Huawei Technologies Co., a key player in China’s tech industry, faces challenges as U.S. sanctions obstruct its ability to innovate in the chip manufacturing space. The company, in its pursuit of AI competitiveness, is working on the development of new Ascend processors that are limited to 7-nanometer (nm) technology.
These limitations hinder Huawei’s capability to rival Nvidia Corp.’s leading-edge AI accelerators, which use more advanced architectures [source].
Sanctions Limit Access to Critical Technology
The U.S. government’s export restrictions, in place since 2019 and tightened over subsequent years, have severely impacted Huawei’s supply chain. One major consequence of these sanctions is Huawei’s exclusion from accessing extreme ultraviolet (EUV) lithography systems, produced exclusively by ASML Holding NV.
EUV is vital for manufacturing semiconductors with node sizes below 5nm, a standard in high-performance chips. As a result, Huawei is restricted to deep ultraviolet (DUV) lithography, which can only support the production of chips at 7nm and larger.
This technological gap has had a tangible effect on Huawei’s output. Before the trade sanctions, Taiwan Semiconductor Manufacturing Co. (TSMC) was responsible for producing Huawei’s top-tier chips.
However, U.S. regulations cut off these ties in 2020, compelling Huawei to shift production to China’s Semiconductor Manufacturing International Corporation (SMIC), backed by a $47.5 Billion government fund. SMIC, though essential for Huawei’s operations, cannot match TSMC’s capability for mass-producing advanced nodes.
Ascend Chips: Huawei’s AI Competitors
Huawei’s Ascend series, including the upcoming 910C and 920 models, seeks to establish a foothold in the AI chip market. These processors use the 7nm node, a technological step that, while sufficient for domestic applications, falls behind Nvidia’s H100 and the anticipated Blackwell GPUs.
The performance gap becomes even more apparent when considering Nvidia’s use of EUV technology, which allows for higher transistor density and greater energy efficiency.
In AI model training, Huawei’s current Ascend 910B chip has shown promise. According to internal tests by Chinese tech firms, the 910B chip reportedly achieves around 80% of Nvidia’s A100 performance and occasionally surpasses it by 20% in specific scenarios. This makes the chip competitive within China, especially given the halted TSMC shipments due to sanctions. Compared to Nvidia´s newest Blackwell chips, however, Huawei’s Ascend chips cannot compete.
The shift to SMIC has brought challenges of its own. Although SMIC has made progress in chip fabrication, it lacks the capability to scale up 7nm production in the same way that global leaders like TSMC do.
This impacts Huawei’s ability to meet both domestic and international demand for high-performance processors. The reliance on a domestic supplier reflects a larger trend in China’s push for self-reliance in tech, where investment in local chip production has been a priority.
|
Related: |
Workarounds and Production Challenges
To counteract the limitations of DUV lithography, some manufacturers, including Huawei, have employed techniques like quadruple patterning. This complex process involves multiple exposures to create finer chip details, which can mimic the effects of EUV lithography but comes at the cost of higher expenses and lower efficiency.
Quadruple patterning helps push the boundaries of DUV technology, yet its labor-intensive nature and scalability issues mean that it’s not a sustainable substitute for EUV. The high production costs associated with this method make it an interim solution rather than a long-term strategy.
Technological Self-Sufficiency is Hard
Huawei’s challenges echo China’s broader efforts to achieve self-sufficiency in semiconductor manufacturing. The Chinese government has invested billions into fostering a domestic tech ecosystem capable of producing advanced chips. However, without EUV lithography, these efforts face substantial obstacles. The inability to access cutting-edge tools means China’s semiconductor sector struggles to close the gap with global leaders like Nvidia and TSMC.
Huawei’s focus on 7nm chips underlines the difficulties in balancing ambition with existing constraints. The company remains a significant force in China’s AI infrastructure, with companies like Baidu and iFlytek adopting its AI accelerators for their data centers. Yet, the technological disparity limits its competitive reach on a global scale.
