Semicon is a series of exhibitions run by SEMI and focused on semiconductor manufacturing. It has everything, from materials to equipment to fabrication. I say series of events because virtually every region in the semiconductor ecosystem gets one – Semicon China, Semicon Europa, Semicon India, Semicon SEA, Semicon Japan, Semicon Korea, Semicon West (USA), and of course Semicon Taiwan, all throughout the year. 

The show this year

Chen Nanxiang, the Chairman of both Yangtze Memory Technologies Corp (YMTC) and the China Semiconductor Industry Association (CSIA) said in the opening keynote, “China’s semiconductor market belongs to the world, while the world’s semiconductor market also belongs to Chinese companies”. He and the China Electronics Chamber of Commerce’s Wang Ning also praised Nvidia’s progress, highlighting generative AI as a key growth driver for the industry, and were very bullish on the industry’s goals to become a $1 trillion market by 2030. Working together globally is key to reaching these goals, they said. This was a very positive and inclusive message, one that representatives of SEMI shared (albeit via video as none seemed to be present in person), but seemed at odds with the language coming from Beijing, which recently banned Intel and AMD chips from government computers, and from Washington, which continues to restrict semiconductor technology exports to China.

I have been attending Semicon China and Semicon Taiwan for several years. Semicon Taiwan last September felt like the largest show I’d ever been to, but Semicon China felt larger again, albeit simultaneously less international. It has grown every year I have attended and this time was no different. Over 35,000 visitors, 1,000-plus exhibitors, and still a number of foreign firms taking up some of the largest booths. Japanese firms were especially present, with the likes of Sumitomo, Canon, Tokyo Electron, Disco, and many more having the largest booths at the show. Korea, Taiwan, and Europe both had a large showing, with Korea and Taiwan having their own pavilions.

When speaking with exhibitors though, not everything was as rosy. Exhibitors at the Korean pavilion did admit the Chinese market was tough for them right now, but did not want to go into details. Surprisingly, despite this, they claimed Semicon China was the only Semicon where they planned on having a specific Korean pavilion this year. Such a decision can be interpreted in several ways, either negative or positive. Another Chinese exhibitor working for an outsourced semiconductor assembly and test (OSAT) vendor admitted that the entire industry was witnessing involution. Like other industries in China, too many players had entered, no one had any clear differentiation, and decent margins were rare.

There was also a clear lack of US companies, just 22 in total, and no household names such as AMAT or LAM. This may not be surprising, but compared to other foreign countries it did feel strange. Cadence’s CEO did appear in person, however, and gave a very informative speech about how AI is driving Cadence’s product development and industry growth.

Power electronics and compound semiconductors were another key theme of the show, having a large area set out just for compound semiconductor companies and also a two-day conference. China now accounts for around 40% of the global power semiconductor market, driven in part by explosive EV sales growth over the past few years. In 2023, BYD Semi overtook Infineon in the Chinese market, and now five of the top ten companies in China are domestic. BorgWarner and Hitachi are no longer in the top ten within China.

Conclusions

The current state of affairs is somewhat contradictory. On one hand, Semicon China and Beijing seem to be welcoming foreign semiconductor companies with open arms. On the other they are pushing companies, especially SOEs, to buy made in China technologies and products. They complain about being blocked from foreign technology, but then proceed to block themselves even further. The conference remarks were very positive, but on the ground, many booths were still pushing the made in China message.

Despite all of this, China remains the largest semiconductor market in the world, and foreign firms, whether they like it or not, need to find ways to tackle it. Chinese firms themselves are facing intense competition domestically so are looking to expand into foreign markets where they hope to find better margins. My impression though is that many foreign semiconductor companies have significant technological advantages over their Chinese competitors and being present in China helps them not only keep an eye on competitor progress but also take away some revenue from these firms that could be spent on R&D to catch up from a technical perspective, or be spent on foreign market expansion.

China may not be the easiest market, but the size and scope of Semicon China, despite everything, shows that there are ways to deal with it.

Traditionally, packaging has been considered a low-end non-critical part of a semiconductor’s design. In the past, it wasn’t overly complicated and keeping it low-cost was key. This led to the growth of back-end packaging plants across Asia. Most large packaging companies are either Mainland Chinese, Taiwanese, or American, and have operations throughout Asia.

But moving towards ever smaller process nodes is increasingly expensive and difficult, so new ways are being developed to continue performance increases year-on-year. Since China has faced certain restrictions on the tools and equipment it can import it has even more reason to place extra focus on methods to improve chip performance without moving to ever smaller process nodes.

What is packaging?

Put simply, a package is a container that holds a semiconductor die. It protects the die, can help dissipate heat, and connects the chip to a printed circuit board (PCB) or other chips. Packaging work is often done through a separate vendor known as an outsourced semiconductor assembly and test (OSAT), although many leading foundries like TSMC are now expanding their packaging capabilities.

Advanced packaging comes in many flavours. It is a general term used to describe many new techniques: 2.5D/3D, fan-out wafer-level packaging, chip scale package, antenna-in-package, and system-in-package, among others. Often the goal is to be able to stack, for example, two 7nm chips to reach the performance of a 3nm chip.

How does China stack up?

When looking at the industry as a whole, Mainland China has around 38% of the global packaging market, the only part of the semiconductor value chain that it leads in, and three of the top ten companies globally. Taiwan has six companies, and the US has one. Mainland China’s leading company JCET has an 11.3% market share and locations in China, Singapore, and Korea. Other Chinese players include TFME and Huatian.

The world’s largest and second largest OSATs, Taiwan’s ASE and the US’s Amkor, are heavily involved in advanced packaging, but as mentioned previously, it isn’t just OSATs involved in packaging, foundries like Intel, TSMC, and Samsung are also more and more involved.

As mentioned, JCET is China’s largest packaging firm. Its HQ is in Wuxi, which has the most packaging plants out of any city in China, and is in Jiangsu Province, which has more than any other province. TFME’s HQ is also in this province, as are plants from major international players such as ASE and Amkor.

JCET’s focus for the past few years and into the future is on nothing but advanced packaging. It often emphasizes that China’s green energy development in areas like electric vehicles and solar power creates opportunities for advanced packaging, as it can be used to ensure the reliable performance of the wide-bandgap semiconductors used in these applications. As well as help improve signal transmission in wireless technologies such as 5G and WiFi.

On the government funding front, in the summer of 2023, the National Natural Science Foundation announced a plan to fund 10-20 small scale research projects focused on chiplets and advanced packaging; committing RMB 800,000 per project, about $110,000, and 7-10 larger projects, committing RMB 3,000,000 each. Resulting in a funding package of around $4m-$6.4m over the next four years. Perhaps this isn’t a lot of money compared to what we hear the Chinese government investing elsewhere. But this isn’t research that requires buying billions of dollars worth of semiconductor manufacturing equipment. These are focused research projects on key aspects of advanced packaging like 2.5D/3D technology, interconnect architecture and optical technologies, and bonding. The final goal of this research is to help improve chip performance by one to two times and create internationally recognized research teams. It is likely breakthroughs from such research can be moved into firms like JCET with relative ease given the strong connections between government, universities, and industry.

Such research is also important to China from a patent perspective. As of 2021, Korea, Taiwan, and the US all led Mainland China when it came to advanced packaging patent applications globally, but China is not standing still, and is now ahead of Japan by quite some margin. It does still have work to do though as even within Mainland China, Taiwanese firms hold more patents than Mainland Chinese, 34% to 23%, and even US firms have 16%.

Despite this industry facing no US restrictions yet, Chinese design firms are still concerned. It has been reported that Chinese design firms are looking to use packaging plants in Southeast Asia, such as ASE’s plant in Malaysia out of concern for future restrictions Chinese packaging suppliers may face. Chinese packaging companies should do more to set up and invest in plants in Southeast Asia as part of their strategy. There will be little use in having the latest advanced packaging lines if even Chinese design firms fear using them.

Conclusions

While I find it unlikely Chinese OSATs and foundries will be cut off from imported equipment as it is much easier to replace locally than front-end lithography equipment. It is still telling that the very thought of this as a possibility has led to Chinese design firms choosing factories based in Malaysia rather than back home in China. As SEA nations fight for semiconductor investment from foreign firms this could be something they can play on. They are attracting not just investment from the likes of Samsung, Intel, and Amkor, but also Chinese packaging firms.

With regard to advanced packaging itself, I see no reason why China cannot be on a par with the rest of the world, however, does being on a par really help? Advanced packaging can, in some cases, get more performance from a chip without having to go to a lower process node, but if your competitors have access to both the latest process nodes and advanced packaging, then one is still playing catch up. Advanced packaging helps Mainland China stay within touching distance, and does provide it with part of the semiconductor value chain where it can say it is at the forefront with peers in Taiwan and the US, but it is not going to provide China with a way out of the lithography bind it finds itself in. Maybe that isn’t the goal though. Sure, as AI applications get ever more taxing it helps to have ever more powerful chips, but if one can achieve the same as your competitors albeit by taking up more physical area and more power, does that matter from a national security perspective as long as the result is the same?

Well, it’s here. In June, the rumors were there’d be a 5G Huawei phone towards the end of the year. It hit the shelves two or three months earlier than expected.

In my previous article, I argued that Huawei’s handset would be more of a domestic play, and I stand by that argument. While I did argue Huawei and SMIC creating a 7nm chip was no surprise, what has been produced has nevertheless surprised some, including myself.

So what is good about this chip? How will Huawei and SMIC progress from here? And what does it mean for others in the industry?

The positives

Let’s start with the positives from a Chinese point of view. Although not 100% confirmed, as we are not sure who else could possibly fabricate this chip for Huawei, SMIC has seemingly produced a true 7nm density chip without EUV. (Some still speculate that SMIC isn’t responsible.) Despite what many believe this was always possible – TSMC did it back in 2018 and SMIC did it by itself earlier in 2023 with a bitcoin mining chip. So no surprises here. When it comes to using DUV equipment to create 7nm density designs SMIC now seems to be on a par with the rest of the world. What the yield looks like is unknown, I can’t believe it is as low as 10%, and also it cannot be as good as an EUV process. I can only speculate that it is good enough and will improve as SMIC gets more customers for this process. With government subsidies, the economics of low yields mean less to SMIC than they might to other foundries.

The other positive is its RF front-end. While Huawei was always a leading modem designer, it previously relied on US suppliers like Skyworks and Qorvo for the RF front-end. This is no longer the case with the Kirin 9000S. It’s impressive that a completely self-developed front-end works at 5G speeds, even if the OS still says 4G.  

The negatives

Nevertheless, there are a few negatives. Despite the rhetoric, this is not a completely indigenous Chinese chip. It is based on Arm IP, it uses SK Hynix memory, it was presumably designed somehow using US EDA tools (I would like to know if it wasn’t), and the fabrication process used foreign equipment from the likes of ASML, AMAT, LAM, TEL, KLA, etc.

Sanctions to date haven’t stopped China’s equipment imports. In fact, they are higher than ever on this front rather than finding ways around sanctions. China hasn’t really needed to do anything. The equipment that can be used for 28nm can be repurposed for 7nm, and perhaps 5nm in a couple of years. While this is a positive for China, it does mean that it is still reliant; products from the likes of SMEE are still far behind. SMIC is keeping quiet. It won’t want to have any stricter sanctions placed on it, but really, the only way to truly stop it would be to limit or ban all sales into China for all such equipment. This is unlikely to happen.

The use of SK Hynix memory is also interesting. This must have come from old stockpiles as SK Hynix was not aware of any recent sales to sanctioned Huawei. This answers the question as to whether domestic DRAM or NAND is ready for such applications yet, and it seems the answer is no, as Huawei opted for SK Hynix memory which was first announced in 2020. We don’t know how much is stockpiled, so it could be possible that future versions of the chip will be forced to change to domestic suppliers.

The fallout

The popular opinion in Chinese society is that China has broken US sanctions, Huawei and SMIC have saved China, and the Huawei phone deserves all the praise it can get. In one sense this is true. It performs like a leading edge chip from a couple of years ago and is easily good enough for any application today. I myself use a phone more than two years old.

There are others outside of China that completely dismiss this chip as a low-yield propaganda project. The likes of MediaTek announced its own equivalent chip using TSMC’s 3nm process almost at the same time as Huawei’s announcement. Huawei itself used to use TSMC’s 5nm process before sanctions, so in fact, sanctions have caused Huawei to go backward.

The truth is in between of course. This is a serious chip, but not surprising. We know 7nm chips can be created using multi-patterning on the ASML 1980i series of DUV lithography machines, and this unsurprisingly is what SMIC has done. We know Huawei subsidiary HiSilicon is great at designing handset chips, and this is what they’ve done extremely well here.

Threats and restrictions remain, however, sanctions could get tighter. SMIC could be punished for supplying Huawei.; it does have a considerable foreign business that could be threatened for example. Could Huawei itself be sued in any way for using SK Hynix chips or perhaps illegally using US tools? If SMIC produced a chip where its customer could not prove it was using properly licensed tools, this could also be an issue for SMIC. I know from my own experience that not having a proper license for EDA tools in China can be quite common. This in turn could restrict any sales outside of China. Even if all this is fine, selling outside of China will still be difficult. This is an expensive $1,000 phone with no Google services installed and a chip performing to the standards of two years ago. The average consumer is not going to want to install Google services manually themselves, let alone fork out $1,000 for doing so. Patriotic marketing does not translate outside of China.

Finally, this new device may mean hard times in the Chinese market for Huawei’s competitors and other chip companies. As Huawei’s sales dropped in recent years, Oppo, Vivo, Xiaomi, and Apple, all took a piece of the pie. This in turn led to more sales for Qualcomm and MediaTek who supply these other handset companies. Will Huawei’s sales rise to eat into that of other Chinese handset companies or Apple’s? If Apple sales in China remain strong then Huawei’s phone will only serve to take market share away from other Chinese brands and hurt Qualcomm and MediaTek as well in the process.

In June, news emerged that Huawei will potentially come back into the 5G smartphone space by the end of 2023. This information apparently was provided to research firms anonymously from “industry sources including Huawei’s suppliers”. Huawei would not comment on the news, and neither would SMIC, the fab Huawei is using.

But what would a 5G phone using SMIC N+1 7nm process look like? Could Huawei reach its 2019 peak again? Who would buy such a phone? Is there any hope for Huawei’s handset future?

The past

In the not-so-distant past, Huawei handsets were the second best-selling brand in the market. In 2019, Huawei handsets accounted for roughly 15.6% of global handset sales, beating Apple but falling just short of Samsung. This amounted to nearly 241 million Huawei phones being sold that year.

Most of these phones – not all, but definitely the high-end phones – used Huawei’s own Kirin application processor chips. They were 5G enabled, ran Android, were fabricated at TSMC on the latest process nodes, and very rarely got anything but positive reviews. At the time, Huawei and HiSilicon had access to the latest EDA tools from Synopsys and Cadence, the latest IP from Arm among many many others, and full access to the Google ecosystem for the export market.

The present

Since 2019, Huawei and HiSilicon have lost access to all these suppliers to varying degrees. No access to properly supported EDA tools and foreign fabs has meant it has moved flagship phones over to Qualcomm as it can no longer manufacture its Kirin AP. Restrictions have also meant its phones are now limited to 4G as it is not allowed to buy 5G-enabled chips and is not able to produce them either. Using Qualcomm chips still means Huawei sells capable phones like the Mate 50, but it is the lack of Google services that means its products are now difficult to recommend in Western markets, even if they do take nice pictures of the moon. 

Even in China, where Google services do not exist, Huawei’s phone sales plummeted in 2020 and 2021, with the brand dropping out of China’s top five. But now, in 2023, Huawei’s sales have increased by 76% year-on-year in Q2 bringing the brand into joint fifth place with Xiaomi at around 13% of the China market. In a market where all other brands (other than Apple) are losing sales, Huawei is somehow managing to increase its own at the expense of BBK, its previous subsidiary Honor, and Xiaomi. 

Why anyone would buy a worse phone at the same or more expensive price can only be down to marketing and a sense of national pride, as although its phones are good, so are Oppo’s and Vivo’s – and they have 5G. Whatever it may be, Huawei has obviously done a great job at reviving its phone brand in the last year within China. Maybe aftermarket 5G enabling Huawei smartphone cases have helped.

The future

So let’s say the rumors are true. Huawei will have its 5G phone by the end of the year using its own chip. Unlike TSMC EUV 7nm, SMIC has only been able to achieve its 7nm through multi-patterning, basically doing multiple lithographic exposures to get the desired resolution. I expect SMIC’s approach will result in low yields and have limited capacity, meaning the resulting chips could potentially be more expensive. This could be compounded by the fact that Huawei handset APs are usually for internal use, not to be sold to other phone brands, so there is limited scale compared to Qualcomm or MediaTek. One could envision Huawei allowing other brands to use it to help SMIC scale up, but how would its performance compare to Qualcomm and MediaTek? Would other brands even want to buy it if it cost more and performed worse?

