SMIC has begun producing 14nm chips, and has joined the FinFET club. The company is on the verge of a stock offering that could reap in excess of $7B...
Semiconductor Manufacturing International Corp. (SMIC) has begun producing 14nm chips, and has joined the relatively small club of semiconductor makers that can build finFETs. The company is on the verge of a stock offering that could reap in excess of $7 billion to keep investing in its business. But with the Trump Administration preventing SMIC from accessing some of the latest manufacturing equipment, can SMIC keep offering cutting-edge process technologies required by leading developers of system-on-chips (SoCs) in the long run?
This is the second part of a two-part series examining SMIC’s business and operations. The first part is here:
There are a number of looming megatrends that will have an effect on society and bring opportunities to high-tech companies. The emergence of 5G networks will drive demand for 5G-enabled applications as well as various SoCs for edge computing, but demand growth will largely depend on rollout of 5G networks. Another developing megatrend is related to artificial intelligence (AI) and high-performance computing (HPC) applications that could certainly take advantage of 5G, but do not necessary need it right away. 5G, AI, HPC, and edge computing applications will consume a lot of chips, so all foundries are going to land additional orders.
SMIC might have some additional drivers.
China’s rapid development in the recent decades has generated rather massive need for semiconductors and helped to cultivate a host of chip developers in the country. The number of Chinese IC design houses skyrocketed from 736 in 2015 to 1,780 in 2017, according to China Renaissance Securities. Meanwhile, the vast majority of chips that Chinese companies use (including those designed locally) are usually made outside of the country. To that end, any disruption in the normal supply chain presents opportunities to makers of semiconductors in China, such as SMIC and Hua Hong. There is a catch though: these manufacturers have to offer world-class products, nodes services, and competitive pricing.
Recently, the U.S. government imposed new sanctions against Huawei Technologies. Under the new policy, any chip supplier that uses tools made in the USA must get a license before selling to Huawei. The Department of Commerce reportedly indicted that it would “narrowly and strategically target Huawei’s acquisition of semiconductors.”
Unless this license is granted, or the rules are changed, this bans any chipmaker — including foundries like TSMC and Samsung Foundry — from working with Huawei. This puts the very survival of the company at stake. At the same time, the strict sanctions make other Chinese chipmakers and producers of electronics reconsider their reliance on suppliers from overseas, a good thing for SMIC.
HiSilicon was the first Chinese developer of SoCs to challenge Apple, Qualcomm, and Samsung Electronics with its own application processors for flagship smartphones and offer truly competitive performance and feature set. There are other SoC designers in the country with similar ambitions, including Oppo (at least according to rumors), Unisoc, and Xiaomi, just to name a few. At present, they have to outsource production of their processors to TSMC or UMC because they do not have any alternative in mainland China. SMIC, which is the leading contract maker of semiconductors in China, is naturally the first candidate to land orders from them once it has the right technologies and capacities.
“In our view, the valuable pool of domestic chip design talent remains pivotal to assisting SMIC, as the leading national foundry player, to advance its FinFET manufacturing roadmap beyond [12/14nm] into N+1/N+2 eras (equivalent to peers’ [8LPP] and N7-non EUV processes),” wrote Szeho Ng, an analyst with China Renaissance Securities, in a note to clients. “This, together with the high degree of tool commonality of 80% among N12-14 and N+1/N+2, should offer comfort for SMIC to potentially lift its 2020E CAPEX guidance further to >US$4.5bn, on our estimates.”
In response to the U.S.-developed CPU import restrictions, multiple processor designers emerged in China in recent years. These CPUs use Arm or x86 architectures, they offer rather promising performance, and they are produced at GlobalFoundries, TSMC, and UMC. These CPU makers transitioned right in time for the upcoming 5G, AI, and HPC megatrends.
Therefore, once SMIC has appropriate technologies and decent volumes, local makers of CPUs will become its potential clients with rather vast financial resources (as, unlike mobile SoC developers, those guys are not engaged in a cut-throat competition). Meanwhile, even today SMIC can benefit from serving 5G, AI, and HPC applications as every SoC needs a PMIC, every smartphone needs an RF FEM, and there are numerous emerging applications that do not necessarily need the latest nodes.
On the node front, SMIC is certainly not standing still. SMIC has been increasing its R&D expenditures rather rapidly in recent years. In 2014, the company spent $189.7 million, or 9.5% of revenue, on research and development. Five years later, in 2019, the company spent $629 million, or 20.7% of revenue on R&D. The increase of R&D expenditures will accelerate SMIC’s manufacturing technology roadmap.
While it is hard to expect the company to challenge TSMC and Samsung Foundry when it comes to leading-edge fabrication processes any time soon, it will likely benefit from the fact that both GlobalFoundries and UMC now focus on technologies aimed at specialized applications. GlobalFoundries will keep offering its 14LPP, 12LP, and 12LP+ nodes that will compete against SMIC’s 14 nm, 12 nm, and N+1 manufacturing technologies, but if SMIC is aggressive enough, its N+2 node will put it into another league that will be ahead of its rivals.
