With the world in the state it is in currently, and the recent tensions around Taiwan, we have been having a lot of discussions about the state of China’s semiconductor industry. We thought we would summarize those here.
The first topic in this conversation is always about semis manufacturing. On the one hand, China’s domestic fab capacity is still far from the leading edge. Despite the recent headlines that foundry SMIC is “producing chips at 7nm”, the reality is that their process is still very far from being commercially available. We tend to think of this as an R&D prototype which is very, very different from a process that is economically viable for customer orders. We think the US government has made it virtually impossible for any Chinese fab to obtain EUV equipment, and so even if SMIC can someday get 7nm working, there is currently no path beyond that.
On the other hand, China’s domestic trailing edge fabs are reaching fairly robust levels. There are something like 120 fabs active or under construction there. The majority of those are owned by non-Chinese companies, but they are for the most part perfectly viable for commercial production. So for these trailing edge processes, call it 16nm and above, China is theoretically self-sufficient.
Turning to the wafer fabrication equipment (WFE) makers, China remains much more dependent on foreign companies. There are a large number of domestic start-ups making various systems, but as far as we can tell these are still even further behind the state-of-art than their foundry customers. Periodically, we see reports in the press of some “breakthrough”, but dig down a layer and the story is once again of a prototype system that is either a dozen nodes behind or not capable of producing at commercial volumes, or both.
Another important area is Electronic Design Automation (EDA) tools, the software systems required to design and produce chips. The story here is somewhat similar to those above – limited capabilities. However, there are dozens of companies plugging away on this part of the ecosystem and we have started to hear rumblings of a few of those making more progress. EDA tools are a tricky market, as their value is dependent not only on the utility they provide customers, but also their integration with the process software at the foundries. Our sense is that there are a few tools here that have achieved integration with domestic fabs, but as far as we can tell none have been fully approved at the leading edge foundries, thus limiting their value. Nonetheless, we are watching this sector more closely than we had been previously.
Beyond this, there is a whole host of systems and inputs required for semis manufacture. This range from HVAC systems, which China can produce, to pre-cursor chemicals where China seems to lack domestic providers of sufficient quality, to advanced robotics and processing equipment an area where China is behind but likely has the ability to catch up over time.
All in all, China’s ability to manufacture semis is still dependent on foreign suppliers, but are making some signs of progress, albeit only up to older process nodes.
Turning to the design side, there are now something like 20,000 fabless “semis companies” operating in China today. We use quotes, because the large majority of these companies are somewhere between not capable to entirely fraudulent.
That being said, there are still several thousand legitimate, capable fabless chip design companies in China today. Determining the prospects of all of these would require a dedicated, specialized effort, beyond our scope today, but stay tuned.
China’s fabless companies are a Cambrian explosion of experimentation and innovation. There is a lot going on in this space, with seemingly a new company getting funded every day. This is not sustainable forever, but does look sustainable for the foreseeable future. There is a near-limitless pool of money available for funding real start-ups in this space.
In general, we see the most activity around a few areas – CPUs, GPUs, analog, automotive, IoT (broadly defined), and the giant of them all – AI. A year ago we would have said that China was going to excel first in smaller chips – AI accelerators, IoT sensors and modules, small automotive parts and some analog. However, in recent months we have seen a big growth in the number of Chinese chip companies developing larger chips like CPUs and GPUS. For instance Biren, a start-up almost no one had heard of before, last week unveiled what appears to be a highly performative GPU.
This is an important development as these large chips capture a large share of the semis value pool. However, a few things about these chips bear mentioning. First, many of these chips are still in very early stages of development, and other than Biren’s GPU, we do not have much in the way of actual performance data. Many of these chips may turn out to be competitive, but many may not. Second, many of these companies are getting considerable design help from companies outside of China. A handful of Taiwanese companies, as well as some of the global semis IP/EDA providers seem to be doing a lot of the heavy lift in design. So while China can say it has domestically designed CPUs and GPUs, that statement should contain an asterisk acknowledging the considerable foreign help.
For all of these fabless companies, the ultimate determinant of their success will be their competitiveness versus global category leaders. This can be measured in the ability of these firms to export or at least sell into foreign companies. By this metric, China’s fabless companies still have a long way to go. We know of very few cases where Chinese companies are selling to non-Chinese buyers. The one exception to this was Huawei’s HiSilicon which was very capable at the global level, but is no longer. The other exception is YMTC, a maker of NAND memory chips, whose performance should scare the memory complex. Of course, China still has a large domestic market to sell into, which is a boon especially to makers of analog and automotive chips, but even here the picture is mixed. We have seen recent, credible analyst tear-downs of various Chinese hardware systems (telecoms, civilian aerospace, automotive, etc.) that still show a preponderance of foreign parts. Those global metrics will become increasingly important when the day comes for these companies to consolidate.
That leaves us with the topic of politics. In terms of domestic politics within China, despite the turmoil around the leadership of Unisoc and the big IC fund, China remains heavily committed to semis. We would not mistake the corruption investigation as a sign of de-emphasizing the sector, and instead would read it as an exercise in focusing industry participants’ attention on attaining tangible results. By contrast, the US policy domain remains complex. The US government appears committed to choking off China’s access to leading edge manufacturing technology, but beyond that its focus remains a bit hazy with a highly opaque process surrounding additions to the entity list.
To sum up, China’s semis industry has made incredible progress over the past ten years, but still has a long way to go.