This is clearly very strategic for China, and would be a big win, but it puts SMIC in a difficult position. Historically it has always kept its head down, but announcing such a feat could put it in the spotlight for further sanctions and equipment it needs to expand its mature node capacity.

Perhaps the chip isn’t the key factor to Huawei’s future global success. I feel it will certainly become a successful brand in China. China-designed chips, China-manufactured chips, and a China-made OS are all very potent selling points in the market. From a global perspective,  it would still have a worse chip, even compared to other Chinese brands, but more importantly, no Google services, which is a killer for most of the rest of the world.

This will be a win for Huawei and for China, but Huawei won’t become the Huawei of 2019 any time soon. In my experience, HiSilicon has always had a great chip design team, but now apparently using its own EDA tools – and I suspect still some Western tools (supported or not) – along with a new process at SMIC rather than TSMC, means this chip won’t be world beating and potentially not commercially viable unless yields and scale are improved. For now, this will be a domestic play.

The semiconductor industry can, on a basic level, be split into parts based on function — sense, transmission, computing, and storage. We tend to focus on news and developments on the latest processor, CPU, or communications chip, but what about the simpler chips at the very edge like MEMS (Micro-Electro-Mechanical Systems) sensors instead of advanced AI edge processors? Not all sensors are MEMS and not all MEMS are sensors, but MEMS sensors are everywhere from smartphones to cars, from smart factories to medical apps. So, what are they, how is China progressing in them, and are US sanctions playing a role?

MEMS involves the miniaturization of mechanical and electro-mechanical devices, with dimensions ranging from micrometers to millimeters. MEMS devices are created using microfabrication techniques like those employed in the semiconductor industry. The devices contain tiny moving parts that sense, measure, and control things such as acceleration, pressure, temperature, and light.

China’s MEMS market

China is the largest MEMS market in the world, worth as much as $14.4 billion in 2022, and potentially as much as $24.6 billion by 2025. Today most MEMS feed into the consumer electronics and automotive electronics industries, almost 77% of sales. There is one problem though: foreign companies dominate the Chinese market, as most Chinese players are extremely small. Over 60% of MEMS are imported, with the figure moving above 80% when just counting high-end MEMS products.

Goertek, at 2%, is the only Chinese company with any significant market share. Broadcom (11%), Bosch (8%), STMicro (4%) are some of the larger players in the Chinese market. The industry as a whole is rather fragmented though, with 54% being ‘other’ companies that are too small to make up even 1% of the market share.

I have counted at least 70 Chinese MEMS companies, ranging from design, to IDM, to manufacturing. Products range from audio, MEMS mirrors, microfluidics, optical, radio frequency MEMS, gas sensors, pressure sensors, and more. Most of these companies are found in Shanghai, Jiangsu, and Zhejiang, with Suzhou in Jiangsu province as a major production base.

Investments in MEMS grow every year, from 43 different investments in 2017 to 144 investments in 2022. Investments focus on consumer electronics MEMS, automotive MEMS, bio-medical MEMS, and industrial MEMS. Bio-medical MEMS companies have seen the most investments but consumer electronics have brought in the most in monetary terms.

So why the growth in MEMS imports and how is it going with boosting domestic know-how and production? First, let’s consider market drivers, and second, government policy.

Market drivers and policy

The need for high-end imported sensors has constantly increased along with growth in China’s consumer electronics assembly industry, as this has started to flatten out, new industries have stepped in to help continue this growth in demand for MEMS sensors. Big data coupled with IoT means more and more sensors are being used to gather data for processing. These sensors may be applied to smart cities and factories and become prevalent in EVs. In EVs some of the most common sensors are inertial sensors like accelerometers and gyroscopes; optical MEMS like mirrors for driverless solutions, pressure sensors for airbag deployment or inside batteries; and thermal sensors to monitor subsystems like BMS. More than 100 MEMS sensors are now used in every car, and this figure will increase.

Automation and safety, whether it be in a vehicle, a factory, or a city, require these sensors.

China’s 14th Five-Year Plan, the 2021 Development Plan for Basic Electronic Components, and the 2021 Three-Year Action Plan for IoT Infrastructure all mention MEMS, the second of which even calls for targeted support of temperature, gas, motion, photoelectric, velocity, and biochemical sensors. The private industry has taken note, and that’s why we are seeing more and more companies entering the field, and local governments adding support initiatives of their own.

US sanctions

While the majority of US sanctions have focused on high-performance semiconductor manufacturing equipment and computing chips, sensors have been caught in the political crosshairs, and the February China balloon incident might mean the sector comes under more scrutiny. Since 2018, I can count at least six MEMS-specific companies that have been placed under some form of US sanction. Including sensors in general this may be over 16 companies. Chinese companies that directly describe themselves as MEMS companies include, MTMicrosystems, Shanghai Nova Instruments, North (Tianjin) Microsystems, Shenyang Institute of Instrumentation Science, Beijing Yanjing Electronics, and Hangzhou Haikang Micro Image Sensor. Only one company sanctioned after the balloon incident is sensor-focused, Dongguan Lingkong, specializing in long-range sensors. Further information is hard to find because its website seems to be down, as is the website of its major 80% controlling shareholder Eagles Men Aviation, also on the list.

Conclusions

Despite these sanctions, most Chinese MEMS companies face no restrictions. The market is dominated by foreign players, but I see no technical reason why Chinese firms can’t compete with the big boys. Right now, though, most lack the economies of scale to compete globally. Government interest in this area could help but we seem to see more local help, from municipalities like Suzhou, rather than central government help, as there are fewer barriers in this space compared to other parts of the semiconductor industry. I expect in the coming years we will see China’s larger players like GoerTek, SMEI, and MEMSRight gradually grow, even into foreign markets, while foreign firms continue to dominate in the foreseeable future. Automotive, bio-medical, and industrial sectors all show strong growth potential in China, which is good for the MEMS industry, and despite a weaker consumer electronics industry and some factories moving away from China, the fact is it is still the largest electronics manufacturing center in the world, and so China’s appetite for MEMS solutions in this space will stay strong.

One of the bright spots when it comes to China’s semiconductor industry is its design capabilities. Yes, these companies mostly rely on US design tools, and when designing leading-edge designs, need to use outside fabs such as TSMC and Samsung, but from a pure design point of view, there is some very skilled talent in China. It’s good enough to get sanctioned at least, with notable examples being Biren Tech and YMTC.

So, how was 2022 for China’s chip design industry then? Is it growing? What problems does it face?

The 2022 Numbers

Last year, 433 new chip design companies were established, bringing the total in China to 3,243, an increase of 15.4%, but the first growth rate drop in four years. Surprisingly, despite talk of companies struggling and sometimes failing, the total number of design companies continues to grow.

Total sales have also increased to roughly RMB 534.6 billion (USD 79 billion). Despite this, given the increase in the number of companies, this works out at sales of around RMB 165 million (USD 24 million) per company, the same as in 2021.

The Yangtze River Delta still accounts for over 50% of design industry revenue but central and western China, despite still only accounting for 15% of revenue, is growing at 49%. Cities like Wuhan, Xi’an, and Chengdu are playing ever more important roles in the industry as they have good universities and lower labor costs. It’s more cost-effective for a lot of these new companies to set up there, compared with the east coast.

Most of these design companies are very small though. Only 566 of them have revenues over RMB 100 million (USD 15 million). In fact, the total revenue of these 566 companies reached RMB 494 billion, meaning they account for roughly 92% of industry revenue with only 8% or RMB 41 billion remaining for the other 2,677 companies, leaving approximately RMB 1.5 million (USD 224,000) each. Suffice to say, the majority of semiconductor design firms in China are making close to zero revenue, have less than 100 employees, and are likely relying on venture capital and government money to survive.

In 2022, we continued to see design firms look to the stock market to raise capital. In total 25 such firms listed publicly, with a combined value of RMB 472 billion.

Overall, the year was a mixed bag, but from a macro perspective at least, Chinese firms are doing a bit better than I expected. However, consolidation would be preferable to having a large number of small firms.

The Problem

Despite COVID-19 and sanctions, officially at least the Chinese design industry continues to grow. In some cases, sanctions have helped firms. Losing access to foreign technology means some Chinese design firms must either find ways around sanctions or buy local. The same can be said of their customers who now may buy local designs to have a more secure supply chain. Of course, it isn’t all rosy in this regard. Just recently, we saw Biren Tech lay off workers and simplify its product to survive. YMTC laid off 10% of its workforce directly due to the sanctions placed upon it. Being forced to buy local, either through necessity or because of orders from on high, can result in a worse end product, slower TTM, more bugs, or lower yields. There are silver linings for certain firms, but overall I’d say Chinese companies would prefer it if there were no sanctions.

Most of China’s chip companies and their revenues still come from the consumer electronics chips sector as well as from telecommunications. Many of these are simpler chips that are designed for mature process nodes. The problem China’s industry needs to overcome is how to move up the chip value chain while avoiding sanctions. At this moment, the design capability is there, but why should companies like Biren Tech bother designing for the leading-edge if they are still going to lose access to TSMC or Samsung? That would entail doing a lot of design work with no way to manufacture it. Recently, a now-blocked article written by a China industry veteran went viral in China. It outlined how China will take until 2030 at the earliest to be competitive at even 28nm for lithography, etching, deposition, and other semiconductor manufacturing equipment. The author also concluded that it would take until 2053 to fully catch up, admitting this was a conservative estimate.

This is the crux of the problem now for Chinese design firms and the Chinese government. They are fully capable of designing powerful high-end chips, but if they risk losing access to the ability to manufacture a design after spending millions creating it, why go to all the effort? On the other hand, if Chinese firms stop perusing such designs altogether, the talent will either go elsewhere or be used to work on things that can actually go to market. China needs to somehow find a way to continue funding such projects to maintain its high-level design capabilities despite being aware that there is a high likelihood of zero returns on investment, while simultaneously looking to find ways to speed up its equipment R&D. This is easier said than done, and in my opinion impossible to achieve during this decade, especially since its own equipment firms are sanctioned, corruption is common, big-fund bosses are under investigation, and private firms are unlikely to continue working on designs without the promise of returns.

Recent joint sanctions from the US, Japan, and the Netherlands only serve to exacerbate and compound the problem. While previous sanctions mainly affected firms working at 14nm and below, and focused on US equipment, new sanctions mean China loses access to Dutch and Japanese equipment that could affect China’s fab expansion for mature nodes. Some DUV machines from companies like ASML and Tokyo Electron are to be banned because they can still be used to create 7nm or even 5nm designs. If this were to be expanded to all DUV machines, it would affect mature node fab expansion in China as well and affect China’s ability to be self-sufficient even when it comes to simpler mature node designs. It will be interesting to see how far these bans go and how well they are enforced. China’s response so far has been rather tame compared to years gone by. The coming year and the rest of the decade will be an interesting watch. Personally, I can’t see any easy route out of this for China in the current climate.

Editor’s note: China’s semiconductor industry has been on edge since mid-July. This time, the threat comes from within. China’s corruption regulator has launched a series of investigations into some of the country’s leading figures in the semiconductor industry. Many work directly or indirectly with the state-backed China National Integrated Circuit Industry Investment Fund, also known as the “Big Fund.” 

Established in September 2014, the fund has been instrumental in nurturing China’s homegrown chipmakers, including successful examples like SMIC and YMTC. The fund runs on a market-orientated model and has more than a quarter of its equity held by the Ministry of Finance. These investigations could reshape the country’s most influential backer in the field and have implications for years to come. Below is an opinion piece looking at what we know so far, why it matters – and where things may go next. 

Leaders under investigation

Xiao Yaqing, head of MIIT; Ding Wenwu, President of China’s National IC Industry Investment Fund (Big Fund); Lu Jun, Chief Executive of Sino IC Capital and asset manager for the Big Fund; Wang Wenzhong, a partner at Hongtai Fund in Shenzhen and asset manager for the Big Fund; Yang Zhengfan, Deputy Head of an investment division of Sino IC Capital; Zhao Weiguo, ousted CEO of Unigroup, one of the largest beneficiaries of the Big Fund. These people may not be household names or even familiar to those who follow the China tech industry but they have all played a key role in government efforts to become self-sufficient in semiconductors. And all of them have now been placed under investigation by China’s corruption watchdog, the Central Commission for Discipline Inspection (CCDI).

What is the Big Fund? How successful has it been? And what problems does it have?

The Big Fund(s)

So, what is the Big Fund? Or perhaps we should say “What are they?” because there are officially two phases of the fund.

Big Fund I raised a total of RMB 98.72 billion in 2014. Its major shareholders are central governmental institutions and leading state-owned enterprises (SOEs). The majority of Fund I went to increasing the capacity and design capabilities of China’s semiconductor industry. For example, approximately RMB 30 billion went to foundry expansion, RMB 20 billion to IC design companies, RMB 19 billion to memory, and RMB 10 billion to assembly, test, and packaging (ATP). Equipment only received RMB 2 billion. The focus is clear: expand the capacity of logic and memory foundries, grow China’s design firms and solidify the country’s strong packaging industry. Most investments from Big Fund I took place between 2014 and 2019 – it planned to exit from these investments between 2019 and 2024 but may remain invested in some after this period.

The top recipients of Big Fund I money include YMTC (RMB 13.5 billion to RMB 19 billion, most that was invested into memory), HLMC (RMB 11.6 billion), SMIC North (RMB 10.4 billion), SMIC South (RMB 6.3 billion), San’an Optoelectronics (RMB 4.7 billion), and JCET (RMB 4.6 billion). 

When we look at how these companies are doing today, we can conclude these investments have been a success. YMTC is beginning to become a respectable competitor to Micron, SK Hynix, and Samsung; SMIC, despite limitations thrust upon it, is experimenting with multi-patterning techniques to create simple 7nm chips; San’an Optoelectronics is China’s leading compound semiconductor IDM; and JCET remains mainland China’s leading ATP firm, perhaps as good as rivals from Taiwan. Chip maker HLMC is maybe less of a success.

Phase two of the fund, Big Fund II, started in 2019 and will finish in 2024. It has similar shareholders but sees more SOEs participating. The total amount raised to date is roughly RMB 204 billion. While a large part of the fund still goes to the capital-intensive foundry sector, Big Fund II aims to focus on large investments to fewer companies in design and materials, though investments don’t seem to follow this stated goal. 

Different from Big Fund I, Big Fund II helps develop key companies through investing in their upstream suppliers and downstream suppliers, usually coinciding with the establishment of an industrial park. Big Fund II also aims to promote downstream applications in key sectors like automotive, big data, and communications. Big Fund II has only invested RMB 86.9 billion to date and expects future investments to focus on key sectors outlined in the 14th Five-Year Plan (2021 to 2025).

To date RMB 55 billion of Big Fund II has been invested in foundry expansion, RMB 18 billion in design, and RMB 13 billion in ATP. Very little has gone into equipment or materials. While this could be more related to the costs involved in chip manufacturing, which far outweigh the cost needed for equipment research, one would think this key bottleneck would have more focus. To date only around RMB 500 million seems to have been invested into equipment. For comparison, Dutch semiconductor equipment maker ASML spent 2.55 billion euros (RMB 17.67 billion) alone on research and development in 2021.

Nevertheless, the Big Funds have helped to drive private capital into areas the government wants. From 2015 to 2019, private capital driven by Big Fund I raised up to RMB 500 billion. It is estimated that Big Fund II might drive as much as RMB 600 billion. This brings the total raised capital from 2014 to the present to approximately RMB 1.1 trillion to RMB 1.4 trillion, to be invested over the next decade. 

Big Fund Problems

While I think there have been some successes, such as YMTC, Changxin Memory, and SMIC (despite more board departures) among others, the funds haven’t always been spent wisely, and this is especially true at the local level. Local government funds have perhaps wasted billions of yuan in the semiconductor industry since 2019.

It is not uncommon for local governments to blindly rush into popular industries, especially those that are key parts of the central government’s Five-Year Plan. This can create room for scams, which we’ve seen in several provincially funded projects due to corruption and poor due diligence. One good example is Wuhan Hongxin (HSMC), which aimed to produce chips from 90nm to 7nm. None of its founding members had any semiconductor experience yet they managed to entice a leading TSMC veteran to join as CEO, convince the Wuhan government to invest, and even managed to purchase some lower-end equipment from ASML.

