Chips engineer handling driving.
Sinology Moment | Getti images
With American limiting China from the purchase of advanced semiconductors used in the development of artificial intelligence, Beijing sets hopes to domestic alternatives such as Huawei.
The task is made more challenging that American curbs not only inhibit Chinese approach to the most advanced chips in the world, but also limit the available technologies from vital to creating AI Chip ecosystems.
Those constraints include the entire chain of semiconductor values, ranging from designer and production equipment used for production and chips to supported elements such as memory chips.
Beijing was mobilized tens of billions Dollars to try to fulfill those gaps, but even though it was able to “roughly force” its way into a breakthrough, still has a long way to go, according to experts.
“American Export Control to Advanced Nvidia and Chips had industry, and at the same time domestic companies, Paul Triolo, partner, and a senior Vice President for China on the advisory company DGA-Albright Stonebridge Group.
Here’s how China agrees against the rest of the world in four key segments needed to build ai chips.
Ai Chip Design
Nvidia is considered the leadership of the World AI Chip, but it is important to understand that this does not actually produce physical chips used for AI training and computing.
Instead, the company designs AI chips or more precisely, graphic processing. The orders of patented GPU companies are then sent to Chip Lisks – manufacturers who specialize in mass production of semiconductor products of other companies.
Although American competitors such as AMD and BroadCom offer different alternatives, GPU designs from NVIDIA widely recognized as an industry standard. The demand for Nvidia chips is so strong that Chinese customers continued to buy any of the chips of the company that can be caught.
But Nvidia fights with Washington tightening restrictions. The company has discovered in april which Additional Ivike It prevented from selling its H20 processor to Chinese clients.
NVIDIA’S H20 was a less sophisticated version of its H100 processor, designed specifically for the skirt of previous export controls. However, experts say it was still more advanced than anything available in the country. But China hopes to change that.
In response to restrictions, more Chinese semiconductor players enters and processor arena. They included a wide range of upstart, such as Enflame technology and technology dial, wanting to be soaked Billions of dollars in the GPU request that Nvidia left.
But no Chinese company seems closer to providing a true alternative to Nvidia from Huawei’s design chips, Hisilicon.
The most advanced GPU of Huawei in mass production is its rise 910b. The next generation was 910C was allegedly The mass shipments are expected to start in May, although updates did not appear.
Dylan Patel, Founder, General Manager and Main Analyst at Semializa, said CNBC that CNBCs, while Chips of Lips remain behind Nvidia, show that Huawei achieves significant progress.
“Compared to Nvidia chips with limited limited, performance between Huawei and H20 is less than a full generation. Huawei is not far behind Nvidia’s product,” Patel said, “Patel said.
He added that the 910B was two years after Nvidia since last year, while the rise was 910c only a year behind.
But although this indicates that Chinese GPU designers have made great steps, design is just one aspect that stands in the way of creating a competitive AI Chip ecosystem.
Ai Chip Making
To produce its GPU, Nvidia relies on TSMC, the world’s largest contractual liver chips, which produces most advanced world chips.
TSMC is in line with American chips controls and is also prohibited from taking any account chips from companies in the American trading blacklist. Huawei was put on the list in 2019. Years.
This led to Chinese Chip designers like Huawei for the enrollment of local left chips, most of whom is Smit.
Lower is far behind the TSMC – It is officially known to be produced 7-nanometer Chips, require less progress technology than TSMC’s 3-nanometer production. Less sizes of nanometers lead to higher power and chip efficiency.
There are signs that Smic has made progress. The company is suspected of behind a 5-nanometer 5G chip for Huawei’s Mate 60 Pro, who had angrily confidence In American controls of the Chip in 2023. the company, however, has a long way before it can be massaged by advanced GPU in an economical way.
According to an independent analyst and technology, Rai Wang, Smic’s famous operational capacity is a TSMC dwarf.
“Huawei is a very good chip design company, but they are still without good homemade chips,” Wang said, noting that Huawei said allegedly working on one’s own capabilities of making.
But the lack of key production equipment stands in the path of both companies.
Advanced Chip Equipment
Leave the ability to meet Huawei’s GPU requirements is limited by a well-known export control problem, but in this case from the Netherlands.
While Netherlands may not have any prominent semiconductor or manufacturer designers, it’s home AsmlLeading in the world The leading provider of advanced chipmake equipment – machines used by light or electronic beams for transferring complex patterns on silicon waffles, forming the basis of microchip.
In line with American export controls, the country agreed to block the sale of ASML’s state-of-the-art ultraviolet (EU) machine. Tools are crucial for GPU advancement on a scale and economical.
The EUV is the most significant obstacle to Chinese production of advanced chips, according to Jeff Koch, analyst on semi-inventive. “They have most other tools available, but lithography limits their ability to scalle at 3NM and below the process,” CNBC said.
Smic found Methods for working around Lithographs on lithographs using ASML less advanced deep ultraviolet lithographic systems, which have seen relatively smaller restrictions.
Through this “rough forcing”, the production of chips in 7 Nm is feasible, but the yields are not good, and the strategy is likely to reach its limit, adding that “in current sink cannot produce sufficiently domestic accelerators.”
Sicarrier Technologies, a Chinese company working on lithography technology, has allegedly associated with huawei.
But imitation of existing tool lithographs could last year, if not decades, Koch said. Instead, China is likely to investigate other technologies and various lithographic techniques for pushing innovation, not imitations, he added.
AI Memory components
While the GPU is often identified as the most critical component in AI computing, they are far from the only ones. To manage AI training and computation, GPUs must work together with memory chips that can store data in the wider “chipset”.
In AI applications, a specific memory type known as HBM has become an industrial standard. South Korea SK HiK He took the leadership industry in HBM. Other fields on the field include Samsung and US-seat Micron.
“High memory of the scope at this stage and progress became necessary for training and launch and model,” Vang analyst said.
As with the Netherlands, South Korea cooperates with the limitation of the chip and began to comply with respect Fresh edge For sale certain HBM memory chips in China in December.
In response, Chinese memory memory memory technology, or CKSMT, in partnership with Chip packaging and testing of Tongfu Microelectronics, is in the early stages of HBM production, according to A Report Reuters.
According to Vang, CKSMT is expected to be three to four years behind global leaders in HBM development, although it faces large blockages of the road, including export control equipment.
Semianalysis estimated in April that CKSMT remained a year away from the reproduction of any reasonable volume.
The Chinese Foundry Vuhan Xinkin Semiconductor Manufacture is allegedly Building a factory for the production of HBM waffles. A report From the SCMP, he said Huawei technologies associated with the company in the production of HBM chips, although companies did not confirm partnership.
Huawei leaned to HBM supplies from suppliers like Samsung for use in their Ascend 910C and Processor, Semializ said in April reportNoting that although the chip is designed in the country, it still rely on foreign products obtained before or despite the limitations.
“Whether HBM from Samsung, waffle from TSMC or equipment from America, the Netherlands and Japan, there is great reliance on foreign countries,” Semiaza said.
