#075 How China is Rising in the AI Chip Packaging Industry, China vs NVIDIA: The Battle for AI Chip Supremacy
Fresh & Hot curated AI happenings in one snack. Never miss a byte š
This snack byte will take approx 5 minutes to consume.
AI BYTE #1 š¢: How China is Rising in the AI Chip Packaging Industry
ā The global competition for AI supremacy is not only about who can design the most advanced chips, but also about who can package them together in the most efficient way.
Chip packaging is a crucial step that improves the performance and functionality of chips by connecting them with other components, such as memory and sensors.
China, which has been lagging behind the U.S. and Taiwan in chip design and fabrication, is aiming to catch up by investing heavily in chip packaging technology, especially for AI applications.
China has a strong domestic demand for AI chips, driven by its ambitious plans to become a world leader in AI by 2030.
According to a report by IDC and Inspur, Chinaās AI server market will grow 82.5% to US$9.1 billion yuan this year, and reach US$13.4 billion yuan in 2027. The compound annual growth rate of Chinaās AI computing power will be 33.9% in 2022-2027, compared with the 16.6% annual growth of traditional computing power.
However, Chinaās AI industry has been facing a bottleneck in accessing high-end chips, due to the U.S. export controls that block leading U.S. chip designers, such as Nvidia and AMD, from selling their chips to China.
To overcome this challenge, China has been developing its own AI chip makers, such as Huawei Technologies, Moore Threads, Cambrian and Stream Computing. Their total shipment reached 1.34 million units in 2023, up 22.5% from 2022, according to IDC.
However, these chip makers still rely on foreign suppliers for key technologies, such as electronic design automation software and semiconductor manufacturing equipment. Moreover, their chip designs are not yet competitive with Nvidia and AMD in terms of performance and power efficiency.
Therefore, China is looking for new ways to boost its AI chip capabilities, and one of them is to focus on chip packaging technology. Chip packaging can enhance the performance of chips by reducing the distance and latency between different components, increasing the bandwidth and data transfer speed, and enabling heterogeneous integration of different types of chips.
For example, TSMC, the worldās leading chip foundry, has developed a chip packaging technology called CoWoS, which stands for Chip on Wafer on Substrate, that bundles logic and memory chips and improves data transmission speeds between them. This technology is widely used for AI and supercomputing applications, and TSMC plans to double its capacity in 2024.
China has already established a strong presence in the global chip packaging and testing market, with a market share of about 30%. The worldās leading packaging companies, such as ASE, Amkor and JCET, have set up factories in China to tap into the local demand and supply chain.
China also has its own packaging companies, such as Huatian Technology and Nantong Fujitsu, that are investing in advanced packaging technologies, such as fan-out wafer-level packaging and 3D stacking.
Chinaās chip packaging industry is also supported by the governmentās policies and subsidies, such as the National Integrated Circuit Industry Investment Fund and the CHIPS Act.
However, China still faces some challenges and risks in its chip packaging ambitions. One of them is the lack of standardization and compatibility among different chip packaging technologies and platforms.
This makes it difficult for chip makers and users to integrate and optimize their chips for various AI applications. Another challenge is the potential U.S. sanctions on chip packaging technologies, as the U.S. seeks to maintain its technological edge and national security interests.
The U.S. has already imposed export controls on some chip packaging equipment and components, such as extreme ultraviolet lithography machines and etching gases, that are essential for advanced chip packaging.
Therefore, China needs to develop its own chip packaging ecosystem and innovation capabilities, and collaborate with other countries and regions, such as Europe and Southeast Asia, to diversify its sources and markets.
China also needs to balance its chip packaging development with its chip design and fabrication efforts, as all three aspects are interrelated and complementary.
AI BYTE #2 š¢: China vs NVIDIA: The Battle for AI Chip Supremacy
ā China is increasingly adopting the RISC-V architecture to design AI chips as part of its strategy to reduce reliance on Western chip technology.
The RISC-V architecture is an open-source blueprint for building chips, which provides a faster and cheaper alternative to design and manufacture chips without relying on Western technologies. The Chinese government has shown seriousness about funding RISC-V initiatives.
Several Chinese companies are at the forefront of this innovation. For instance, the Chinese Academy of Sciences (CAS) is developing an advanced RISC-V chip called XiangShan-v3 in collaboration with top Chinese companies, including Alibaba, Tencent, and ZTE.
When it comes to comparison with NVIDIAās chips, itās important to note that NVIDIA is a leading player in the AI chip market. However, due to U.S. export controls, NVIDIA has had to develop China-specific chips with lower computing capabilities.
For instance, NVIDIAās H20 AI chip, which was expected to hit the market in November, has been delayed until the first quarter of 2024. This chip can only deliver a maximum compute performance of 4800 TOPS over the PCIe interface, much lower than the 10,000 TOPS that the A800 and the H800 offers.
On the other hand, scientists at Beijingās Tsinghua University have developed a new kind of computer chip called an all-Analog Chip Combining Electronic and Light computing (ACCEL), which is more efficient than NVIDIAās A100 chip.
ACCEL is fast and capable of performing 4.6 quadrillion operations per second, in comparison to NVIDIAās A100, which delivers 0.312 quadrillion operations per second of deep learning performance.
In conclusion, while NVIDIA continues to be a dominant player in the global AI chip market, Chinese companies are making significant strides in developing their own AI chips using the RISC-V architecture and other innovative technologies.