It is common for the managing company to invest very little or even nothing in such projects, while most money comes from the local government. The Nanjing Huaiyin District government for instance invested more than RMB 2 billion in Dehuai Semiconductor in 2017, while its own annual public budget was only RMB 2.56 billion. More financing rounds followed, with most funds coming from the Huaiyin District government. Before mid-2020, it was common for local governments to conduct such billion yuan investments, and there are many more examples of wasted investments and unnecessarily large industrial parks that remain mostly empty. Phytium’s large 30-story building in Tianjin with only two floors used springs to mind, although this may have started to fill since my last visit in 2019.

In my experience, local investment decisions are often made by rooms filled with politicians and only one or two industry experts whose views may get crowded out. Such a trend was also criticized by the State Council in October 2020, but such investments have continued albeit in a more hidden manner.

Conclusions

We must remember the original goal of these funds: reducing foreign dependency on semiconductors. Yet China is importing more chips than ever. Despite its local production growing, imports have too, and the deficit has grown from a pure monetary perspective. 

In its updated ‘Made in China 2025’ plan revised in 2019, China explicitly states that it is targeting 58% of the chips it uses to be made in China by 2020 and 80% by 2030. In 2021, chips manufactured within China’s borders only accounted for 16% of the chips the country was using. It has massively failed at reaching this goal. Moving from the unrealistic goal of 40% by 2020 to 58% by 2020 really highlights to me just how out of touch with the industry Beijing is. Even the head of the Big Fund, Ding Wenwu, who is now under investigation, once warned it was “unrealistic” to cut corners. Well, it seems that is what was expected.

Chinese investment into the semiconductor industry in my view has faced the following problems:

1. Unrealistic targets set by people with little knowledge of the industry – some of the timelines and goals were way too ambitious – which lead to inefficient investments a lot of the time.

2. Lack of industry experts making decisions or being listened to. 

3. Tens of thousands of “semiconductor” companies appeared overnight to scam government cash; local and central government have a hard time telling which businesses are serious and which aren’t. Once they do get their hands burnt the process to get funds becomes harder for legitimate businesses as they are fighting against imposters.

4. Money is spread out among many companies, a lot is wasted, and a lack of investible ideas also means money often goes to companies that are unsuitable.

5. With this amount of money flowing, not having any corruption would be a big surprise.

In my opinion, China needs to realize the semiconductor industry is a global industry – no single country can be self-sufficient and such goals lead to spreading oneself too thin. Investments should focus on where China is strong: ATP, design, memory, and mature process nodes. YMTC’s success is a positive sign for Beijing. Becoming strong in memory first is a path Korea and Japan took, and perhaps Beijing can learn from them. Right now though, it risks becoming a jack of all trades. 

And with the US CHIPs Act passing, China may well feel more threatened than ever, so doubling down on its current path and investigating some executives to be scapegoats for any potential failures may be what it has decided is its best course of action. One has to ask oneself though, would the CHIPs Act even be a thing if China had not so aggressively pushed for the self-sufficiency dream?

The semiconductor supply chain faces a new problem: the Shanghai lockdown. Shanghai is an important center for the semiconductor industry in China holding a complete supply chain of design, fabrication, and ATP (assembly, test, and packaging). In each of these verticals, Shanghai accounts for roughly 20% to 25% of China’s sales. The city is also famous for SMIC, China’s premier chip fabrication company.

In addition, Shanghai’s semiconductor industry is connected to its own electronics industry as well as nearby tech zones in Kunshan and Suzhou, which house electronics manufacturers such as Luxshare, Wistron, Pegatron, Foxconn, Logitech, Bosch, and Plantronics, among others. Many of these companies are customers of the semiconductor companies and also have operations or factories in Shanghai linked with the international ports there.

Closed loops and working from home

The center of Shanghai’s chip industry, and indeed China’s high-end semiconductor manufacturing industry, is Zhangjiang High-Tech Park in the city’s eastern Pudong district. While there are supporting areas in Shanghai and a strong ATP industry in Suzhou, Zhangjiang is the core where companies such as SMIC, ASE, and Huahong can be found.

To maintain production, Shanghai’s fabs have created closed-loop systems. The majority of employees are working on-site. Huahong is the most publicized of these. It has over 6,000 employees working and living at its Shanghai facility, converting meeting rooms and corridors into bedrooms, even making shower rooms. SMIC and TSMC Shanghai operations claim to be maintaining full production through similar closed-loop systems with some teams working from home but most on-site.

Shanghai is also a center for semiconductor materials such as polishing liquids and photoresist removers from companies such as Shanghai Xinyang and Anji Technology. Sales of these materials to fabs are said to be ongoing but experiencing some problems with “pandemic-related delays and customer production line conditions”. This suggests things might not be fully up to speed at fabs and ATP facilities in Shanghai.

Semiconductor equipment is also experiencing logistics delays, meaning fab expansion plans could be delayed too. For example, SMIC is one key company expanding its capacity and is experiencing hold-ups receiving equipment. Logistics in neighboring Jiangsu and Zhejiang provinces are reportedly experiencing setbacks and congestion. The main issue seems to be getting equipment or other supplies from the port to the fab or from neighboring provinces. Despite the fact that ports and customs clearance seem to be running smoothly, and online portals using WeChat have been set up to help semiconductor companies get goods through as fast as possible, equipment deliveries are still struggling to get from the port to the customer. If such delays become frequent it could increase the time new lines aren’t making money for fabs and are a burden. 

Equipment is still being manufactured though. Like the fabs, equipment companies like AMEC, an etch equipment company, have also implemented a closed-loop system to maintain operations.

It’s a different situation for chip design companies, however. Roughly half of China’s major chip design companies are in Shanghai, and it is home to most large international players’ R&D offices. HiSilicon, Unisoc, Nvidia, and NXP are among the numerous firms to all have operations in Shanghai. For the most part, these integrated circuit (IC) design companies can carry on working from home but there are limitations. A small number like Espressif Systems, therefore, have staff including its CEO working and living in their offices. 

Where chip design companies are working from home, there may be further disruption. Company workstations may not have been able to be brought home so employees may be using their own computers to VPN into office workstations. Many companies will have limited VPN access or none at all though, and even with a VPN there could be security issues they would not face otherwise. Using home connections may also result in limited bandwidth and a slowing down in the amount of work that can be completed each day. If any hardware expansion is required, this also isn’t possible – no one can visit the equipment room to install a new server or storage rack. Many companies have multiple locations around China or even the world, but at best this means there will still be delays as part of their team is working from home. There is potential for global projects to be held up, including for foreign companies; of course, many companies don’t even have this luxury, with their whole team in Shanghai. Finally, some work just can’t be done from home: how does one go about contacting upstream companies, testing wafers in labs, getting them to ATP facilities, and then to electronics companies? Working from home and with the current logistics situation, this seems like a difficult task.

Conclusions

While semiconductor manufacturers and design companies have found ways to carry on their work, we can expect significant delays. Working and living in the same space in unfavorable conditions isn’t consistent with good work morale and high efficiency, especially when there is no clear end date in sight. A single Covid case in any of these facilities could spell temporary closure; we have already even seen one key oxygen facility stop production after staff tested positive for Covid, despite them producing something you’d think would be essential to many of the patients in the city, so there is no reason to think semiconductor plants couldn’t also be closed if one experienced an outbreak.

While working on-site in a closed loop is a good idea to keep production moving under China’s Covid policies, it’s no good if the fabricated chips cannot get to downstream industries or if production line expansion is held up due to equipment shipping delays. OnSemi recently stated its China distribution center was forced to shutdown due to Covid – the product may be there but there is no way to get it to customers. In a best-case scenario, we see lags in getting chips to customers and delays in capacity expansion; in the worst case, there is a complete stopping of production. Chip design industries, while coping better than fabs, will also face design delays and security issues. Cloud-based electronic design automation (EDA) solutions may help with some of these bottlenecks, but they seem unlikely to solve all of the current issues. These design setbacks could add to chip production lead times for the next year or two.

Chinese fabs and fabless companies potentially face the most delays in getting products to market. Overall, this means downstream industries like consumer electronics devices and automotive could face shortages and price hikes; the consumer and economy will therefore suffer.

The real solution is a policy one. Working from home keeps chip design going but naturally incurs delays. For chip manufacturing, it is impossible in the long-term; living in the fab can only be temporary. Only policy change therefore can get the industry back to where it needs to be. A band-aid approach simply won’t cut it.

A lack of key semiconductor talent and the rising cost of such talent are hot topics in the industry in China right now.

Previously, it was difficult to attract the brightest talent into the industry. There were several reasons for this, with two key factors chief among them. First, semiconductors require an insane level of detail and accuracy and any part of the semiconductor design or manufacturing process requires years of study even to get to an acceptable level. Second, and perhaps most importantly, all this studying and effort doesn’t get the student a better-paid job than simply learning to code application-level software. 

The brightest talent, even those that studied semiconductor design, would therefore go to the big internet companies and get much better pay than China’s struggling semiconductor companies could ever afford.

However, in the past couple of years, we have seen a huge increase in semiconductor talent demand. Companies are becoming increasingly vertically integrated, launching their own semiconductor departments, and startups are appearing everywhere. Simply having a lot of investment and demand does not mean core talent appears overnight, however. Talent takes time, meaning the best workers right now can demand high wages.

The upward trend is a good thing for the industry in China, but it needs to be managed. Ensuring the new influx of talent doesn’t all fall into the design space to the detriment of other key parts of the value chain – especially chip fabrication plants and equipment manufacturers where the country faces the strongest pressure from foreign governments – is crucial.

Skyrocketing salaries for chip engineers

Semiconductor startups face a problem: they either increase operating costs and burn more money by attracting the best talent with high wages, or ignore this trend and risk recruiting less than ideal candidates. HR personnel in the industry, including people I have spoken to myself, are even saying that the best graduates may have multiple offers and play them against one another in an attempt to get the highest salary possible. 

The best graduates – those graduating at Masters or PhD level – can demand around RMB 400,000 per year ($62,000) per year; such a figure has been touted in the media as the graduate salary Oppo has been offering to semiconductor engineers. This is an extremely high wage for someone with no actual work experience. 

Regardless, salaries still seem to be rising, with Alibaba offering up to RMB 500,000-600,000 per year to newbie chip engineers. For comparison, the average annual salary for a semiconductor engineer in the US is around $110,000 (a little over RMB 700,000).

This upwards trend in salaries is putting huge pressure not just on startups but on all traditional semiconductor companies. It is simply too hard to compete with the internet firms that have moved into this space – and on top of that, smartphone companies are poaching employees too.

There is a double whammy of increased cost and competition for talent as well as a reduction in customers: Alibaba or Oppo may have been your customer before, but now they’ve taken your employees and are making their own chips.

Remember as well that these are graduate wages. Experienced semiconductor veterans can demand much more not just for their talent but also for the simple fact there are so few of them. This situation makes it easier for them to frequently switch companies to gain higher wages or at least threaten to move as a bargaining chip for a better deal. That’s not good for the long-term development of a semiconductor company and its products, where designs can take more than 18 months from start to finish.

Structural problems in the industry

It is worth pondering the effect all this will have on chip manufacturers like SMIC and semiconductor manufacturing equipment (SME) makers like AMEC – companies propelling the twin strategic priorities of China’s chip independence plan.

New talent is attracted into the semiconductor industry due to increasing wages, but these wages are mainly focused on the chip design industry. Internet companies and handset companies are pushing salaries up, but none of these companies are developing equipment or setting up fabrication plants. New talent will want to move into design potentially at the expense of China’s key chokepoints – fabs and equipment. 

Wages here will have to keep up as well if they are to make sure they attract all the new talent they require. Of course, this adds to costs, potentially causing Chinese fabs to face the same cost problems those in Western countries face compared to the likes of Taiwan. Indeed, Chinese fabs such as SMIC have been offering much higher wages to foreign employees in order to attract them away from Taiwan or South Korea. It’s unclear whether they are willing to increase wages for local employees as well.

Naturally, foreign semiconductor companies’ China operations will not be unaffected by such trends. While Chinese companies are receiving various forms of government backing, foreign companies with design centers in China aren’t. It is hard to retain the best talent in your China operations if a startup, internet company, or phone company next door suddenly offers double the salary. China is no longer the low-cost design center it once was.

High salary doesn’t mean high quality

Not all graduates are receiving such high wages though. Some people I talked to in the industry argue that so many new graduates are entering the semiconductor industry, especially digital design, that most of them can’t demand high salaries because companies can just choose a different graduate. Indeed, companies I have spoken with have admitted there is no lack of junior engineers. 

A small percentage who are postgraduates and have real-world experience can choose between 10 or 20 different offers. But most of the new graduates have taken self-study or a training class to switch to the semiconductor industry from other majors.

Semiconductor talent on paper has increased, but most of it has limited ability and experience. There aren’t enough quality graduates in the market, and 3-6 months of crash courses is not enough to create high-quality talent. 

Wages may have increased, but the difficulty hasn’t. Once a chip is taped out, you can’t edit it like you can a piece of software – the industry has a very low tolerance for errors. HR needs to be more aware of the actual abilities of these new semiconductor engineers and stipulate appropriate wages and positions for them.

Conclusions

There is no denying salary increases were needed if China was to attract talent into the industry, so from this perspective, this trend is a good thing for the country.

However, the money could be flowing to the wrong places and could harm many in the sector. Startups and traditional semiconductor companies are struggling to compete as their former customers start poaching talent, and to fill the talent gap the market is now flooded with new semiconductor engineers with little real-world experience or deep enough knowledge to make sure the final product is bug-free. 

What’s more important is this increase in talent and wages doesn’t seem to be heading to where China needs it most – essentially fabs and SME companies. The heightened interest and rush to learn semiconductor design is no bad thing for China, but right now it is unbalanced and potentially detrimental to some parts of the industry.

Let’s be clear: Arm China was created by SoftBank and Arm to make more money out of China by presenting itself as a local company as much as possible. It was always SoftBank’s plan to have Arm China create intellectual property (IP)  for the Chinese market. It wasn’t in the plan for Arm to lose the power to choose who runs Arm China or for Arm China CEO Allen Wu, 53, to run investment companies competing with Arm’s own, or for the renegade CEO to set up an “Open NPU Innovation Alliance” (ONIA) that potentially competes with Arm globally.

Given the geopolitical nature of the semiconductor industry right now, whether or not Wu had taken over the joint venture, Arm China would still have attempted to market itself as a Chinese company. With Arm’s architecture facing stiff competition from the open-source RISC-V architecture, marketing in China would not have been very different.

Arm China doesn’t go so far as to say it is independent now, but it claims it  is “independently operated and Chinese controlled.” No matter what it calls itself, Arm’s company in China is nonetheless still 49% owned by Arm Ltd. One Chinese investor in Arm China, Ningbo Meishan Bonded Port Area ARM Investment Management Partnership, has sued the company in a Shenzhen court over the standoff with Wu, but cases like this may take years to be resolved.

This whole situation is a red flag for any foreign tech company considering a JV in China. There are other ways of entering the market that might be more suitable for some companies. You should explore these before going the JV route, and safeguard your company chops! 

Perhaps most important of all, the conflict in China threatens the entire Nvidia-Arm acquisition deal. What can Nvidia offer China in order to get the deal through? Can Arm China still get access to its UK parent’s IP, notably the next-generation Arm v9 architecture?

With a Chinese face

Around 27% of Arm’s revenue originated from the Chinese market in 2020. From its launch in 2017, Arm China was intended by Arm and SoftBank to appear more Chinese and to allay any fears of its large Chinese customers that supplies from a foreign company could be abruptly cut off.  

Seeing the threat of RISC-V and self-developed instruction set architectures (ISAs) combined with sanctions on ZTE taking effect only months before, Arm in 2017 was making a move to cast itself as a local option and continue the China gravy train. At the time, it seemed like a shrewd move to maintain sales growth in China. So was appointing Wu, a China-born US citizen, educated at Michigan and Berkeley, who had worked for Arm in China since 2014.

June 2020:  Boardroom showdown

Arm discovered in 2019 that Wu had been attracting investments to Alphatecture, his own fund for investing in tech startups, when he should have been bringing them to the Hopu-Arm Innovation Fund. (Not surprisingly, Hopu Investment Management Company, one of the Chinese investors in Arm China, has been siding with the mother company). In June 2020 Arm China’s board of directors, four of whom were appointed by Arm Ltd., voted seven-to-one to oust Wu. He refused to leave, kept the registration documents and all-important company chops, and is alleged to be paying his own legal expenses from Arm China’s bank account. Over a year later, he remains in charge at Arm China, despite attempts by both foreign and Chinese owners to remove him and to appoint new executives. 

