Despite hurdles both technological and political, SMIC is intent on being as close to the state-of-the-art in production as it can get...
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?
SMIC is China’s leading contract maker of semiconductors. It offers the most sophisticated process technologies in the country and is expected to be instrumental to the government’s Made in China 2025 plan. The company is serious about competing against market leaders Taiwan Semiconductor Manufacturing Co. (TSMC) and Samsung Foundry in the extreme ultraviolet lithography (EUVL) era, but last year the company could not obtain the EUV scanner it had already purchased because of export requirements.
Meanwhile, U.S. restrictions prohibit leading foundries from selling chips to Huawei Technologies, which now has to find another way to build its processors.
SMIC’s unique business model
TSMC is the largest foundry in the world. By revenue, it is roughly triple the size of Samsung Foundry, which in turn is triple the size of GlobalFoundries and of UMC, the third and fourth largest foundries, respectively. SMIC, the world’s No. 5 foundry, lags those two by a large margin.
Unlike GlobalFoundries and UMC, which both decided to focus on specialty manufacturing technologies, SMIC is still working on leading-edge nodes and can theoretically compete against the more financially powerful market leaders — TSMC and Samsung Foundry.
Advanced nodes cost billions of dollars to develop and require capital-intensive 300-mm fabs (that also cost billions) to pay off, which is why many manufacturers in the recent years ceased to pursue the most advanced technologies. To that end, SMIC’s leading-edge efforts might seem somewhat surprising, though there is an explanation for that. SMIC has a unique business model that involves rather massive funding from local governments, state-owned enterprises as well as joint ventures with large multinational corporations, which significantly reduces its own capital and R&D expenditures.
It is not uncommon for makers of semiconductors, such as Advanced Micro Devices Inc., GlobalFoundries, and Intel Corp., to get incentives from local and federal governments when they build new fabs (or in AMD’s case “built” — it’s done building new fabs). Semiconductor production generates loads of well-paid jobs, directly and indirectly, so authorities are generally inclined to entice companies to build their fabs on their territories. Meanwhile, even with incentives and tax reductions, companies like Intel or GlobalFoundries still have to spend billions of their own money on their plants.
SMIC has a different approach. The company leverages the ambitions of Chinese municipalities to attract more technology companies to their areas in a bid to create high-tech clusters and generate well-paid jobs. Instead of asking for incentives, SMIC partners with local governments that co-invest in SMIC’s manufacturing facilities, a model called ‘reversed-BOT.’
At present, SMIC tends to maintain majority ownership of its fabs (something that does not always happen), but back in 2000s, early in the company’s history, municipalities would essentially build fabs for SMIC, which would operate them. Partnerships with local authorities greatly reduce SMIC’s capital requirements and give the contract maker of chips a number of benefits when compared to rivals.
The model is not without caveats though: a modern fab can easily cost well over $10 billion and municipalities may not be inclined to invest half of that sum. As a result, SMIC has to apply to the state government for help. For example, SMIC got $2.25 billion from two Chinese funds back in May to ramp up production using its 12nm and 14nm fabrication processes in Shanghai. SMIC lost its majority stake in the SMSC company which owns the Fab SN1 as a result of the investment, but will naturally continue to operate the plant.
In the semiconductor world, joint path-finding and research operations are quite common, there are numerous research alliances that jointly explore future materials and transistor structures. The results of such joint fundings form the foundation for upcoming commercial nodes, but it usually takes some time before they get implemented.
For example, IBM-led Research Alliance together with the SUNY Polytechnic Institute’s NanoTech Complex (in Albany, NY) developed the foundation for GlobalFoundries’s and Samsung Foundry’s silicon nanosheet GAAFET-based process technologies and even demonstrated feasibility of the approach with a 5nm manufacturing technology that relied on silicon nanosheet GAAFETs back in 2017. Eventually, GlobalFoundries abandoned development of leading-edge nodes and focused on specialty technologies, whereas Samsung Foundry announced plans to use silicon nanosheet MBCFETs within its 3nm node sometimes in 2022 – 2023. Meanwhile, the foundation for that technology was created by a joint team of scientists quite a long time ago.