Competitive process technologies are important, but so are production capacities. In Q4 2019, SMIC’s 14nm production capacity was only about 1,000 wafer starts per month (WSPM). In March 2020 the company could process up to 4,000 wafers using its latest manufacturing technology. SMIC expects its 14nm capacity to increase to 9,000 WSPM by the end of July and to 15,000 by the end of 2020.
“Our customers’ feedback on 14nm is positive,” said Zhao Haijun and Liang Mong Song, SMIC’s co-chief executive officers. “Our 14nm is covering both communications and automotive sectors with applications including low-end application processors, baseband and consumer related products. We decided to increase capital expenditure by $1.1 billion to a total of $4.3 billion, to meet the market demand.”
Understanding its rather good prospects, SMIC sharply increased its 2020 CapEx from $3.2 billion to $4.3 billion. Right now, the company needs equipment to ramp up its 14nm and 28nm capabilities (as well as 7nm a couple of years down the road), so buying advanced DUV tools is something that the company has to do as soon as possible.
When it comes to funding, SMIC will be able to use the money it is going to raise in Shanghai. If additional money is needed, the company could apply to various state funds, but that will mean an increase of the total state shareholding in SMIC. China’s authorities are inclined to make SMIC more competitive, according to analysts, so they will fund. As far as money is concerned, there should not be a problem, but the state control over the company might be.
“China’s Ministry of Industry and Information Technology (MIIT), during the State Council’s January 20 press conference, reiterated the commitment to centralize resources/groom those early entrants in the capital-intensive segments of the chip supply chain (i.e. manufacturing) to be globally competitive, instead of seeding more newcomers to create vicious rivalry within the ecosystem at the ramp- up stage,” wrote Szeho Ng. “We think this bodes well for established players such as SMIC and Hua Hong in the foundry space.”
From MNC to almost government-controlled company
SMIC was established back in April 2000 by Richard Chang, a semiconductor industry veteran who was born in China, graduated from a Taiwan university, went to the U.S. to complete his post-graduate studies, and then landed in Texas Instruments, where he served for 20 years and helped to build and operate about 10 fabs in different parts of the world. To build up SMIC in a short period of time, Mr. Chang used funding from Chinese banks, investment arms of municipal governments, investors from Taiwan (even despite the fact that all investments in China have to be approved by the Taiwanese government), and investors from the U.S. In fact, prior to its IPO in Hong Kong and New York in 2004, SMIC raised about $2.5 billion from various private and institutional investors from the region and the U.S. Furthermore, Mr. Chang hired hundreds of engineers from TSMC as well as those from the U.S. To that end, while SMIC is headquartered in Shanghai and operated fabs located solely in China, to a large degree this used to be a multinational corporation (MNC) in the early 2000s.
Things have changed by now. When SMIC went public in 2004, it was revealed that the Chinese state held around 14.75% of its shares, by mid-2012 China’s stake in SMIC increased to 36.9%, whereas by late 2018 the government controlled at least 46.36% of the company. The company serves customers from all over the world, yet over 60% of its revenue came from China in Q1 2020.
Financial analysts believe that despite of massive infusions from the state, SMIC is still not formally controlled by the government and is an independent equity. Meanwhile, investors like international banks or TSMC now either hold very small SMIC stakes, or have sold out completely.
Strong financial backing is crucial for SMIC, but getting money from various government-controlled entities increases risks of getting under U.S. sanctions.
While the number of contract makers of semiconductors with leading-edge ambitions has decreased in the last 10 years, the competition between the remaining players has intensified to a large degree because the key fabless developers of chips demand very advanced process technologies, high yields, and capacities that are enough to satisfy demand for the latest smartphones that are sold in massive quantities. In fact, even to be in the game these days, one needs to have advanced technologies with healthy yields, vast production capacities, competitive services, and customers with deep pockets that can engage early.
SMIC’s current roadmap includes process technologies that use FinFET transistors and will rely on deep ultraviolet (DUV) lithography. The company bought an extreme ultraviolet lithography scanner in 2019, but so far it has not been delivered. ASML says that to deliver an EUV tool to a-no-matter-which Chinese company, it needs to have a license from the Dutch government.
“We need an export license from the Dutch government (Wassenaar Arrangement) to ship EUV to China,” said Monique Mols, the head of media relations at ASML. “We are waiting for a decision from the Dutch government and until we have the export license, we cannot ship EUV to China.”
Interestingly, even though ASML’s light sources for DUV scanners as well as droplet generators for EUV scanners are produced in the USA and then shipped to the Netherlands, they do not require an export license from the U.S. government (but they cannot be used to make chips for Huawei).
“Our DUV light source is made in the U.S. and from there it is shipped to the Netherlands, no license needed” said Mols. “Parts of the source for EUV are also made in the US. They are shipped to the Netherlands and no license is needed.”
Because SMIC does not have an EUV tool, its roadmap beyond its N+1 and N+2 technologies may not be finalized at this point, so there is no visibility beyond 2022 – 2023 in terms of technologies. Meanwhile, to better compete against bigger rivals in the next two or three years, SMIC needs to procure DUV scanners and other factory equipment to boost capacity of the fabs designed for advanced nodes (i.e., 14nm/12nm as well as N+1/N+2).
