BLUF: Made in China 2025 and the National IC Plan (I will conflate the two for convenience and hereafter just refer to them as “MIC 2025”) are going to fall far, far short of the goal that China produce 70% of the semiconductors it consumes by 2025. However, China is in this for the long-haul, so this failure should not make US policymakers complacent in their efforts to protect and promote this industry.
China’s goal to produce 70% of the chips it consumes by 2025 will not be met. Optimistic industry estimates suggest that China will produce less than 1/3rd that amount. The basic trade statistics are indicative of the larger story:
Between 2015 and 2019, Chinese trade in Integrated Circuits (ICs) grew from:
Imports of $230 billion in 2015 to imports of $305 billion in 2019.
Exports of $70 billion in 2015 to exports $102 billion in 2019.
Interestingly, the trade statistics also show China’s reliance on US supply of semiconductor goods increased significantly during this same time period, in spite of trade tensions:
US exports of ICs to China grew from $4 billion in 2015 to $6.5 billion in 2019.
US exports of semiconductor manufacturing equipment doubled during this time period, growing from $1.65 billion to $3.5 billion. Data here.
China continues to consume far more chips than it can produce domestically. That trend shows no signs of stopping, and by some metrics is in fact increasing. There are three factors working against China’s goals for its semiconductor industry self sufficiency at this point:
Competition (more math)
Bankruptcies, debt, and trade wars
MIC2025 called for China to produce 40% of the semiconductors it consumed by 2020. It didn’t. If you’re not even half way to 40% in 2020, 70% in 2025 looks awfully optimistic. Yet for some reason, China raised this target in 2019 and now wants to produce 80% of the semiconductors it consumes by 2030.
MIC2025 suffered from a lack of definitions: what is “Chinese” chip production? Is it:
(A) China-headquartered chip companies operating fabs in China? (SMIC, Huahong etc)
(B) all China-based fabs (this would include TSMC, SK Hynix, Samsung, & Intel’s fabs in China)?
If we go with Option B, the report card looks bad: China-based chip production supplied $19.5 billion of China’s $125 billion IC market in 2019, or 15.7%.
If we go with Option A, the report card looks worse: China-headquartered chip companies supplied $7.6 billion of China’s $125 billion IC market in 2019, or 6%.
So as we sit here at the end of 2020, China isn’t even half way to its goal for 2020 by even the most generous definition of what constitutes production of the Chinese semiconductor industry. Anyone who has looked at the numbers of fabs under construction/expansion in China and crunched the wafer starts per month (see below) can tell as we sit here at the end of 2020 that China is not going to produce 70% of the semiconductors it consumes in 2025.
Yet by some metrics, China went from having eight 300mm wafer fabs in 2015 to twenty-four in 2019. The lead time needed to build a fab shell (6 months), fill it with equipment (12 months), get up to commercial yields (another 12 months), and get products qualified with customers (easily 6 months) is considerable and its impressive China tripled its number of fabs during a 5 year period.
But when you start to crunch the numbers on wafer-starts-per-month, things look a lot less optimistic:
This isn’t an exhaustive list of China’s fabs, but it is indicative. Focus just on the yellow bars in the image above and its apparent where things stand:
27% of installed capacity for wafer-starts-per-month for foundry and memory is coming from Chinese-headquartered companies. Put another way:
73% of “China’s” chip production is currently coming from non-Chinese companies operating factories in-country, and they’re doing it at more advanced nodes than their Chinese-headquartered competitors.
This deficit is even more acute when the market share numbers are broken down for foundries and memory chip firms:
From 2018-20, TSMC’s share of the China foundry market increased from 55% to 61% while UMC’s went from 7-17%. Meanwhile, over that same period of time, SMIC’s market share declined from 19% to 16% and Huahong’s was flat at 8%.
Chinese fabless firms still choose Taiwanese foundries over Chinese foundries. Taiwanese foundries’ market share is growing while the Chinese foundries’ market share is declining or flat. That pretty much says all you need to know about how China’s domestic foundries are doing.
The news on the memory side of the industry is slightly better for China, with YMTC and CXMT beginning honest-to-God commercial production of memory chips 2020. But again, these numbers are behind schedule and much lower than China hoped for. A helpful reality check comes from IC Insights earlier this year:
CXMT began limited production of its first DRAM products in 4Q19. This company has a few thousand employees and a capital spending budget of about $1.5 billion per year. In contrast, Micron and SK Hynix each have well over 30,000 employees and Samsung’s memory division is estimated to have over 40,000. Moreover, in 2019, the combined capital spending from Samsung, SK Hynix, and Micron was $39.7 billion.