SMIC has rather vast R&D operations and has been accelerating its expenditures in the recent years. The company also works with the international research & development and innovation hub Interuniversity Microelectronics Centre (IMEC) as well as the Institute of Microelectronics of the Chinese Academy of Sciences (IMECAS). In addition, SMIC also collaborates with its own customers, including Huawei Technologies and Qualcomm, and partners, such as Brite Semiconductor and CEVA, to develop actual commercial nodes and platforms. Such approach helps to further optimize expenditures and offer better services. Such collaborations also ensure that big players like Huawei and Qualcomm actually want SMIC to succeed and produce chips they need in China.
Modern process technologies cost billions of US dollars to develop and while SMIC uses various ways to optimize its expenditures, rising costs require additional funds. In a bid to raise from $3.2 billion to $7.5 billion for R&D and expansion, SMIC plans a new IPO at the Shanghai Stock Exchange later this year. SMIC indicated that 40% of the money raised will be used for its 300-mm Fab SN1 equipment, 20% for R&D on advanced and mature nodes, and the rest for the replenishment of working capital. If everything goes as planned, adding hundreds of millions to SMIC’s research and development budget is a rather big deal as the company spent $629 million on R&D in 2019.
Today, SMIC operates seven fabs: three 200-mm fabs and four 300-mm fabs. Back in 2016, the company bought a 70% stake in Italian LFoundry to enter the automotive IC market, but sold this fab in 2019 for an unknown reason. Meanwhile, given SMIC’s business model, this is by far not the first time when the company sells its stake in the facilities it operates.
Early in its history, SMIC built fabs that processed 200-mm wafers and focused on proven technologies. Thanks to the help from the Shanghai authorities, the company completed its first fab in 2001, 13 months after the company was established, a very high pace. In 2008, the company began building China’s first 300-mm fab for logic, another indicator how fast this company was progressing back then.
These days SMIC’s 200-mm fabs are used to make chips using mature logic nodes as well as special-purpose process technologies. In fact, SMIC has a variety of specialized nodes, including those for mixed-signal & RF, MEMS, PMIC, and eNVM. All of these technologies are in high demand and analysts expect demand for 200-mm capacities to grow in the coming years.
SMIC, with its 233 thousand 200-mm wafer starts per month (WSPM), has enough capacity to serve its current customers, but since demand for 200-mm capacity is set to grow, SMIC will need to increase its 200-mm production in the coming years. Meanwhile, keeping in mind rising demand, new and used 200-mm fab equipment is hard to find.
SMIC’s 300-mm projects have been a mixed bag for the company. These fabs are much more expensive to build, equip, and operate. Some of 300-mm fabs that SMIC once operated had to be sold because they were unprofitable. At this point, the foundry has five 300-mm fabs. Two of the fabs are located near Beijing: the Fab 2 P1 can process wafers using 55nm – 180nm technologies and has a capacity of up to 52,000 wafer starts per month, whereas the Fab 2 P2 processes wafers using 28nm or 40nm nodes (equipment used for the former can be used for the latter technology) with a monthly output of 50,000 WSPM. The company also operates the Fab 8 P1 300-mm fab in Shanghai with a productivity of only 2,000 wafer starts per month as of Q1 2020, and the Fab JN1 that is designed for FinFET-based process technologies. In addition, SMIC is also ramping up its 300-mm fab in Shenzhen.
Capacities of SMIC’s 300-mm fabs are not as vast as those of TSMC’s 300-mm GigaFabs, though it should be noted that WSPM output heavily depends on process technologies used as increased usage of multi-patterning requires wafers to spend more time in the cleanroom. Volume of scale drives costs down (which is why GigaFabs manages to reduce operating costs per wafer) but requires tremendous upfront investments as well as guaranteed high utilization rate, so sticking to mid-sized fabs might be the right idea for SMIC given the fact that the lion’s share of its revenue comes from mature and proven process technologies.
The foundry’s management fully understands that SMIC is going to need larger fabs for more advanced process technologies. According to Christopher Yim, an analyst with Bocom International Holdings Co., SMIC’s latest fabs in Beijing and Shanghai are huge and can accommodate loads of tools. For example, the 300-mm Fab SN1 in Shanghai is going to cost around $10 billion when fully equipped and ramped up to 70,000 WSPM. However, fabs like this one require investments from the government, which somewhat changes SMIC’s original business model of working with municipalities.