As long as Wu controls the company chops, the board members can’t get rid of him because company decisions need the chops to become official. Over a year ago, SoftBank and Hopu asked Shenzhen regulators for a replacement seal. Some observers have speculated that Wu has backing from high-level Chinese officials, but that cannot be verified. Arm China’s Chinese state investors include sovereign wealth fund Chinese Investment Corp. (CIC) and Shenzhen government-owned Shum Yip Group.  

In fact, Wu is suing the three executives the board tried to reinstate. In July 2020, Wu even wrote an open letter on Arm China’s WeChat account, asking the Chinese government to help him fight Arm. A year on, they still haven’t publicly come to the rescue. Arm China owners last year offered Wu tens of millions of dollars to leave but he still occupies Arm China’s head office in Shenzhen.

The chops are still in Wu’s hands, and he isn’t budging. I’ve witnessed first-hand the entourage of bodyguards he has at events. Could the mysterious chops actually be on his person at all times?

August 2021: From Arm China to Anmou Technologies

On Aug. 26, Arm China officially launched a new brand, Core Power (Hexin Dongli), to promote self-developed IP (CPU, GPU, XPU, SPU, VPU, ISP, NPU) and services. It vowed to continue the localization of Arm’s CPU architecture and to create products suitable for the Chinese market. Much emphasis was placed on self-development, how Arm China  is “independently operated and Chinese controlled”. Wu said at the event, “Since it was established in 2018, Anmou Technologies has not only inherited Arm’s CPU business in China, but also deployed new businesses for the digital age”. Rumor has it that Arm China employees used the words “peace and love” to describe the relationship between Arm China and Arm, but I can’t confirm this. The company was described, however, as “China’s largest CPU IP supplier.”

So, it is claiming to be Chinese: Chinese controlled, Chinese run, with Chinese IP and, even now, is not just licensing Arm IP but developing its own. From a marketing standpoint, its press releases no longer say “Arm China” but “Anmou Technologies,” the legal name in China. It is clearly trying to erase any indication it has foreign connections.

Arm China may now be an Arm rival

The most startling announcement at the Aug. 26 event concerned self-developed IP, especially the neural processing unit (NPU) microprocessor. 

Another interesting bit of news was that Arm China had created an “Open NPU Innovation Alliance” (ONIA) in June, with Wu as the alliance’s chairman. The alliance’s website says the NPU’s ISA is open source and will be promoted globally, following a business model that seems similar to RISC-V Alliance’s. While intended for worldwide participation, so far all alliance members are Chinese entities. Besides Arm China, the 54 members include AllWinner, Changan Auto, Rockchips, Sword7, Sanechips (ZTE), TCL, and Tsinghua University.

This alliance has the potential to compete directly with Arm’s own NPU cores. Surely, this was never the vision of Arm executives in Cambridge or of Arm’s owners in Tokyo. Having said this, the whole endeavour feels like a difficult undertaking, different NPUs may be good at different applications. I guess this group will need to decide on what their applications are rather than trying to have a solution for everything.

A cautionary tale

As stated above, this situation seriously puts in doubt the viability of any tech JV between Chinese and foreign partners. It isn’t a good look for China if Beijing wants to attract future foreign investment in this sector. 

Arm and SoftBank wanted to find ways to make more money out of China. Yes, China is a different kind of market. In other markets it is unlikely a JV would be considered by a chipmaker or designer, but Arm executives wanted as much access as possible, so pretended to be as Chinese as they could to curry favor. The strategy backfired in a spectacular way.

JVs are often touted as the best way to deal with the Chinese market but, all too often, the foreign partner is left with a husk of a company while local management runs off to set up a competitor,  taking all the well-trained staff with them. By not even bothering to set up a new company, Wu can be credited with an innovation in the decades-old scheme. Keep the chops and it’s all yours, baby. There are many other ways to be successful in China, don’t let anyone talk you into a JV without doing some homework on other options! Even if you have the most shares of any single party and on paper have control like Arm, you may find out the hard way how little you may have. 

How will this affect the Nvidia-Arm acquisition? Arm’s ownership shifting from Japan to the US will make it be subject to US sanctions. It certainly could erode the value of the deal. In my opinion, Nvidia, the world’s biggest maker of graphics and AI chips, will have to offer the Chinese government something big in order to get the deal through. 

As of now, Arm China can license Arm’s v9 architecture if it wants to, but it has so far chosen not to. It essentially acts as a distributor of the IP to Chinese companies but does not touch the IP, which goes directly from Arm to the Chinese licensee. Indeed there are several Chinese licensees of the v9 IP, including at least one architectural licensee. It is unclear why Arm China doesn’t want to license v9 and make modifications like what it does with v8. Either way, Arm, being a British firm, faces no restrictions on licensing to Chinese companies right now. But this could change with an Nvidia acquisition. With v9 already being in China, including architectural licensees, it is hard to see what Nvidia can offer to convince China to let the acquisition go ahead.

Correction: The last two paragraphs of this article have been updated to reflect the fact that Arm Ltd is still licensing the Arm v9 IP to Chinese companies. A previous version of the article incorrectly states that Chinese companies have no access to the IP.

Chip design companies have to plow a high proportion of revenue back into research and development (R&D) in order to catch up with competitors or to just stay ahead of them. The more a company spends attracting the best researchers and licensing the best tools, the more likely a company is able to innovate, keep up with Moore’s Law, stay ahead of the competition, and win the most market share. If a company doesn’t spend enough and falls a generation behind competitors, then it will end up either having to burn money to catch up — or find something new or niche to do.

Spending 18% of revenue on R&D for a long time has been considered a healthy share. For the semiconductor industry as a whole, this share may even be 22% now, perhaps making it the industry where the most is spent on R&D as a percentage of revenue. But are Chinese companies keeping up with this? Some state-backed chipmakers are attempting to help fulfill Beijing’s ambitious goals of replacing much of the nation’s imported semiconductors with homegrown products. Are they spending a larger proportion of their revenue to catch up with their international peers?

Looking at the numbers, we see that most Chinese chip companies are spending about 18% of their revenue on R&D. That’s on par with the global norm. But since their revenue tends to be smaller, you would expect they’d have to spend an even higher percentage to catch up.

US vs China fabless R&D spending

The US Semiconductor Industry Association estimates that US chip companies, on average, spent about 20% of their revenue on R&D in 2019. Some spend much more: Marvell spent around 40% in 2020, Nvidia 26%, and AMD 23%. Microchip Technology, however, spent only 16.6% in 2019.

Looking at Chinese fabless companies listed in Shanghai and Shenzhen, we can see the average is around 23% of revenue in 2020. Removing the one obvious outlier, Cambricon, which spends 167% of revenue on R&D (it’s in the burning-money-to-catch-up stage and should be counted as a startup in some ways, despite being listed), this percentage drops to 17.5%. That is almost the aforementioned healthy 18% share, but less than what we are seeing in the US.

But hang on a minute, if Chinese fabless companies are spending similar percentages or less than US counterparts, how does this compare in pure dollar terms? How large is the gap? Let’s compare a few similar US and Chinese companies. 

The largest disparity is between two designers of graphic processing units (GPUs): Nasdaq-listed Nvidia and Shenzhen-listed Jingjia Micro. While Jingjia, founded in 2006, clearly has no plans to replace Nvidia anytime soon, it does promote itself as the creator of a domestic GPU/domestic graphics card. It also spends a healthy 27% of revenue on R&D, slightly more than Nvidia’s 26%.

Jingjia and Guoxin 

As you can imagine, the two companies’ revenues are quite different. Jingjia’s revenue was $100 million (RMB 645.6 million) in 2020, while Nvidia’s was $10.92 billion. In other words, the $27 million Jingjia spent on R&D was equivalent to less than 1% of Nvidia’s R&D budget for the same year. Despite its best efforts and even adding in some government investment, it isn’t likely Jingjia is ever going to compete with Nvidia. Instead, it will likely maintain a niche within China and win some government-related business.

Guoxin Micro is a bit of a mixed bag. It designs microcontrollers, field-programmable gate array (FPGA) IP, and smart card chips, among other things. Let’s take Shenzhen-listed Guoxin as an example of a Chinese FPGA company—a company that makes chips that can be configured after they are produced. Guoxin provides the FPGA intellectual property (IP) for Pango Micro, a Tsinghua Unigroup company. It currently only spends around 11% of revenue on R&D and that is split between various products, not just FPGA IP. For the benefit of the doubt, let’s say all 11% is invested into FPGA R&D; that works out to $55 million. Xilinx invested 27% back into R&D in 2020; that works out to $195.2 million, meaning Guoxin’s investment is around 28% of Xilinx’s. In short: China’s FPGA efforts right now lag far behind and will struggle to catch up if this remains the case.

Okay, so maybe these are two extreme examples, but who seems to be doing better? Well, while Taiwan’s application-specific integrated circuit (ASIC) maker GUC is larger than China’s Verisilicon in terms of revenue ($475 million vs. $260 million), GUC is spending around $89 million, or 19% of its revenue on R&D, while Verisilicon is spending a whopping 35%, or $91 million. It seems as though Verisilicon wants to match GUC’s dollar R&D spend, a wise decision given GUC is essentially part of TSMC, the world’s largest contract chipmaker. Shenzhen-based Goodix, famous for its fingerprint sensor chips, is spending around 26% on R&D, which equates to $260 million, clearly world-leading in this area.

Conclusions: spend more

While R&D spend as a percentage of revenue and total R&D dollar spend are things we can look at to determine how innovative a company may be, they are not everything. Many of these companies will have other sources of money for R&D spending, such as VC investment or government funds that may not show up in these statistics. 

Also, a large proportion of any R&D budget is spent on salaries, and while we see headlines of huge salaries being handed out by some companies in the Chinese semiconductors industry, most companies cannot afford to pay them. On average, salaries are below those in US, European, and Taiwanese counterparks. This means some of these R&D dollars may go further in China than they might in other countries. Indeed, the majority of these companies are expanding their R&D teams quite significantly each year. Goodix added 578 people to its R&D team in 2020, for example, while Verisilicon added 168.

In conclusion, these listed Chinese companies are far behind international competitors when it comes to R&D spend. Only companies with considerable backing such as Cambricon can afford to run at a huge loss in order to reach world-beating status. Many will stay in their corner of the domestic market, largely because that’s all they ever intended. And that’s fine. Yet for Chinese companies that have higher aspirations to catch up with or even surpass foreign competitors, it is essential that they crank up their spending on R&D.

Last week, Chinese processor company Loongson announced plans to release a new instruction set architecture. Loongson is known for processors based on the MIPS architecture, and is linked to the Chinese Academy of Sciences.

The company said that the architecture has done away with “outdated content” found in traditional instruction sets and is more suitable for high-performance, low-power design. The new architecture, it claimed, makes it easier to compile software and develop operating systems or virtual machines. It is also compatible with mainstream instruction sets, so software designed for x86 or Arm should be able to run on LoongArch.

China’s quest for a homegrown CPU

The global CPU market has been dominated by the x86 architecture for years, essentially controlled by two companies, Intel and AMD.

For several years now Chinese companies have been trying to break this duopoly, with some success domestically but definitely not globally. Huawei and Phytium both used the Arm v8 architecture to create powerful 64-core server chips used in data centers and supercomputing. Under US pressure, it is difficult for either company to continue creating such chips.

Hygon and Zhaoxin design x86 processors through joint ventures with AMD and VIA, although Hygon fell into geopolitical trouble as well. Another company, Sunway, has always used the lesser-known, US-designed Alpha architecture, but as far as I know Sunway processors were only ever used within the government.

A few companies, most notably C-Sky and China Core, tried to promote their own architectures or variants of older ones like PowerPC into the commercial market. Both more or less failed and have since latched onto the much talked-about open-source RISC-V architecture. Alibaba acquired C-Sky in 2018. It’s now a leading RISC-V processor company under the name T-Head.

Loongson has always used the MIPS architecture. MIPS ISA has an interesting history, but it is going out of fashion—even its owner, MIPS Technologies, has ditched it in favor of RISC-V.

There has never been a successful Chinese architecture. C-Sky failed to scale and moved to RISC-V. Other companies that claim to be “made in China” have used or use existing open-source or licensable architectures.

Starting from scratch to build an ISA is a big challenge. It’s faster to design your CPU based on a mature architecture, because there is an existing hardware and software ecosystem to latch onto.

However, with Huawei, Phytium, Hygon, and Shenwei on the US entity list, China is worried that it doesn’t have a completely independent architecture. RISC-V may be a great platform for Chinese companies to go overseas with their designs, but it is a global initiative, and in some cases, China may want something that is totally its own.

READ MORE: China’s chipmakers could use RISC-V to reduce impact of US sanctions

No patent infringements

You may be wondering if LoongArch infringes on patents from other architectures. To allay such fears, Loongson paid for a third-party IP agency last year to analyze whether LoongArch infringed on other architectures including Arm, x86, RISC-V, and MIPS. They concluded that the design is unique and independent, that its manual was clearly different to others’, and that it didn’t infringe on Chinese patents for any of the major international architectures.

Perhaps the key phrase here is Chinese patents, rather than global. This may be something to keep an eye on. Loongson says they will analyze international patents as well but have so far concluded that the architecture is completely independent and controllable.

It seems to me that in order to avoid patent infringements and at the same time support emulation of other architectures they have ended up increasing the complexity of their instruction-set: 2,000 instructions is more than other mainstream architectures.

Loongson 3A5000

The Loongson 3A5000 CPU, announced last month, is already using this new architecture and has already been successfully “taped out,” and sent to a fabrication plant for production, at 12nm.

This CPU is aimed at the PC market. The interesting thing here is the process node. Loongson has always used GlobalFoundries to tape out chips based on ST-Micro’s FD-SOI process. One might presume they would continue to use GlobalFoundries for the new generation chip, but they have not announced what process it uses.

Some have said it will use the TSMC 12nm process, while others suggested it could be using SMIC, which now boasts the ability to tape out 12nm. SMIC may be not be ready for mass production yet, but for a test chip, this should not be a problem. This could be a Chinese architecture manufactured at a Chinese fab—just hearsay right now, but something to consider. TSMC or GlobalFoundries are still more likely, as SMIC 12nm would be new to the company, and SMIC has recently come under more restrictions from the US.

It’s also worth noting that Loongson moving from 28nm in previous chips up to 12nm shows development in its design capabilities. It also has a new server chip 3C5000 using the same process but it is said to be much more powerful.

Why not RISC-V?

Since Wave Computing became MIPS Technologies and ditched the MIPS architecture in December, there have been rumors that Loongson would follow. Most in the industry surmised the company would move to RISC-V like many others have. 

RISC-V seems to be the easiest route for a company like Loongson, but there are some reasons why it might have chosen not to. First, there are other companies doing this, so it would be difficult to differentiate. Secondly, it’s clear Loongson wanted something 100% Chinese, not reliant on an international architecture. Finally, Loongson might be planning to follow the RISC-V model and actually open the architecture.

According to its press release, once the IP patent situation is confirmed globally, they plan on creating a LoongArch Alliance where members can access the architecture and Loongson IP cores for free. While the company did not say the instruction set will be open to members, it is certainly possible.

It is rumored though that the company will join the RISC-V consortium. Prior to the LoongArch announcement, executives have said they are “looking forward to join the open-source instruction consortium.” Most thought this meant RISC-V, but they could have been alluding to their own alliance.

I wouldn’t be surprised if the company joined RISC-V. Its own architecture could be used within China for military or government applications, while RISC-V would be a better platform for Loongson to finally go global.

Time will tell

“Only by achieving independence in the root of the instruction system can the software ecosystem chains be broken,” Loongsoon management said in its press release. Such statements make it clear that the main purpose of LoongArch is for China to have its own fully independent instruction-set architecture.

Since C-Sky moved to RISC-V, this hasn’t been the case. While I do not see LoongArch becoming a globally competitive architecture, as ecosystems are difficult to build up, it could be another string in China’s self-reliance bow.

It will also be interesting to see how the LoongArch Alliance develops. Will it open the instruction-set architecture? If cores designs are free, is that just for research or for commercial use as well?

This whole initiative definitely has government support. Loongson came out of the Institute of Computing Technology, China Academy of Sciences, which is still a major shareholder, and a RISC-V consortium member, with one person on its board.

I will be keeping an eye on whether LoongArch infringes on any global patents, any processor benchmarks out there, how its alliance develops, and whether it does make a move to RISC-V in the end as well. It will likely be used in government and military PC and server applications, but can it move beyond that? Time will tell.

Last week, UK-based semiconductor design company Arm announced plans for the next generation of chips. The v9 architecture comes ten years after the release of v8, which is currently the standard used for mobile phone central processing units (CPUs) and many other processors.