ASML does not need to get any export license to sell new DUV equipment produced in the Netherlands to Chinese companies, including SMIC. There is a small catch though. If ASML exports a refurbished DUV tool from the U.S., it needs to get an export license from the USA.
“In principle we do not need an export license from the U.S. government to sell DUV systems,” said Mols. “If we export a DUV system from the U.S. a license might be necessary. We need to comply with U.S. regulations of course. DUV systems we ship from the U.S. are refurbished systems [and] the vast majority of the DUV systems are shipped from the Netherlands.”
In a bid to procure equipment that uses technologies developed in the U.S., SMIC persuaded the American government that it had never fabricated chips for the Chinese army and identified measures it had introduced to comply with U.S. regulations back in the early 2000s. As a result, it became a consignee to a special comprehensive license (SCL) granted to Applied Materials in 2003 and then became a validated end-user (VEU) of American technology exports to China in 2007. Both licenses eliminated individual export licensing requirements on shipments of controlled items to SMIC, which greatly reduced lead times when the company purchased tools. Regrettably, the SCL program was canned in 2015 with all SCLs expiring in 2016 and SMIC requested to be removed from the VEU list in 2016 for some reason. To that end, while SMIC will not have any problems getting ASML’s new equipment from the Netherlands, it might be a little difficult to procure used tools from the U.S. Furthermore, equipment made by American companies in the U.S. will require an export license and there is no more Hong Kong ‘loophole’ for Chinese makers of semiconductors.
Keeping in mind that SMIC is raising billions of dollars to increase production capacities, it certainly has a plan, but procuring some of the equipment made in the USA might be time challenging. Meanwhile, there are also risks associated with funding by government-controlled entities and sanctions by the Trump administration.
Just like other foundries, SMIC relies on its so-called alpha clients that adopt its latest nodes early and help to tune technologies to maximize yields and other critical parameters. SMIC’s 28nm fabrication process was primarily developed with Qualcomm in mind, whereas the 14nm node was aimed at HiSilicon. Because working with Huawei’s semiconductor arm is risky, SMIC’s 14nm and 12nm ramp ups will be slower unless the company finds another partner. There are enough companies in China who need advanced nodes that could help SMIC, according to analysts, but everything takes time. In fact, some market observers believe that even if HiSilicon will break up, this will positive for the Chinese semiconductor industry in general.
“We expect the impact of HiSilicon’s potential absence from co-working with SMIC to optimize 12/14nm process parameters and yield to be short-lived, as we see a proven engineering pool in China with experience of co-working with TSMC down to N5,” wrote Szeho Ng, an analyst with China Renaissance Securities, in a note to clients. “In addition, the potential breakup of HiSilicon may accelerate the growth of China fabless start-ups: a plus for China-oriented design service houses such as Alchip and VeriSilicon, in our view.”
SMIC needs to accelerate its technology development in order not to be four years behind TSMC and take advantage of AI/HPC as well as China silicon megatrends in the coming years. To that end, the sooner the company lands a major customer that needs its FinFET nodes, the better. Meanwhile, right now SMIC does not have vast FinFET capacities, which limits its abilities to compete for high-volume orders that essentially pay for R&D-intensive modern nodes.
SMIC has demonstrated a very rapid development throughout its history as municipalities co-funded its fabs and it could land customers as a second-source foundry. This model enabled SMIC to leave behind numerous foundries both in terms of nodes and production capacities as well as allowed the company to join prestigious FinFET club. The situation changed drastically in the FinFET era as fabs got so expensive that SMIC now needs additional support from the state to build them, whereas its new nodes are so different from those offered by others that it cannot be considered as a second source contract maker.
For now, SMIC has strong R&D operations, a clear process technology roadmap that spans for several years down the road as well as a long list of potential customers from China and elsewhere that will take advantage of the looming 5G, AI, and HPC megatrends. But SMIC’s EUV efforts are greatly slowed down by the lack of an appropriate scanner because of export regulations. Furthermore, to recoup its R&D expenses on these 12nm, N+1, N+2, and future nodes, SMIC is going to need vast production capacities and make huge volumes of chips.
Without advanced technologies and competitive volumes that will interest big names, SMIC will hardly be able to buy its way into the GAAFET and EUV future, so the company is now at a dramatic point of its history. To equip fabs for leading-edge technologies, SMIC will go public in Shanghai later this year, eventually it might also need to take loans or apply for additional funding by various entities involved in the ‘Made in China 2025’ project.
Getting too much help from the state amid the U.S.-China trade war means additional risks for SMIC. But without the combination of advanced process technologies and production capacities, the company will lose any chance to close the gap with TSMC and Samsung Foundry and will eventually have to focus on specialized technologies and niche markets.
Whether or not SMIC’s management will find the right balance between funding by various entities and mitigating risks will be an interesting thing to watch over the next several quarters. One thing that is clear is that even with bright prospects, SMIC is going to need more and more capital from wealthy external sources to continue its journey.