Nvidia GPUs. The video discusses the challenge Chinese companies face in building a competitive software stack, emphasizing its critical role in leveraging hardware capabilities for AI applications and the strategic importance of software in the AI hardware ecosystem.

πŸ’‘Nvidia CUDA

Nvidia's CUDA is a parallel computing platform and programming model that allows developers to use Nvidia GPUs for general purpose processing. The video mentions Chinese companies attempting to make their hardware compatible with Nvidia's CUDA platform, illustrating the widespread adoption and significance of CUDA in the development and optimization of AI applications, and the strategic moves by Chinese firms to ensure compatibility and ease of transition for developers.

πŸ’‘Packaging Technology

In semiconductor manufacturing, packaging technology refers to the methods used to encase semiconductor chips and connect them to external devices. The video discusses advanced packaging technologies like TSMC's CoWoS (Chip on Wafer on Substrate), highlighting their importance in integrating multiple dies and memory into a single package to enhance performance. This technology's relevance is underscored in the context of China's challenges and innovations in semiconductor manufacturing.

πŸ’‘Generative AI

Generative AI refers to algorithms capable of creating content, such as text, images, and music, that resembles human-generated content. The video highlights the booming interest in generative AI technologies, mentioning Nvidia's strategic positioning through the development of efficient processing for deep learning algorithms. Generative AI's significance is illustrated through its potential for innovation and the competition it drives in the development of specialized AI hardware.


Semiconductor technology is central to economic competition between the US and China, leading to restrictions on China's access to critical tech.

Chinese companies like Huawei, Alibaba, and startups are developing advanced AI chips to reduce dependency on foreign technology.

Huawei's in-house EDA software is in pilot phase, capable of handling chip layouts down to 14 nanometers.

Huawei's 910b GPU is comparable to Nvidia's A100 GPU in performance, with 512 TeraFLOPS at 8-bit precision.

SMIC's limited capacity of 25-30,000 wafers per month constrains the supply of AI GPUs in China.

Chinese companies face challenges in designing chips, manufacturing, and developing software stacks for new AI hardware.

Building a software stack from scratch is a significant challenge for Chinese companies to fully utilize their AI hardware's potential.

Biren's BR100 GPU, built on TSMC's 7nm process, has faced manufacturing suspension due to export regulations.

Chinese companies are exploring alternative lithography techniques to overcome limitations of DUV machines.

CXMT is ramping up production of high-band memory to address one of the bottlenecks in GPU performance.

MetaX is working on making their hardware compatible with Nvidia's CA platform, while others develop their own software stacks.

Hygen Technology's Shensuan 2 GPU is reportedly compatible with Nvidia's CUDA platform, offering an easy transition for Nvidia customers.

Intelifusion's Deepedge10 chip claims compatibility with Nvidia's H20 GPU, though detailed specs are not available.

Chinese startups are pitching hardware to companies even before prototypes are ready, showing the growing interest in domestic AI chip development.

In 5 years, China may have developed the manufacturing process to create their own domestically designed and manufactured AI hardware.

Nvidia's long-term strategy and software stack development, including CUDA and Tensor Cores, has solidified its position as an AI hardware leader.

More Chinese companies are expected to overcome manufacturing challenges and contribute to the AI hardware landscape.