There are some good articles on the new features v9 brings to the table, most notably the Realms feature, which promises to increase security by running applications while data is protected from inspection or intrusion by the host or any other software running on that host. The new architecture will also bring AI/ML extensions for AI support across its CPUs, network processing units (NPUs), and graphics processing units (GPUs), and the ability to improve performance by accelerating workloads in a CPU environment in ways that previously required external hard accelerators.

Licensing architecture

I’ve written an overview of major architectures in China elsewhere, but here’s a brief recap: Many Chinese companies design Arm-based chips, but most will license complete Arm cores on a single-use or multi-use basis, so they don’t have to design a core themselves.

More ambitious chip design companies may get an architecture license, which allows the licensee to change the design itself. This is what you need to create a customized core like the Kirin line of phone CPUs, designed by Huawei’s HiSilicon for use in its phones. But it’s difficult to build a core from scratch, so you have to be highly skilled.

Currently, only two companies in China have an architecture license for v8: Huawei and Phytium Technology, a fabless chip design company focused on Arm server chips..

Notably absent

Arm’s press release included several quotes from high-profile partners around the world, including representatives of three major Chinese smartphone brands; Xiaomi CEO Lei Jun, Vivo CTO Shi Yujian, and Oppo Head of Research Levin Liu. Notably absent were Huawei and Phytium Technology.

Both Huawei and Phytium previously bought architectural licenses from Arm, in part as a way to advertise their independence and control. To help them sell such a message Arm also created Arm China, a separate company that has its own issues.

The smartphone makers that did make the press release, have never been architectural licensees of Arm v8. This could change as they look to develop their own chips. All these companies have been investing heavily in building their own internal chip design capabilities.

However, I think for the time being they will stick with application processors from Qualcomm or MediaTek. As part of Arm’s presentation MediaTek announced that its first smartphone chip using the v9 architecture will be available by the end of 2021, sooner than any Chinese handset OEM would be able to design their own. That’s a lot sooner than they’re likely to be able to make their own.

Chinese handset companies will likely license Arm cores for individual designs, such as Xiaomi’s recent image signal processor design. Xiaomi’s previous attempt at an application processor was somewhat of a failure, and it makes sense for the company and others like Oppo and Vivo to focus first on simpler designs that can help them differentiate their products and also help them gain valuable real-world chip design experience.

So what are we to make of the absence of current licensees Huawei and Phytium? Are they not considered key partners, can they license v9 architecture, and does it even make sense for them to?

Can Huawei buy the v9 architecture?

Huawei has struggled to access semiconductors and IP since the US placed it on a list of companies which require licenses to buy US or US-linked technology. The absence of either company in Arm’s presser could imply that one or both won’t be able to upgrade to v9.

In response to such speculation, Arm has said that it can continue to license its IP to China including Huawei, concluding that its IP is of UK-origin and so not subject to the US ban. Ian Smythe, vice-president of solutions marketing at Arm said, “Following a comprehensive review, Arm has determined that its Arm v9 architecture is not subject to the US Export Administration Regulations,” adding that Arm had informed US government agencies of this conclusion.

That might not be the last word for Huawei. Ultimately, the US government may conclude that Arm’s Austin facility, which contributes to a lot of its high-performance architectures, means that Arm’s IP is sufficiently of US-origin to face export restrictions.

Phytium on thin ice

Phytium is not on the export ban list, and as such does not face the same restrictions as Huawei. However, it is on a list of “military-linked” companies that face restrictions on cross-border investments.

Also, the Washington Post reported today that that the Trump administration was planning to put Phytium on an export blacklist, but “ran out of time”. The article also reported Phytium chips are used at supercomputing centers that design advanced weapons systems for the People’s Liberation Army. This could heighten Washington’s scrutiny of the company, potentially leading to sanctions.

My best guess is that they will go ahead and secure a v9 license without much trouble, but they may be trying to keep a low profile in the hope that the US will not decide to target them. Watch this space.

Now or never

An architectural license gives Huawei and Phytium a certain amount of security: Once granted, the license is permanent, meaning Huawei would be able to continue designing new v9 chips indefinitely whatever actions Washington takes. But under present circumstances it might not be too useful.

An architectural license does not mean the licensee is licensing a specific core. They receive a set of specs for Arm’s cores and a testing suite. This allows the licensee to customize their own processor to fit their application. They can make cores that are faster, smaller, or less power hungry than standard Arm cores, or otherwise differentiated from standard Arm licensees.

Qualcomm and Apple rely on such licenses to create their chips, as did Huawei for its Kirin series. There are only a handful of such licensees globally, mainly because it costs a lot and requires a lot of time and internal expertise to create your own custom Arm core, while there are perfectly good cores available to license at a much cheaper price.

A license alone isn’t enough to make chips. If Huawei is able to buy an architectural license and does so, it still has no access to the EDA tools it needs and the fabs to actually manufacture a high-end Arm-based chip.

But it could be now or never. As competing companies move to v9, Huawei’s v8 license will soon be obsolete. It could actually make sense for the company to go in on an architectural license it can’t use for now in the hope that further down the line either restrictions on the company are removed or domestic self-sufficiency gets to a point where Huawei can get back into the high-end chip game.

With Nvidia’s acquisition of Arm also on the horizon, Arm could soon become a US-owned company. It could make sense for Huawei to lock in access to its IP now, although the same concerns could also motivate China to block the deal.

Conclusions

Access to IP is a chokepoint for semiconductors in China. As I’ve written before, RISC-V may help with this to some extent, but it isn’t as mature as Arm yet, and processor cores are just one of many different types of IP within a chip.

Despite RISC-V’s growth, Arm’s v9 architecture will be a core component for handset, server, IoT and automotive chips for the coming decade. For Huawei it may make sense to get in now, while it still can.

For its part, Arm will want to be free to license to Chinese companies and will be happy to take Huawei’s money. However, the reach of the US government can be long and if the Nvidia acquisition goes through I struggle to see companies on the entity list being allowed access. 

Some domestic analysts argue that Huawei should not rely on architectural licenses. “You may get a v9 license this time, but what about v10 or v11, etc? Does endlessly licensing foreign IP mean independence?” (my translation).

It would be strange to see China without any Arm architectural licensees, but that is a prospect.

We may also see new Arm licensees. Perhaps the likes of Oppo or Vivo will decide it makes sense for them. We all know they are investing huge sums into their own IC design capabilities.

As I outlined in my previous article, designing and manufacturing automotive semiconductors that are up to industry standards is difficult. Chinese semiconductor companies have not focused on this area until relatively recently, especially because it has been much easier to scale fast and make money in the consumer market.

The main types of semiconductors that go into a car are control chips, analog and mixed signal power chips, sensors, wireless communications, interface chips, and memory chips. I will concentrate on the areas I think China is growing: MOSFETs, memory, image sensors, and autonomous driving chips. It happens that these are the areas where I have the most hands-on experience.

Power electronics

Power transistors are abundant in the high tech cars of today: windscreen wipers, windows, and sunroofs use metal-oxide-silicon field effect transistors (MOSFETs). Roughly speaking, a MOSFET uses an electric field to control the flow of electrical currents. Metal-oxide-silicon (MOS) is the material they are made of, and field effect transistor (FET) is the type.

MOSFETs are relatively simple and cheap to produce, so automakers use them for controlling and converting electric power—what is known as power electronics. What is making them more and more interesting for China is their use in power electronics for electric vehicles: DC/DC converters, on-board chargers (OBCs) that allow electric and hybrid vehicles to charge from any AC power supply, and traction inverters that convert electricity from the battery to AC power that can be used by the engine. 

As EVs become more common, use cases for newer materials are becoming more apparent. Wide-band gap (WBG) materials like gallium nitride (GaN) and silicon carbide (SiC) in FETs are newly applied in power electronics, and allow for higher voltages, which are required for faster switching speeds. This in turn improves the power conversion efficiency, and therefore the range, of EVs.

Tesla has gone the route of using SiC MOSFETs from ST Micro for its inverter in newer models. It previously used insulated-gate bipolar transistors (IGBTs). Others, like Nexperia, have chosen to use GaN instead.

There are concerns that WBG materials are unreliable, such as being extremely sensitive to gate voltages with absolute maximum values close to recommended operating conditions. But that’s what automotive standards regulate, and some GaN and SiC field effect transistors have already passed the Automotive Electronic Committee’s Q100 and Q101, the basic stress tests that guarantee a certain level of reliability acceptable to automakers. I expect in 2021 we will see them being used in more and more EVs. 

Pricing may be an issue at the beginning because WBG FETs individually are still more expensive than IGBTs or MOSFETs. However, WBG materials can lower overall costs due to the simplification of the surrounding circuitry. As EV brands compete to achieve longer range vehicles, demand will increase and with it will come a reduction in pricing.

In the global power electronics semiconductor market, Nexperia, a spin off of NXP that is now Chinese-owned, makes up about 7-10% of the market, and it accounts for more than 13% of the MOSFET market. It is ranked number two globally for automotive grade MOSFETs behind Infineon. 

Huawei invested in Oriental Semiconductor, a MOSFET IDM, which to date has very limited market share.

The purpose here is not to debate SiC vs. GaN, there are advantages and disadvantages to both, but to make clear that there is a Chinese-owned company, Nexperia, at the forefront of global EV power electronic semiconductors. Nexperia is head to head with famous global names in the industry such as TI, NXP, Infineon, ONSemi, and Rohm. I expect to see Nexperia grow in China along with the domestic EV industry.

Memory

Today’s cars use local memory primarily for infotainment and driver assistance. Automotive grade memory does not account for as much of the memory market as consumer electronics and telecommunications, but it is still a market worth around $10 billion—and growing. The smarter the car, the more memory it needs. Autonomous cars will have to make calculations really quickly, so they will need high-performance local memory.

DRAM, NOR, NAND, and so on, are all memory types used in the industry, but of course must go through stringent testing and pass standards such as AEC-Q100 to be acceptable to automakers. All the usual suspects in memory ICs are prevalent here, Micron, Samsung, and Infineon (Cypress), as well as smaller companies like Macronix and Winbond.

NOR is easier to develop. The market is dominated by Taiwanese companies like Macronix and Winbond, but there are also China mainland companies like Gigadevice doing well. 

Gigadevice’s overall memory market share is about 18%. It has also developed automotive grade products and is a majority shareholder of Changxin Memory Technologies (CXMT), a Hefei-based foundry specialized in DRAM chips. Through CXMT, Gigadevice has a route into the much larger DRAM market. 

Yangtze Memory Technologies (YMTC) is focused on NAND but has yet to produce an automotive grade product. I don’t think it should yet. It has a lot on its plate: Its consumer products are not yet a success, its production capacity is still lacking, and it is facing legal challenges from Micron over patent infringement. Its funder Tsinghua Unigroup has other problems it needs to deal with before expanding into even more new areas: In December it defaulted on $450 million of debt.

DRAM is a more difficult but more rewarding design task; it makes up around half of the memory used in the automotive market. CXMT to date has no automotive grade DRAM product, so that leads us to Integrated Silicon Solutions (ISSI).

ISSI is a US company headquartered in California. Its core competency is in DRAM, SRAM, and NOR flash. Automotive is one of its key markets, boasting customers such as Bosch, Delphi, and Continental. Back in 2015, Cypress Semi (now part of Infineon), looked to acquire ISSI to add DRAM as the last piece of its automotive semiconductor puzzle. Chinese investment vehicle Uphill Investment outbid Cypress and acquired ISSI. 

Fast forward four years and ISSI switched hands again when it was acquired by Chinese fabless company Ingenic in a RMB 7.2 billion deal. This was a somewhat strange deal, in that a small, relatively unsuccessful MIPs-based fabless CPU company acquired a much larger relatively successful US memory company. 

The deal passed CFIUS review, maybe because ISSI was already owned by a Chinese consortium since 2015. By contrast, Tsinghua Ungroup’s attempts to acquire Micron in 2015 were blocked. It is likely that ISSI was not considered as important, and the Committee felt that it couldn’t be seen to block every single tech deal. 

Like Nexperia, this ISSI acquisition gives China another route into DRAM, and also a route into automotive grade products and knowledge transfer of what is actually required to be successful in the industry. 

CMOS image sensors

The use of complementary metal-oxide-semiconductors (CMOS) in the automotive area is driven by growth in autonomous driving applications. CMOS are a type of high-resolution imaging transistor that is used in most cameras, from DLSRs to smartphones. 

Autonomous cars will usually come with a mix of radar, lidar, and CMOS image censors (CIS) to cover all bases. CIS, for example, may not work well in low light conditions and to reach level 5 autonomous driving more and more sensors are needed on a vehicle. A car produced in 2021 may be loaded with 8 image sensors, and this number is only growing. 

In fact, although demand dropped over 2020 due to Covid-19 related externalities, there are now not enough CIS chips in the market to meet demand, and prices are going up over 40% (in Chinese).

As with power electronics and memory, China’s leading image sensor company also came about through acquisition. Omnivision was originally acquired by a consortium of Chinese investment companies in 2015 and then by Chinese company Will Semiconductor in late 2018. The acquisition instantly made Will Semiconductor one of the most valuable Chinese semiconductor companies, which was hardly a household name before. Even today most people in the industry are more familiar with Omnivision than Will, and Omnivision’s headquarters are still in Santa Clara.

In 2019, Omnivision accounted for around 10% of the global $19.3 billion CMOS sensor market. The same year, it was beaten into third place by Samsung with a 21% market share, and king of CMOS Sony with a 42% share. 

But when it comes to CMOS for the automotive sector specifically, Omnivision is doing better than Sony. It holds around 22% of the market, second only to US company ONSemi at 36%, Sony can only muster 10%. 

Technically, Omnivision’s products are just as good as ONSemi or Sony. All these companies offer similar 8.3MP front-view camera CMOS products for autonomous driving. Omnivision sells a lot of its products into European automotive OEMs and is well placed to grow in China as well, especially with demand outstripping supply.

Autonomous driving

You need a chip to process what your sensors are detecting and there are several Chinese startups specializing in this space. Rather than gaining momentum and market share via acquisitions, this area is characterized by established foreign players and local Chinese companies, and it’s highly competitive. 

Startups like Horizon Robotics and Black Sesame face competition not just from the likes of Huawei, with its MDC chip, but also from a whole host of foreign companies that are already more established automotive semiconductor suppliers. These established players have other revenue streams which means that they don’t just rely on the automotive market, or even this specific subsection of it. This allows them to grow into the market without having to burn through investor cash in the hope of future revenues.

One might argue that these Chinese companies have an advantage domestically, but that isn’t necessarily the case. NIO announced last week that it will use Nvidia’s Orin system on a chip (SoC) in its automotive processor (ADAM), indicating that even Chinese carmakers might opt for foreign processors. SoCs are integrated circuits that combine all the main components of a computer; memory, processing, etc. NIO’s ADAM will use four Orin SoCs to push above the 1000 TOPS required for level-5 autonomy. 

At our company we have met with most of the automotive OEMs and “tier-1s”—direct suppliers to OEMs—in China on behalf of our clients. The vast majority are developing autonomous vehicles using foreign SoCs like Nvidia’s Orin. Others we usually come across include Nvidia Xavier, TI’s TDA4X, Japanese Renesas’s V3H, and Ambarella CV22. Sometimes Horizon and Huawei are mentioned, but Black Sesame is nowhere to be found. Based on my experience, even in China, American companies and Renesas are outperforming their local counterparts.

This isn’t to say local companies have no hope. Huawei obviously will face problems supplying high-end autonomous driving SoCs if it continues to face export controls from the US, but I wouldn’t discount them yet. 

Horizon Robotics has already partnered with key tier-1s and some OEMs, including Audi. Its Journey 2 automotive AI processor is said to have shipped 100,000 units, and its level 3-capable Journey 3 is said to be going into mass production in Q3 2021. The company also has a clear roadmap to L5 for its future SoCs.

Conclusions

This is just a snapshot of a part of the industry I have had most contact with. China’s largest and most global players in the automotive chips sector came to be Chinese through acquisition. Some of these acquisitions may have struggled to go through in today’s climate, but the fact they were done earlier shows some foresight on these Chinese companies’ part. At the same time, in fields like autonomous driving, homegrown companies are rising. 

The acquired companies are in a good position to take advantage of the growing EV and AV industries, but the home grown companies may struggle to compete with the size and scale of their foreign counterparts.

Nexperia, ISSI, and Omnivision have all kept their HQs in their respective home countries, but are concurrently operating strong R&D or manufacturing facilities in China—and in my experience Chinese owners are rarely hands off. ISSI and Omnivision have design teams in China, whereas Nexperia operates packaging R&D on the mainland. 

There is nothing nefarious about this, it is quite normal and makes sense. But technical know-how is transferred naturally as part of the work process, so even if these companies switch owners in the future I expect some skills and knowledge will have been transferred to Chinese employees.