The high points of China’s semiconductor industry in 2020 are all non-Chinese firms that have factories in China: Samsung (S. Korea), SK Hynix (S. Korea), Intel (US, soon to be owned by Hynix), TSMC (Taiwan), and UMC (Taiwan).
Bankruptcies, debt, and trade wars
China’s semiconductor sector has seen a string of high-profile bankruptcies in 2020, all of which are ringing alarm bells about the sustainability of the government’s support for the industry. These alarm bells are probably unfounded; the Chinese government is never going to stop throwing money at this industry to facilitate its domestic growth, development, and sustainability. But these losses show that they’re going to lose a ton of money in the process.
Bankruptcies and Debt
There have been six notable bankruptcies, four of which stand out in that they were clearly partnering with/poaching personnel from some of the best in the business and for one reason or another it just didn’t work out:
Tacoma Semiconductor Technology: Nanjing-based Joint Venture (JV) with TowerJazz (Israel), filed for bankruptcy in July 2020.
Wuhan Hongxin Semiconductor Manufacturing: Wuhan-based and headed by the former COO of TSMC, closed in 2020.
GexinIntegrated Circuit Manufacturing: Chengdu-based JV with GlobalFoundries (US), closed in July 2019.
Huaxintong Semitech Co. Ltd.: JV with Qualcomm (US), shut down in April 2019.
Huaian Imaging Device Manufacturer
Shaanxi Kuntong Semiconductor Technology
This is on top of the mounting debt troubles facing Tsinghua Unigroup, China’s national champion for all-things semiconductor, which doesn’t look like it will end well:
A shotgun marriage involving a Chinese SOE stepping in to rescue a Chinese national champion [Unigroup] is certainly not the outcome Beijing had in mind for one of the leading lights of its semiconductor industry. Though Beijing is clearly committed to supporting the semiconductor industry in the long-term, Unigroup’s struggles are just another reminder that it has yet to figure out how to win in this industry in the short term.
In addition, several notable Chinese semiconductor firms have been caught in the trade war. This isn’t a death knell necessarily, but the US actions certainly dont help.
SMIC: The news this week wasn’t great for SMIC, but it also wasn’t the end. Keeping SMIC from manufacturing <10nm and below doesnt mean much when they only just got their 14nm FinFET to commercial yields. SMIC is years away from 10nm so the new rule doesnt effect their day to day. Honestly, if I’m SMIC, I feel like I got off lucky compared to what the US government has done to other Chinese chip firms. In a way, its actually a fantastic excuse to end their internal debate over whether to pursue profits or state demands for cutting-edge technology. Clearly cutting-edge is out of the question.
HiSilicon/Huawei: it took the US government a while to craft a rule that had teeth, but the fact that HiSilicon cant partner with TSMC (or any other fab filled with American-origin semiconductor manufacturing equipment) and use American-origin EDA tools to design their chips is a death-knell for now.
Fujian Jinhua (JHICC): The lesson with Jinhua is simple - dont brazenly steal chip IP from Micron (/any US chip firm) or the DOJ will build a case, flip your former collaborators, and put you out of business. One under-appreciated aspect of this case: UMC’s decision to cooperate with US authorities against Jinhua seems to indicate they value business coming from US fabless companies > business coming from Chinese fabless companies and keeping the US more happy than the CCP.
Everything described above should paint a very humbling picture for China. After five years, the sole point of optimism for MIC 2025’s semiconductor aspirations is in memory chips. And even here, China is behind schedule, over budget, and below yield. Memory was supposed to be the easy part. At this rate, it will take decades for China to become competitive outside of memory chips.
A few concluding notes of caution:
These numbers show huge huge demand for semiconductor manufacturing equipment in China and if the leading U.S. firms (KLA, AMAT, Lam) are to maintain their leading position in the industry they are going to need to be able to sell to China.
China is in this for the long haul. Five years of false starts and slow progress mean little in the eyes of policymakers who are thinking about this industry on a 50-year time horizon.
NOTE: These views are my own and drawn from the material I cite.