Chinese chips are artificially slowed to avoid US export regulations

Once upon a time, countries protected their domestic industries with tariffs on imports. This gave the country a price advantage over companies operating overseas, but the practice has gone out of fashion somewhat in recent decades.

Nowadays, governments are altogether more creative, using fanciful export controls to protect their interests. To this end, the United States has adopted an export restriction on high-powered computing devices. In response, Chinese designers are trying to artificially slow down their hardware to dodge these rules.

Companies like NVIDIA and AMD had to rework some products to comply with US regulations against Chinese exports. The A100 Data Center GPU was banned from export, so NVIDIA developed the lower-spec A800 instead. Credit: NVIDIA press site

The new export rules come as the US government grapples with the rise of China’s military, both in size and technological sophistication. Regulations limit the export of advanced integrated circuits, but the regulations do not stop there. Tooling, software and other manufacturing equipment required to manufacture such material are also subject to the rules. The often stated goal is to slow or stop the development of advanced military devices that could be used by the Chinese government or sold to other countries. Alternatively, it could be presented as an attempt to preserve the advantage of existing players in the semiconductor market.

Chips capable of an “aggregate bi-directional transfer rate over all inputs and outputs” exceeding 600 GB/s, not including volatile memory, cannot be exported or re-exported to China, under the new rules. Advanced manufacturing tools used for electroplating, chemical vapor deposition and other chip production processes are also banned from export. Just to cover all bases, software packages for the design, manufacture, or use of these chips or associated hardware are also subject to the sanctions. Companies can apply for a license to export this material to China, however, as with most of these restrictions, there is a presumption that such licenses will be denied. Other restrictions apply to chips exceeding certain machine learning performance limits and powerful supercomputers.

Additionally, with respect to exports of items not subject to the above restrictions, “U.S. Persons” must have a license if the items will be used in the “development” or “production” of integrated circuits in China meeting to certain criteria. This includes chips that use a non-planar architecture or are manufactured at a technology node of 14 nm or less, as well as NAND memory with 128 layers or more and DRAM manufactured at a node of 18 nm or less. The category of “U.S. Persons” is also broad, including U.S. citizens, permanent residents, and corporations and legal entities established in the United States, even when operating overseas.

Chinese tech giant Alibaba and small startup Biren Technology have since found themselves struggling with the restrictions. Along with a variety of lesser-known Chinese chip companies, they have invested heavily in designing new chips for high-power computing applications. These include new chips to compete with GPUs from companies such as Nvidia and AMD, as well as processors for machine learning applications.

But Alibaba and Biren are factoryless, outsourcing the actual production stage. Many of these companies have their designs produced by Taiwan Semiconductor Manufacturing (TSMC), considered one of the world’s premier silicon foundries.

Some of the latest designs from these companies violate new export rules, in terms of data rates or other factors. Although they are intended to be produced in Taiwan, US export regulations still have an effect. This is because the vast majority of semiconductor fabs around the world rely on US-made equipment and software. If foreign factories started shipping such designs to China, they would quickly be cut off from the American equipment and software needed to work the installation. China is developing its own semiconductor production facilities, but they are currently years or decades behind the cutting edge of technology and therefore cannot produce such advanced designs.

Biren Technology touted its new data center GPUs as performing better than NVIDIA’s A100 offering. Since the latter is no longer legal to export to China, having a local replacement is essential. Credit: Biren Tech News Site

Biren Technology may push the limit with its BR100 GPU, aimed at machine learning applications. Early statements cite a figure of 640 GB/s, beyond the stated limit. Since then, the company’s website has listed the card’s bandwidth at various figures ranging from 512 GB/s to 448 GB/s. According to some researchers, the company could disable parts of the BR100 chip to exceed the limit, while potentially allowing it to re-enable it later.

Alibaba’s own efforts face similar problems. The company has been working on advanced machine learning chips for AI work on TSMC’s 5nm tech node. Apparently, the team plans to rework the designs to avoid issues with regulations, but this is an expensive exercise that would take several months and millions of dollars.

Engineers have complained that the rules are unclear, as there are different ways to calculate the bidirectional transfer rate. Anyway, many are already working to reduce the speed of processors in order to circumvent the rules. The key remains discreet, according to a source speaking to ArsTechnica. Some companies have released press materials about chips with transfer rates above regulations, alerting authorities to monitor shipments of such parts. In cases where a chip’s capabilities are not yet widely known, engineers have more potential to work with the factory to find a redesign that could circumvent regulations.

Part of a trend

This is not the first time that US export regulations have tried to clip the wings of Chinese tech companies. Huawei’s semiconductor arm, HiSilicon, broke a previous set of export rules in 2019. Initial sanctions were imposed on the company over backdoors allegedly found in Huawei’s communications equipment. These rules cut off Huawei’s access to software and hardware from companies like Intel, Google and Qualcomm. However, Huawei persevered with its own chips and apps, with Chinese buyers supporting the company’s sales as international business plummeted. HiSilicon quickly became the leading smartphone chipset supplier in China.

From there, the US government moved up the chain, making it illegal to supply equipment to HiSilicon for its semiconductor factories. This was the death knell for the company’s flagship smartphone chipsets, and it was quickly overtaken by other companies in the market.

Access to the world’s best silicon fabs is essential for building high-performance chips. Currently, the United States holds the key, thanks to a monopoly on the supply of advanced manufacturing equipment and design software. China will continue to develop its own capability, just as it has continued to produce its own jet engines and other technologies. However, in the same way, it is a long and slow road to travel, and an expensive one at that.

Banner image: “Silicon Wafer” by Enrique Jiménez