China is driving up the automotive value chain, and is shooting to go all the way up to automotive chips. Only a few years ago Chinese cars consisted of cheap knock-offs of western brands—who remembers the SCEO HBJ6474Y? But over the past decade, China has gradually made progress.

The modern car is extremely complex. The average car has at least 50 chips, and electronics account for over 40% of the entire bill of materials. Who is supplying these chips, what do is the chips’ function, and what companies in China are moving into this market?

These questions have been brought to the fore recently as Chinese automotive companies faced a chip supply shortage that has led to some minor production halts.

The pecking order

The global automotive semiconductor market is worth around $41 billion and may grow to $65 billion in the next couple of years. At $41 billion, it accounts for around 12% of the entire semiconductor market.

Less than 3% of global sales of automotive semiconductors come from Chinese companies. European firms make up about 37%, American ones around 30%, and Japanese ones around about 25%. Only one of the 20 top global automotive semiconductor companies is Chinese, and even that one is a spin off from NXP that was acquired by a Chinese company, its headquarters is still in the Netherlands.

With the growing need for autonomous driving capabilities, processing power within cars is increasing. So much so that a car today is more of a computer with wheels.

There is a range of different types of chips in a car, from simple to complex. The main types are control chips, analog and mixed signal power chips, sensors, wireless communications, interface chips, and memory chips.

It is no secret that China has huge automotive ambitions, but why does it still make up such a tiny portion of the overall automotive chip market?

Well, one big reason is that this market is difficult. It’s difficult for a lot of reasons, but not so difficult they can’t be overcome. Any company new to the market needs to be patient and prepared to spend a lot of time not making money before they get anywhere. Some companies used to consumer market chips just aren’t prepared for this.

Product and supplier requirements

Unlike chips for normal consumer products—which China is quite good at designing — automotive chips, like any component going into a vehicle, have much more stringent requirements. Automotive chips must be able to withstand much wider temperature ranges, be resistant to vibrations, shocks, anti-interference, and have very low failure rates.

Automotive companies usually require single digit defects per billion parts, and even sometimes zero defects. By comparison, industrial grade chips usually require less than one part per million, and consumer grade chips a few parts per thousand. All this reliability and consistency, must be achieved at mass production and each part of the product must be traceable, including packaging and even raw materials.

That’s not all.

Having the best and most reliable chip for a certain function out there isn’t always the most important thing for automotive companies. They need to know that the chip manufacturer can keep producing the same chip consistently over a long period of time.

The chip must last not only at least as long as the vehicle is on the road, usually over 15 years, but also be available for as long as the vehicle manufacturer produces the car model, at least 30 years. So, supply chains must be reliable and stable for decades.

Industry standards

To make sure semiconductor suppliers meet the requirements, carmakers require their suppliers to pass industry standards tests. Using these benchmarks, they can identify suitable suppliers. The most common standards are AEC-Q100 for reliability, ISO 26262 for functional safety, and ISO/TS 16949 for quality management.

All these standards make it difficult for any semiconductor company to enter the automotive industry. Completing the relevant tests, submitting the documents, getting certified for all relevant standards for your chip, making sure your suppliers meet the standards too, and then becoming an approved supplier for an carmaker, can take two to three years—at best.

Hidden costs

Manufacturing and legal costs compound on these quality management bills.

The level of quality required in automotive chips means that much of the industry players are integrated design manufacturers (IDMs), meaning that they manufacture chips as well as design them. This ensures that not just the design process is automotive compliant, but also the manufacturing and packaging processes. This means there is much more upfront capital expenditure to enter the market than if one was just setting up a fabless company.

Legal costs can also rack up. Semiconductor suppliers in the car industry often have joint liability if something goes wrong with the chip, and so may bear some costs for product replacement, compensation, and fines. Any company thinking about entering the industry will be overly cautious and may decide it is not worth it.

Even if a new entrant decides it is willing to bear all these costs and passes all the standards requirements, convincing carmakers to buy their chips will be an uphill battle. Older semiconductor suppliers, and carmakers already have strong supply chain relationships that can be very difficult to break into.

Who is doing well and what can China do?

Chinese automotive chip companies can be placed into three main categories; acquired, mature companies moving into automotive space, and newly emerging companies.

China’s largest automotive chip companies have come via acquisition. The likes of Nexperia (acquired by Wingtech), ISSI (acquired by Ingenic), and Omnivision (acquired by Will Semi), are all world leading in their specific fields, MOSFETs, memory, and image sensors respectively. Companies in the second category, like Huawei, or new entrants, like Semidrive and Horizon, are China-focused, for now—but they have global ambitions.

I think it is foreseeable China takes up more of the market, especially domestically. China could even start creating its own automotive standards to make it easier for them.

In the next article I will discuss what some of these Chinese companies are doing in the field of automotive chips, what their plans are, and how successful I believe they will be.

READ MORE: SILICON | Can China make chips?

Arm is barely out of the news these days. When the UK-based semiconductor IP company was acquired by Softbank in 2016, the whole industry wondered what it meant for the company and how it would change. At the time, many Chinese licensees weren’t too happy about having to license key technology from a Japanese corporate overnight. But in the end, Softbank was rather hands off and things went on as normal.

Finally, this week, the US’s Nvidia announced plans to acquire Arm from SoftBank. The deal would combine Nvidia’s world leading GPU and AI capabilities with Arm’s dominance in mobile and edge chipsets, potentially creating a world leading semiconductor company capable of servicing from the cloud all the way to the edge.

But the deal needs approval from the US, UK, and Chinese governments, and China could be the toughest hurdle. The country let Nvidia’s Mellanox acquisition go through, but also famously blocked Qualcomm’s acquisition of NXP. My bet in this case is that they will not allow the Nvidia-Arm deal to go through. Arm architecture is critical to China’s emerging chip design industry, and there’s no upside to China if they’re owned by a US company—and a competitor.

It may take some time before we know for sure—the company says it may be over a year—but if it is approved would it be a success for Nvidia and Arm? What might it mean for the industry? And what will China focus on as it evaluates the deal?

A good deal for Nvidia

It’s easy to see why Nvidia is interested in Arm.

Nvidia’s focus has been on AI in the cloud. Its GPUs are market leaders in this space, with the global top four cloud services using Nvidia GPUs for 97% of their accelerators. This includes Alicloud, which although more diversified, still uses Nvidia for over 80% of its accelerators. Tencent’s gaming as a service offering uses Nvidia GPUs, and even Chinese server manufacturers like market leading Inspur are releasing ever more AI servers based on Nvidia GPUs. However, Nvidia’s business does not cover the edge market: its products are not suitable for handsets, internet of things (IoT) devices, among others. This is the area where Arm is dominant.

Essentially all handsets and the vast majority of IoT devices are based on Arm architecture. Arm licenses its processor cores to companies like Qualcomm, Mediatek, Unisoc, and Hisilicon. The cores are then used as the foundation of the companies’ system on chip (SoC) products like Snapdragon, Helio, S500, and Kirin, not to mention all the IoT chips out there. Through this acquisition Nvidia would gain access to a part of the market it had previously had no presence in.

This isn’t all though. Arm-based CPUs like those from Huawei (Kunpeng), Ampere, Marvell, and Phytium have been attempting to mount a challenge to Intel in high-performance computing (HPC). Intel is currently dominant, with over 90% market share, in HPC CPUs, while Intel’s x86 architecture holds over 98% of the market. But it is feasible a Nvidia + Arm CPU solution could begin to eat into Intel’s HPC market share. Nvidia would have the whole stack—even the networking side, via its acquisition of Mellanox last year.

So, it makes sense for Nvidia, at least if it is playing the long game. Arm’s profits are something like $300 to 400 million per year, meaning if nothing changes it would take over 100 years to make back its reported $40 billion price tag . Not all things can be measured in such a way though, and I expect Nvidia CEO Jensen Huang thinks he can grow this profit and build an all-encompassing ecosystem.

A farewell to Arm?

Arm’s founders certainly aren’t happy about the deal, as they made clear in an open letter/petition called “savearm.co.uk,” asking the British government to block it.

To be a successful silicon IP company, in general, it is expected you aren’t also a competitor.  With Nvidia, Arm would lose its independence.

Nvidia has promised that Arm’s business won’t change, that it will remain independent, continue its open licensing model, and maintain customer neutrality. But other than trust what is there guaranteeing this? Softbank was rather hands off with Arm, but Nvidia is a deep tech company, and it’s my opinion that its management will want to put their own spin on things. Will server chip licensees now see Nvidia as a competitor—and what about automotive chip giants like Qualcomm or NXP? Fabless companies around the world will be having discussions right now about how they plan for this.

One option, as I have written on a few occasions, is RISC-V. I don’t see RISC-V replacing Arm as the core general purpose processor in handset chips just yet, and I also don’t see IoT companies being as concerned about this acquisition as others as Nvidia doesn’t compete in this space.

What I do expect to see happening now is Arm licensees—many of whom are also RISC-V members—stepping up their investment in the open instruction set, and perhaps using it more often in complex heterogenous designs where RISC-V cores may act as accelerators on a chip that has Arm as its main processor. Indeed, I have seen this kind of chip quite often in China already. In the IoT space, I already see companies moving to RISC-V who previously used Arm M-series cores, but this is something that was already happening, and I don’t see the Nvidia acquisition making much difference.

Overall, in the handset, server, and automotive chip markets Arm licensees will be worried, and I expect them to come up with medium-to-long-term contingency plans. There is potential in this time frame—perhaps five to 10 years—for RISC-V to become more feasible in these chips as the main processor, and that could cause Arm/Nvidia problems. I should add RISC-V’s limitation isn’t technical, but more related to building an ecosystem to match Arm’s. It would be wise for Nvidia’s Huang to stick to his word, keep Arm independent or it may be the case we have already reached peak Arm.

China

China is going to “love this deal.”

That’s what Huang told the EE Times, promising that the structure of Arm’s China JV won’t change. I can’t say I agree though, unless he knows something the rest of us mortals don’t.

Chinese media have reported that Arm may be able to resolve the crisis with its China JV before the sale. According to The Paper, they’ve reached a compromise under which Arm will drop charges against Wu and he will step down soon. But this was contradicted by recent news showing that an outgoing investment company owned by Allen Wu “Ningbo Meishan Bonded Port Area ARM Investment Management Partnership” is suing Arm China.

So the JV stays how it is and Arm gets rid of the CEO it doesn’t like. It doesn’t sound like Nvidia or Arm is giving much to China. Huang hasn’t addressed the elephant in the room—if the deal goes through, China’s AI chips companies will be dependent on a competitor for key IP, and Arm will be US owned!

The merged company would threaten some of China’s star AI companies. AI, and especially AI semiconductors, are a key part of China’s self-development strategy, and China has some strong companies in the inference space like Horizon Robotics, Intellifusion, Iluvatar, Sensetime, Artosyn, and others all using Arm architecture.

Nvidia will have strong AI training capabilities with its GPU but now also strong inference capabilities, giving it an advantage over these companies. Perhaps this will force some firms to move to RISC-V for future designs. AI companies like Canaan have already done so.

READ MORE: China’s AI chip startups: how many will survive?

Worse from China’s point of view, the deal would make Arm’s IP US-owned, as Washington is cutting off Chinese companies like Huawei from American-owned technology. Along with fab equipment and electronic design automation (EDA) tools , semiconductor IP is an area where the US is strong and China is weak, and Arm is the world’s largest silicon IP company. More than 40% global IP sales are from Arm, and 95% of chips designed in China use its IP. In short, it adds another weapon for the US to use against China.

As mentioned in the savearm.co.uk letter, Arm would become subject to US Treasury Department regulations. This could mean that any device in the world using Arm architecture will have to comply with these regulations, potentially giving the US government the ability to cut off Chinese companies it doesn’t like not only from designing chips using Arm IP, but also from buying semiconductors with these chips in (e.g., Oppo buying Snapdragons), or buying a device that uses Arm chips (e.g. a Chinese IoT company buying Telit IoT modules). It could all be blocked if the US government choses—at least, there is risk of this.

Does this sound like a deal the Chinese government will like? I don’t think so, and I struggle to think what it would want in return for letting such a deal through.

In recent years, handsets have been key to semiconductor industry growth. So when analysts predicted a grim 2020 for the handset markets, things didn’t look great for semiconductor companies either. In December 2019, analysts expected handset sales to drop 2% to 270 million units in 2020.

Predictions for the rest of the year are weaker, but still miles ahead of the handset market. Some analysts expect a 5-10% drop in global chip sales for 2020 as wireless, automotive, industrial, and general consumer electronics sales all fall. Others predict 3.3% growth for the whole year. It might not be huge growth, but it’s growth nonetheless.

If consumer electronics sales are in freefall, what’s keeping the semiconductor industry from dropping further, even giving it hope of growth—and is this the opportunity Chinese chip makers have been waiting for to make their mark on the industry?

TSMC saw sales of every chip it manufactures drop in volume in H1 2020 growth. Except for one, which grew by 12%: high-end server computer chips.

Less phones, more laptops

While the Covid-19 pandemic accentuated the trend of falling handset sales, it put fuel on the fire of cloud computing.

Cloud services were already moving data storage and processing from edge devices, like phones and laptops, to data centers. But then the pandemic and lockdowns made work from home the norm around the world.

This has not only led to an increase in PC and laptop sales, which grew 11.2% year-on-year globally in Q2 2020, but also an increase in the use of video conferencing, distance learning, and video streaming services.

In China, up to 300 million people were working from home in the first quarter of 2020—and tech companies jumped at the opportunity. Virtually every major internet company brought out new apps to deal with this demand.

Baidu brought out its collaboration tool Baidu Hi. Alibaba released DingTalk 5.0. Bytedance created Feishu. Tencent pushed Tencent Meeting (which it had luckily just released in December 2019). Even Sohu and Pinduoduo got in on the action with Little E and Knock. 

This surge in remote working services has led to a surge in internet traffic, which demands more processing power from cloud providers. More processing power needs more servers, and servers are made of chips: general-purpose CPUs and accelerators like graphics processing units (GPUs) and field-programmable gate arrays (FPGAs).

READ MORE:  SILICON | China’s progress on homegrown CPUs

I don’t believe we will see internet usage drop back to pre-pandemic levels. The amount of data collected by individuals and companies has been exploding for a while now, and it will only grow.

The sudden change in human behavior brought by Covid-19, along with increasing workloads and 5G connectivity, represents an opportunity for Chinese companies to break into a market dominated by Intel, AMD, and Nvidia.

Do one thing, do it well

More people working from home doesn’t just mean more servers; it means a greater mix of servers to cater to the varying needs of different applications. Some servers need to be flexible; some need to be low-cost, and some need to have specific accelerators designed for specific applications.

Not only will the world need more chips, but it will need a greater variety of them. This gives Chinese companies the chance to pick a market and develop a product.

Different types of chips have made their way into the server space in recent years and ever increasingly so. General purpose Central Processing Units (CPUs) aren’t suitable for some applications, so they need help from various different kinds of accelerators.

Accelerators are processors to which the main general-purpose CPU offloads some workload. Graphics processing units (GPUs), used for image and video processing, and more flexible field-programmable gate arrays (FPGAs), and application specific integrated circuits (ASICs) are the main types of accelerators in use.

In 2012, Nvidia found that its GPUs, normally used for processing images and video, were great for AI applications. It has ridden the AI wave to now be worth more than Intel. Some applications have required more flexibility, so FPGAs from Xilinx and Intel have also made their way into data centers.

Chinese Jingjia Micro, and recently Zhaoxin, are working on GPUs, but at this stage they are low-end laptop/PC offerings that don’t meet the demands of servers. The same can be said for Gowin, Anlogic, Pango, and others doing FPGAs; Chinese players are still far behind the likes of Intel, Xilinx, and Achronix.

READ MORE: China’s first homegrown x86 PCs are here, but don’t get too excited

Sometimes it makes sense to create a chip for a specific purpose and to do that one thing really well. Enter Application-Specific Integrated Circuits (ASICs).

China’s chip sector has proven to hold its own in at least one type of server ASIC: cryptocurrency mining rigs. Bitmain and Canaan are the world’s top producers of crypto mining equipment. This suggests that it is possible for China to lead innovation in at least one kind of server chip.

But many companies have popped up in China looking to ride the AI server ASIC wave in recent years, and none have found great success yet. Like most industries in China, lots of people jump on the bandwagon and make large profits difficult for one another. Many will die, but a few will survive and prosper.

The old ISA conundrum?

While ASICs are probably the best opening, Chinese companies in the server space now are focused on general-purpose CPUs. Companies working on both types of chips have chosen a variety of Instruction Set Architectures (ISAs).