When it comes to implementation of new process technologies, SMIC’s progress has been quite impressive. In 2001, SMIC’s Fab 1 offered a 0.25-micron manufacturing technology, but already in 2002 it implemented a 0.18-micron process technology for logic. By Q1 2008, the company offered a host of nodes and even announced its 65nm technology. Eventually, SMIC would start making 40nm chips in late 2012, launch 28nm product line in 2015, and initiate production using its first FinFET-based 14nm process technology in late 2019. Meanwhile, SMIC has always been about four years behind market leader TSMC, a cadence that broke a bit with the transition to FinFET transistors which took 4.5 years.
In addition, it has signed process licensing agreements with Chartered (180nm) and IBM (for 40/45nm, 32nm, and 28nm). Obtaining fabrication processes and related technologies from others brought SMIC necessary IP and allowed the company to land customers who wanted a second supply source for their mature chips, but did not want to significantly redesign them. These actions enabled SMIC’s rather impressive rise in the foundry business: by 2005, it became the world’s No. 4 contract maker of semiconductors behind TSMC, UMC, and Chartered.
The meteoric liftoff, as well as technologies that enabled it could not go unnoticed by TSMC, which accused SMIC of stealing its trade secrets along with 0.18-micron process and related production technologies. In 2005, SMIC agreed to pay TSMC $175 million in an out-of-court settlement. In 2006, TSMC accused SMIC of stealing its 0.13-micron process and related technologies. In late 2009, the two companies entered into a settlement agreement under which SMIC paid TSMC $200 million, granted an 8% share in the company, and an option to purchase an additional 2% to increase its stake to 10%. Eventually, TSMC would increase its share in SMIC to 9.54%, but then that share would decrease as a result of additional funding by various state-owned companies.
While it is evident that early in its history SMIC licensed manufacturing technologies from its competitors or, at times, obtained fabrication processes using controversial methods, the company’s strategic goal has always been to develop everything it needed in house. To gain this capability, SMIC cultivated local semiconductor engineers both internally and by inking deals with universities as well as hired talents from Taiwan and the U.S., who then helped to train local talent. For example, around 400 engineers brought by SMIC from abroad in 2001 – 2003 trained 800 local engineers. Back then, SMIC had a slogan: one old one brings along two new ones, which definitely helped the company to build its very strong R&D and manufacturing operations. On the downside, when some of microelectronics giants came to China to build their fabs, they would hire engineers from SMIC.
By 2010 – 2011, SMIC had several R&D teams working on new fabrication technologies together with partners from various organizations. Multiple R&D teams enable to mitigate risks as well as develop differentiated processes aimed at specific applications. For example, SMIC was among the first to announce a PolySiON-based 28nm technology. It is also the first company to announce a 7nm-class node for low-cost applications (which will arguably compete against Samsung Foundry’s 8LPP and eventually against GlobalFoundries’s 12LP+).
While SMIC has always been behind TSMC in terms of leading-edge process technologies, it could offer competitive pricing, differentiation, and a comprehensive supply chain, which was good enough for its customers, including the largest ones, such as Huawei’s HiSilicon, Qualcomm, and Fingerprint Cards.
Mature technologies — from 40/45nm to 250/350nm — accounted for 92% of SMIC’s revenue in Q1 2020, whereas 28nm and 14nm nodes brought the company 6.5% and 1.3% of revenue, respectively. HiSilicon was among the first companies to adopt SMIC’s 14nm node to make its Kirin 710A SoC for entry-level smartphones (and, perhaps, automotive industry) that was originally launched in 2018 on TSMC’s 12nm process.
Rather slow ramp of SMIC’s 28nm and 14nm nodes and domination of mature technologies in the company’s revenue demonstrates that the company makes chips that do not necessarily need the latest nodes. For example, many ICs for communications, consumer electronics, automotive, and industrial applications have rather long life cycles. Furthermore, many Chinese chip developers tend to design for proven nodes. These factors explain SMIC’s 98.5% fab utilization rate in Q1 2020. Meanwhile, since SMIC’s FinFET-based 14nm process is unique for SMIC and the company cannot land customers looking for a second source, it is going take a while before the technology takes off, especially considering limited 14nm production capacities.
While SMIC is years behind TSMC and normally it would have taken at least a decade to catch up, the intensifying trade war between the US and China as well as the upcoming industrial megatrends present the company both remarkable opportunities and massive challenges.
This is the first of a two part series. The second will be published later this week. -ed.