Instruction set architectures (ISAs) are a set of instructions that control communication between software and hardware in processors. They are owned and licensed by western companies, which means Chinese chipmakers rely on deals with IP licensing firms like the UK’s Arm.

Can Chinese companies even begin to make inroads into a market that is 98% x86 architecture, of which almost 90% is Intel and 10% AMD?

It’s difficult, for all the same reasons why Chinese companies can’t wrangle US superiority in semiconductors.

Whether because of luck, economic planning, or market forces, a couple of companies have emerged around each ISA, spreading China’s bets. Huawei and Phytium are using Arm; Zhaoxin and Montage are using x86 (I consider Hygon defunct); Loongson is using MIPS; and Sunway, something else altogether, possibly developed in-house.

Huawei’s Hisilicon has been by far the most successful in the server CPU space. Some Chinese analysts say it may sell 1.5 to 2 million of its Kunpeng server chips this year. Its Taishan server chip might see its market share grow to 3% share globally by the end of 2021. We all know Huawei’s current troubles, so such predictions aren’t exactly watertight.

One way out of the ISA conundrum, as I’ve written before, is using the RISC-V open-source architecture. Huawei and others are jumping on the bandwagon, trying to develop high-performing chips using the free-to-use architecture, and should continue to. It won’t be a fast transition.

READ MORE: China’s chipmakers could use RISC-V to reduce impact of US sanctions

But when it happens, it will remove one key weapon from the US arsenal. The US won’t be able to block Huawei and other Chinese companies from getting their hands on the fundamental architecture.

However, even if one of them created a CPU, based on any of these ISAs, that was superior in power, performance, and area, there are other barriers to entry.

Snatching some of the global market share is not just about having a great performing chip. The software, applications, standards bodies, etc. create an entire ecosystem that can help customers integrate, optimize, and get to market faster.

Huawei has tried to create such an ecosystem by opening up OS source code, compilers, tools, etc. it has done better than any other Chinese company, but still relies on the Arm ecosystem.

Too many cooks

Whether we like it or not China is looking to design and manufacture homegrown chips to replace US imports, and server chips are key to this. The stability and growth in this market means it’s ripe for investment, even if barriers to entry are high.

Making server chips is a long and painful process. But the Chinese companies listed above have identified the cloud as an opportunity and have been investing in it.

One extreme example of trends in China’s server chips industry came from Tencent earlier this year. The Shenzhen company announced it would buy 1 million servers over the next five years, spending around $70 billion.

Domestic demand for server chips will only grow in the coming years. With government preference for domestic chips in this industry and a need for custom accelerators, it could be one of the better semiconductor verticals for Chinese companies to build a customer base in.

However, like many industries in China, there is increasing risk of over-fragmentation, which will make it difficult for everyone to make solid profits. Inspur and Sugon, two leading Chinese server companies, recently set up their own chip divisions, adding to market fragmentation. While I doubt they will start with CPU design as their first foray, it might be coming in later years.

It remains to be seen if Chinese chip makers can compete internationally. But increasing revenues from China will give them better footing to go about global business development, especially in China-friendly countries.

Chinese companies need to pick their fights. Competing in the general-purpose CPU space is an uphill battle. RISC-V could provide a long-term self-reliant option, but dominant players Intel and AMD are strong competition.

Custom ASICs for specific tasks is where China already has plenty of talent and companies that are up to the task. While it would be nice to see extra competition in the GPU and FPGA space, Chinese companies here face the same barriers as those creating general-purpose CPUs.

Most importantly, ecosystems need to be built. It is a difficult and time-consuming endeavor. Even a recent SOE I met with preferred to use Intel, simply because he understood it; it works, it’s mature.

With this mindset, even grabbing the domestic market is a far cry from where we are now. The government needs to step in and use “Made in China” incentives.

China has talented engineers in the ASIC and FPGA design space, but there are simply too many companies for any one of them to have the economies of scale and R&D spend to truly compete. Consolidation and collaboration are needed if Chinese design companies are going to seize the server opportunity.

New export ban rules announced by the US Department of Commerce could be the blow that finally incapacitates Huawei, cutting off its ability to create advanced semiconductors.

Despite these restrictions, Huawei has continued to use US technology.  

Hisilicon has been able to continue designing chips, relying on existing licenses from key US Electronic Design Automation (EDA) tool companies Synopsys and Cadence. The rules limited these companies’ ability to provide updates, patches, and technical support, but Hisilicon could continue using the software, even if it wasn’t quite up to date.

It also didn’t block Huawei from contracting chip fabrication to Taiwan Semiconductor Manufacturing Company (TSMC), a Taiwanese company, and to Semiconductor Manufacturing International Corporation (SMIC), a Chinese company, both of which use US equipment in their production lines.

The rules did prevent Huawei from offering Google Services on its phones, a significant blow that led to reviews like “a stunning phone you shouldn’t buy.” Indeed, Huawei’s handset shipments have started to suffer outside of China, but overall sales are up due to a huge increase in domestic demand—where Google Services are not allowed anyway.

The US now says it will apply the rules to indirect relationships like TSMC, meaning anyone in the world using US technology or software to design or manufacture semiconductors for Huawei must now obtain licenses from the US.

To directly quote from the briefing, “Huawei benefited from a loophole that allowed it to make use of US electronic design software and manufacturing equipment to continue to produce its own semiconductors. That ends today.”

I’m not a lawyer, and I can’t tell you if this version of the ban is watertight. People are already suggesting loopholes on Twitter. But at the end of the day, TSMC can’t afford to give up on its US market and will comply if its lawyers can’t find a work around. SMIC, as a Chinese company, could be another story.

How does this affect Huawei?

Under these rules, many more suppliers will need a US license to work with Huawei. Fabs owned by TSMC and Samsung will need a license; many semiconductor IP companies, even some non-US ones, will need a license; outsourced design service companies will need a license. One would assume that most of the time the US will deny licenses, or at least hold the threat of denial over China and Huawei if they don’t play ball.

The process of making a semiconductor is complicated and has multiple phases, going from raw materials, to design, to fabrication, to packaging and assembly. The new rules threaten Huawei’s ability to make chips in two central phases, by limiting access to fabrication plants (“fabs”) and preventing the use of EDA tools in design.

And the rules in theory also affect Chinese companies. Just like TSMC, SMIC will have to apply for a license to manufacture Huawei’s chips, as it too uses US equipment. Imagination Technologies may be Chinese owned these days, but it still uses US EDA tools, so its IP couldn’t be used by Huawei without a license unless the company moves away from these tools. All Chinese design service companies, such as Verisilicon, use these tools—there just aren’t any realistic alternatives.

Huawei is said to have prepared by stockpiling a lot of chips, and the rules came with a 120-day reprieve for orders already in place, so Huawei’s next Kirin chip (Kirin 1000), which is in production at TSMC, should be good to go. Production is expected to stop by mid-September, so I imagine Huawei will look to manufacture as many of this chip at TSMC as possible between now and then. Plans for the 5nm Kirin 1100 for next year may have to be scrapped, as only TSMC can do this. Any future high-end designs at 7nm and 5nm will have to be scrapped or moved to another less advanced process.

Of course, even this is possible only if there is a fab that can set that up without US equipment in that time period. There isn’t an obvious loophole.

The fab problem

Not having Google Services is one thing, but if you don’t have a chip you don’t have a product, even for the domestic market.

But Huawei has surprised us before and may continue to do so. It would have known this was coming, and as such will have some contingencies. But it’s hard to conceive of a plan that would cover this situation. Huawei does have a stockpile, but you can’t stockpile chips that haven’t been manufactured yet.

Without loopholes, there are more or less no existing fabs that can work with Huawei for now. In the short-term, this means Huawei has nowhere to manufacture its chips. In the medium-to-long term, there are some answers, if costly ones.

One, TSMC, Samsung, SMIC, and other fabs could create Huawei-specific, or China-specific, production lines with zero US equipment. This would be a huge investment just to deal with one customer, but if US restrictions spread to all Chinese companies it could make sense economically. Even Huawei alone could still make sense to TSMC, which relies on Huawei for 10-15% of sales, but this risks the wrath of the US government. The Chinese government might also push SMIC to set up a non-US line. It announced a $2.2 billion investment into SMIC straight after the US announcement, perhaps to create such a production line, but this wouldn’t replace TSMC’s 5nm and 7nm, and current SMIC free capacity is not enough to deal with orders from Huawei.

Two, Huawei could start fabricating its own chips, like Intel or Samsung. It would have to create its own chip production line free of US equipment. This isn’t something that can happen overnight, and would not be cheap, but would give it more control.

It makes more sense for it to work closer with domestic fabs to create US-free production lines, as it is more economical and lets both companies focus on their core expertise. Either option could result in no longer having access to leading edge process and so a worse product than today.

But either option could fail, depending on what the US does with international equipment makers. While there are non-US manufacturers, they are probably vulnerable to US pressure just as TSMC and Samsung are. Leading Dutch equipment company ASML has previously followed US export rules, and without ASML you can’t have a high-end chip.

The EDA problem

On the EDA front, Huawei’s research and innovation lab, called the 2012 lab, has been rumored to be working on its own set of tools. It is unclear how ready these are, but this could be one area where the company surprises us all.

Read more: SILICON | China’s design tools conundrum

Domestic tool companies already have tools for certain parts of the design flow, but nothing that covers the entire design process from architectural exploration, to RTL verification, to physical design, etc. The Department of Commerce has made it clear it wants to stop Huawei using Synopsys, Cadence, and Mentor tools, and I interpret the following to mean its partners can’t use them either to supply Huawei with design services or silicon IP:

This expanded rule will impose a US licensing requirement, an export-control licensing requirement whenever anyone anywhere in the world uses US technology or software to design or produce semiconductors for Huawei. Companies wishing to sell certain items to Huawei produced with US technology must now obtain a license from the United States.

That brings us to the IP problem. Although Huawei has a make rather than buy philosophy, it does rely on several IP companies that will be affected by the new rules, and these IP companies often use US EDA tools to design their IP. Although Arm cores were previously deemed to be UK origin technology and Huawei could continue to access Arm v8 and v9 architectures Arm uses EDA tools from companies like Synopsys, so could Arm IP be back on the chopping block? 

As I have written before, the new open source architecture RISC-V could be Huawei’s way out here. But while RISC-V is growing fast and is extremely versatile, its ecosystem does not match Arm’s yet, so it will be a few years before it is viable in consumer electronics like handsets.

But it’s not just Arm. There’s a lot of scattered IP in a design: the GPU, communication interfaces, on-chip monitors, etc. In addition to EDA, Synopsys, is also an IP provider. Just one example is its USB IP: Huawei uses Synopsys USB 2 and USB 3 PHY IP, and it no longer can. This IP is not something that can just be designed overnight, and Huawei will need to find an alternative that doesn’t come from a company using US EDA tools.

System-on-a-chip design companies like Hisilicon invariably rely on IP for some parts of their designs, in order to speed up the design process and create the best performing design possible. For some IP, it seems Huawei will have to design itself using a mix of its own tools and other domestic tools, as well as encourage its non-US suppliers to verify RTL using other tools. I spoke to one non-US IP company, whose lawyer confirmed it won’t be hit by the rules and can carry on licensing to Huawei, so there will be some IP suppliers Huawei can still rely on.

Production, EDA, and to a lesser extent, IP are the three main areas of concern, but there are many others, field programmable gate arrays and emulators for chip prototyping being one, 5G test and measurement equipment being another.

Deus ex Biden unlikely

I expect the US tech lobby will be going crazy right now, as many companies are not just losing their Huawei business, but also Chinese business. The first question I often get asked in meetings these days is: “is your IP American?”

The environment in the US is very anti-China. As Trump and Biden attack each other as soft on China, they’re bidding up the confrontational attitude, and I don’t expect that to change any time soon. There may be change rhetorically, but not in US policy goals.

China has, of course, been just as brash and undiplomatic. Its media are calling for a strong counterattack, and the government itself has threatened to use what it calls its “Unreliable Entity List.” That threat hasn’t been made concrete at this time, but Apple, Boeing, Qualcomm, and Cisco are have been rumored as targets.

I have considered some options for Huawei here, but none of them feel very realistic or short-term. The best option would be to find a way to avoid the ban coming into force.

Meanwhile, TSMC will be trying everything it can to help. Losing Huawei’s business would also a huge blow for the Taiwanese fab, and I’m sure it will be taking legal advice as well as lobbying the US government to let it continue its work with Huawei. Its announced fab in Arizona seems not entirely certain, so this investment could be used for leverage in any negotiations.

Will there be enough chips?

A worst-case scenario sees Huawei without enough chips for its next flagship product, and stuck with either nowhere to manufacture chips, or an inferior US-equipment free production line that means future products are no longer world leading, possibly after many months of interrupted production. That’s not even considering a scenario where the company can’t even viably design chips at all.

I’ve focused on phones, but Huawei also needs the chips it designs for all its other product lines: servers, laptops, switches, base stations, AI, cameras, etc. It will have been stockpiling a lot of these, especially 5G base station chips, but it still faces the problems highlighted above across all its product lines.

The company has surprised us before though, and perhaps it can again. But if it can’t, then we can expect a very different Huawei, and US tech can look forward to retaliation in China.

Recently I have written about China’s problems with EDA tools and chip fabrication. It isn’t all doom and gloom for China, though, and I’d like to talk about some areas where China is doing better. Central processing units (CPUs) are a bright spot.

Much has been written about the Sino-US trade war, especially about Huawei, with a lot of this discussion revolving around access to x86 CPUs from Intel and AMD, and also access to Arm core IP. Currently, Huawei and others still seem to have some access to Intel and Arm technology, and probably will unless sanctions become more robust. If Huawei and other Chinese companies lose access completely, while it would be a heavy blow, it wouldn’t necessarily be a death nail. Chinese companies such as Loongson, Phytium, Huawei, Zhaoxin, and Alibaba have developed CPUs and core IP which can fill in some of the gaps that would appear.

What are CPUs?

Just so we are all on the same page, let’s quickly describe a CPU. The CPU runs the OS and various applications on a device, processing data and giving an output. CPUs used to contain one processor but now usually contain more than one processor (cores). You will have heard of dual-core, quad-core, etc. For example, the latest Qualcomm chips are octa-core based on Arm Cortex processor cores, just like Huawei’s Kirin chip. A computer may even have more than one CPU, each with multiple cores. Multiples CPUs are more common in the server space.

To make a CPU, one of the first things a design company will decide is the core their CPU will use and thus also choosing the instruction set architecture (ISA), an instruction set provides commands to the processor, it is the link between software and hardware. Software designed for one ISA may not work on another without emulation.

ISAs come in two basic types: faster CISC (complex instruction-set computer) architectures, and more power-efficient RISC (reduced instruction-set computer) architectures. A CISC ISA can do multiple things in a single instruction whereas a RISC ISA may need multiple different instructions to complete the same task. Arm, MIPS, and RISC-V are RISC architectures, while x86 is CISC.

Traditionally, if you’re building a server, you need performance and you’ll choose CISC for its speed. If you’re building a phone, you’ll choose RISC.

Once you’ve chosen CISC or RISC, you have to pick a specific processor core to base your design on, and thus the ISA. Choosing a core commits your CPU to its ecosystem of compatible tools, apps, middleware, etc., so it can be hard to switch if your team has already become accustomed to a certain ecosystem. I’ll look at the field by dividing it into processor cores or ISAs.

Arm

The smartphone industry grew with Arm and its ecosystem and today basically all handsets use an application processor based on the Arm architecture.

Although Qualcomm left the Arm server market, along with its China JV Huaxintong, and there is endless debate as to whether Arm can ever replace x86 in the server space, there are still a number of western companies such as Marvell, Amazon, and Ampere developing Arm server chips, and for China, it has become even more important to develop them.

In the Arm camp we have Huawei, and the lesser known Phytium. With the sanctions ongoing rumors have arisen that Huawei’s HPC Compute and HPC Storage business lines may be closed sometime this year as they have limited access to Intel CPUs and Intel support. This has increased Huawei’s efforts to replace Intel with its own Arm-based server grade CPUs. The fruit of these efforts to date was the impressive Kunpeng 920. One of the most powerful Arm CPUs in the market Kunpeng boasts 64 cores running at 2.6Ghz, at 7nm. The chip will mainly target cloud services and big data applications in Huawei’s Taishan server range. And while I believe it and future generations of the chip can be a success in this space it does not address Huawei’s concerns at the bleeding edge of HPC or even supercomputing.

The other Arm-based CPU player is China Electronics Corporation subsidiary Phytium. The company has a good relationship with the Kylin OS team, their offices are next to one another, along with the Beidou team. Indeed, much of Phytium’s team is also part of the National University of Defense Technology (NUDT). Although its HQ is in Tianjin, much of the R&D is in Changsha, near NUDT. Phytium has a range of desktop PC and server CPUs but has more of an HPC focus. Its latest Arm chip is also 64 core but runs at around 2.2Ghz and uses a 16nm process. Phytium’s claim to fame is of course its use in supercomputers like the Tianhe-2, but actually the majority of the processing here was done by Intel Xeon and Intel Phi CPUs, Phytium’s was mainly processing front-end tasks. This changed though due to an Obama ban resulting in the removal of Intel Phi, NUDT replacing it with its self-developed 128 core Matrix-2000 processor. This wasn’t as powerful as Intel’s latest offering at the time, but still more powerful than the older Intel Phi processors they were replacing. NUDT and Phytium while cannot instantly replace HPC workloads are moving strongly in that direction.

MIPS

MIPS, like Arm, is based on a RISC ISA, but it’s less successful. It never got traction in the handset world, but it’s got a long history in China.

The first commercial Chinese CPU I know of (please message me if you know of earlier commercial CPUs), was the Loongson (aka Godson) which began life in the Institute of Computing Technology (ICT) back in 2001 and was commercialized by Loongson and Lemote. You can Google image some nice pictures of the Lemote laptop and desktop PC. These days, its latest CPUs are based on MIPS architecture which is now owned by Waves Computing out of the US and are fabricated by ST Microelectronics from Europe. The latest Longsoon CPU runs at 2Ghz and has four cores on a 28nm FD-SOI process, fine for government civil servant desktop use. For next year it does have some plans to move to a 16 core 2.5Ghz 16nm CPU, not world beating, but a big improvement. Whilst it isn’t going to become a household name it can be used in certain government or military use cases.

While Loongson may have been the first, it’s the only general purpose CPU right now based on MIPS, another company, Ingenic is also using MIPS but for low-power application specific processors. Other Chinese CPU players base their designs off Arm, x86, or RISC-V.

X86

Intel’s x86 is the only major option for CISC, and as such dominates laptop and PC CPUs, and has a leading position in servers.

Some of you may have read my article on AMD’s JVs in China, so will have seen the name of Zhaoxin Microelectronics before. Zhaoxin is a JV between Taiwan’s VIA Technologies and the Shanghai government. Along with Intel and AMD, Zhaoxin has access to the x86 ISA. While I do not see the company competing with its peers at the high-end, Zhaoxin is showing progress in the low-end server, desktop PC, and notebook space. Its latest KX-6000 16nm, 8-core, 3Ghz CPU performs on par with a 2017 Intel i5, enough for China to be independent at the low-end. Zhaoxin is planning a 7nm CPU for 2020 as well. Let’s keep an eye out for performance figures.

Zhaoxin, along with AMD’s JV (THATIC), discussed in my previous article, are China’s attempt to remain independent when it comes to x86 CPUs. Right now, THATIC is limited to AMD’s older Zen architecture, rather than the newer Zen 2, and in all honesty its Dhyana CPU is said to be more or less a rebranded EPYC CPU. Rather than being world leading, the aim has just been to have access to decent x86 CPUs domestically.

Zhaoxin has more freedom to develop than THATIC so I expect to see more from it in the coming years, where as THATIC seems to be in more of a frozen state given restrictions placed upon AMD.

RISC-V

For now, RISC-V, is the open source budget option, mainly seen in low margin IoT applications where price beats ecosystem. But I foresee that changing as China pushes to reduce exposure to US export regulations, its ecosystem grows, and is proven in more and more applications.

I have previously written about the importance of RISC-V to China. In my opinion it is China’s best bet at becoming relatively self-reliant in CPUs. Details are in my previous article, but in brief, despite originating in the US, RISC-V is sanction proof and can be used as a base for CPUs ranging from low-power IoT to high-performance computing.

Whilst there are many companies working on RISC-V based SoCs, MCUs, and CPUs, Alibaba gained most of the headlines in 2019 when its new subsidiary T-Head announced the RISC-V-based c910 processor and RISC-V-based Hanguang 800 AI processor. The C910 CPU and Hanguang NPU help Alibaba improve AI features in its cloud services and allow it become less reliant on foreign chip suppliers like Nvidia, as well as giving it new revenue streams from by potentially selling its CPUs or licensing core IP to third parties.

Conclusion

China is on the road to CPU independence in the low and mid end, but it will be a few years yet before it can be independent in the high-performance computing space, but I am confident it can get there. The current geopolitical situation is only going to speed up the process and even if true, loophole-free sanctions are put into place China will only be slowed down temporarily, life finds a way. China will struggle to break into the top tier with most architectures, which must be licensed and do not allow licensees access to source code. RISC-V gives it the opportunity to do so. Ironically, it is a US open-source processor movement that has provided this opportunity to China.

The coronavirus epidemic is affecting every industry, and semiconductors are no exception.

While most industries have shut down, necessities in the medical, food, and logistics industries have carried on working. Semiconductors are one of the industries that have carried on production—even in Wuhan itself. There couldn’t be more of a striking example as to how important the semiconductor industry is to the Chinese government. It can’t stop for a week, even for covid-19.

Will this crisis have lasting effects on the Chinese semiconductor industry?

Wuhan memory

Two key companies in China’s semiconductor plans, YMTC and XMC, are located in Wuhan. For years, China has complained about the US-Korea-Japan memory cartel, and what it sees as price fixing. Since virtually every electronic device we use requires some form of memory, and China is the largest manufacturer of said devices, buying memory from the likes of Micron, SK Hynix, Samsung, and Toshiba is one of the key factors adding to China’s semiconductor deficit.

While there are other key Chinese memory companies—e.g. Changxin memory in Hefei—the two Wuhan memory makers have both have confirmed they are not stopping production. While the rest of us are forced to work from home due to covid-19 fears, two companies at the heart of the epidemic cannot be allowed to discontinue production.

It may sound crazy to carry on working in Wuhan right now, but shutting down a fab, even temporarily, is very expensive. Fabs usually run 365 days a year. They may sometimes undertake “warm” shutdowns for a few days for maintenance work, but almost never come to a complete stop. In a warm shutdown, the machines are kept on doing dummy runs to keep the equipment stable so production can continue straight away, meaning staff must be on site.

To get staff out of the plants would mean a full “cold” shut down in which all equipment is turned off. Some parts of the process can’t be stopped without destroying product. Once turned back on, it takes a number of dummy runs for each equipment before they can get back to normal. Since this could mean over a month to restart, cold shutdowns nearly never happen, and companies will do nearly anything to avoid them.

The companies claim to be taking all precautions necessary to ensure no infected employees return, including having employees live on-site, but from what we know about covid-19 it may be hard to even know one is infected for up to 14 days. Of course, in fabs everyone wears masks, goggles, and gloves all the time anyway. Let’s just hope they don’t face a shortage like the rest of the country.

YMTC claims production won’t be affected. This may be true, but I would worry about the company’s long-term goals. Chinese memory companies have made news with the large pay packets they rely on to attract experienced Korean, Japanese, and Taiwanese employees with, but even these may not be able to convince these employees to come back after they’ve been evacuated from a disaster zone. This may affect these companies’ long-term goals to catch up with their international rivals. MYMTC’s goal of moving from 64-layer NAND flash to 96-layer and above becomes a lot more difficult without international talent. Perhaps they can set up an offsite R&D center for such talent to work remotely from outside Wuhan. Moving staff out of Wuhan for a long while may be necessary if they are to retain talent but may not be workable from a practical standpoint.

China is taking a gamble here. While a cold shut down would be a big set back for China’s memory ambitions, even one infected employee on a production line would be much worse. Not knowing who could have been cross infected or what surfaces are contaminated would mean a much longer shutdown. Since it may be months before Wuhan returns to any kind of normalcy, perhaps carrying on is a risk worth taking.

HiSilicon doesn’t stop

Huawei’s HiSilicon is perhaps just as important to China’s semiconductor plans as memory self-sufficiency. It’s China’s largest semiconductor design company and key for the country’s wireless chip development. It’s no surprise then that, as contacts have confirmed, many of the Guangdong company’s employees continued working at their offices throughout the government-mandated “work from home” period. HiSilicon is a fabless semiconductor design company, not a fab, but a physical presence on-site is still necessary for much of the work involved. This is a highly secure company—they don’t just let employees log into their workstations from home.

With this being a key year for China and 5G it is even more important Huawei keeps the wheels moving. The big trade show of the year, MWC, may be difficult for some key Huawei employees to attend. MWC is requiring proof that all travelers have not been in China for at least 14 days before the event. This means the event will be difficult for any Chinese company to visit or exhibit. It will be interesting to see how this plays out.

Conclusion

It seems in Wuhan they have concluded the costs of a cold shutdown outweigh any risk of infection at these facilities. Fabs are the cleanest places around, so they should know what they are doing when it comes to prevention.

I don’t expect Hubei’s memory situation to have any effect globally. The main effect on memory pricing will be more related to the electronics industry as a whole slowing, and thus memory demand dropping. However, in the long-term, we may see these companies struggle to retain international talent, and thus lose ground in their bids to catch up.

HiSilicon has been dealing with enough troubles as of late. This is yet another, and affects all in the industry in China, but is a challenge that I believe they are best placed to cope with. Shenzhen is a hotbed for covid-19 cases, though, so they need to be careful.

Without any doubt though we can conclude that like supermarkets, delivery services, and the medical industry, the semiconductor industry is one key sector the government cannot allow to slow at any cost.

There has been a lot of hoo-hah recently about Huawei semiconductor subsidiary HiSilicon moving away from being captive to its parent company and selling chips to the open market. This story started gaining traction around September last year and became more mainstream in December, when the IC design firm showed its 4G Balong chip at the Elexcon Expo in Shenzhen. This led to speculation that HiSilicon would also be releasing its Kirin mobile chipset to the open market as well, and this was a “new strategy” for HiSilicon that would pressure Qualcomm. But the truth is there is no new strategy, HiSilicon always had a non-captive side, and has no plans to release 5G Kirin chips to the open market. In fact, there probably isn’t even a market for them.

HiSilicon never was 100% captive

Most people assume HiSilicon is captive to Huawei. The media always refers to it as such. People are most familiar with its Kirin series of handset chips and would be correct in saying these are for Huawei’s use only, but this isn’t all HiSilicon does. Its Kirin (handset Application Processor), Balong (handset baseband), Tiangang (5G base station), Ascend (AI), and Kunpeng (server) chips are captive to Huawei, but for many years now it has sold a whole range of chips into the open market: video processing, camera, STB, TV, and NB-IoT. These chips don’t show up in ads for products, so they don’t grab headlines, despite selling in the millions. Industry insiders in China often talk about “Big HiSilicon” and “Little HiSilicon” to differentiate between the captive and non-captive sides of the business.

The 4G Balong chip it has begun selling to the open market is mainly targeting the IoT segment, not handset, and isn’t even a new chip—it has been around since 2014. A CAT-4 4G multi-mode chip can and will compete with Qualcomm in the IoT space—think bike sharing or point of sale devices. Balong is mainly made up of three other chips—baseband, RF, and power management—and doesn’t have the super powerful application processor it would need to be a full on handset chip. This is not the attack on the handset industry some suggest it is.

What if Kirin was available to all?

HiSilicon has given no indication that it plans on selling its more advanced Kirin 5G handset chips outside of Huawei. But what would happen it did take them to the open market? Who would buy them and what challenges would it face?

It is hard to see who would actually buy Kirin chips. Apple and Samsung certainly wouldn’t, having their own chip teams. I would expect other foreign firms, despite their struggles, to stay clear of Huawei technology as well.

That leaves local Chinese handset players such as Oppo, Vivo, and Xiaomi. The problem here is of course one of competition. Huawei handset sales, while dropping in the west, have grown significantly within China as the company has pivoted to some extent. This has eaten into local competitors’ sales, especially the likes of Xiaomi, which even saw co-founder Lei Jun step down as chairman.

While these local companies also have dreams of chip design themselves, they still use Qualcomm and Samsung chipsets for their high-end phones. Using a Kirin chip would mean losing any differentiation they had over Huawei and may even mean Huawei makes more profit from their phone sales than they do, not something I think they will want to do. There could also even be integration problems: HiSilicon is used to integrating with Huawei but integrating with other manufacturers may mean it has to design its chips differently in the future, adding to complexity and cost.

Conclusion

The mobile chip market is notoriously difficult to enter, and a number of companies like Renesas and Ericsson have given up over the years. The low and mid-end of the market already has Unisoc and MediaTek, and in the high end there is Qualcomm and Samsung. I do not see space for Huawei. At best, in some dystopian tech decoupling future domestic companies could be forced to use Kirin chips or other domestic companies like Unisoc’s or even ASR’s.

Perhaps it makes sense for domestic companies to use non-5G Kirin chips in cheaper phones. Samsung has done something similar before with Unisoc, but again, low end isn’t where the growth is now, so why bother going through the trouble?

As of now, fighting to get its chips in non-Huawei phones isn’t something HiSilicon plans to do, and I don’t think it would be beneficial to the company. It has bigger fish to fry.

Stewart Randall continues his series on China’s efforts to achieve independence from integrated circuits imports.

At the risk of sounding too negative, let’s discuss China’s semiconductor fab/foundry situation. Fabs are one of the big reasons it’s hard to imagine China getting completely independent from integrated circuit (IC) imports: there isn’t anywhere in China that can make cutting-edge chips. The semiconductor independence “Big Fund” has prioritized the area.

I promise I will write some positive articles on China’s semiconductor developments, but this is a hot topic now. Rumors—denied by TSMC—swirled last week about US pressure on fab leader TSMC to cut sales to Chinese companies. So let’s look at this key part of the IC industry: physically making the chips.

What’s a fab?

A fab or foundry is a factory where semiconductors are fabricated. In order to manufacture the most cutting-edge chips, fabs need capital equipment from all over the world, the most advanced of which are from Europe, US, and Japan.

What goes on inside fabs is akin to alchemy. In a sense, at least when discussing silicon chips, they are using equipment from companies such as ASML and LAM to turn sand into a chip. This is not an industry you get into on a whim. One ASML photolithography machine—a single part of the production line—costs about $100 million.

Integrated device manufacturers like Intel operate their own fabs, but today we’re looking at “pure play.” Pure play fabs do not design chips, although some work closely with design service partners. They generally make chips to order for “fabless” IC companies like HiSilicon or Qualcomm, receiving designs in the form of GDSII-type files.

How fab are China’s fabs?

The imaginatively named “Taiwan Semiconductor Manufacturing Corporation” (TSMC) dominates the global pure-play fab market with close to 50% market share. Samsung is a distant second at 18%. Third place Global Foundries from the US has only around 9% market share.

There are two main Chinese companies in this industry: the “Semiconductor Manufacturing Industry Corporation” (SMIC)—I think they were trying to beat TSMC in the imagination department—and Hua Hong. These two have 5% and 1.5% global market share respectively, with most of their sales within China. While these two fabs are broadly competitive in anything 28 nanometers (nm) and above, anything below is a struggle, and the smallest processes are where the high-end chips are.

Chinese fabs are behind, and it’s going to be very hard for them to catch up for a few reasons: access to the latest equipment, a lack of talent, and being late to the game.

Equipment

It takes a long time for leading equipment companies to manufacture the latest equipment. Companies like ASML and LAM Research naturally put their biggest customers first in the queue. TSMC and then Samsung will always be first in line.

Further, equipment makers can’t easily expand production capacity to get Chinese fabs faster access to equipment like extreme ultraviolet photolithography (EUVL). The skills needed to make them are few and far between.

Geopolitics also gets in the way of delivering the latest equipment on time—or worse. A memory maker called Fujian Jinhua was cut off from this equipment in late 2018 by a US export ban. With no alternative to US and US-linked European suppliers, the company had to cancel plans for a $6 billion plant, and it appears is still not producing any chips. As of today, the company’s website is up, but its products page (in Chinese) is empty.

Talent

The best talent naturally goes to the best companies. If you are a fresh graduate looking at the industry you will want to go to TSMC or Samsung. Otherwise, you may even choose a different industry.

It’s rather anecdotal but every SMIC engineer (outside of management) I have ever spoken with has complained of low wages and extreme working hours—think 996, and sometimes even worse. The night shifts aren’t fun either. Now the extreme working hours may be similar at the likes of TSMC, but the wage situation is not.

To fill the talent gap SMIC and others have been targeting Taiwanese, South Korean, and Japanese engineers with large salaries and other perks. A lot of Big Fund money has gone into such talent recruitment, which has helped win some recruits—but could also cause even more dissatisfaction among local engineers who aren’t getting paid as much as their foreign peers. To date, the extra talent has not helped Chinese fabs catch up. The other hurdles are just too great.

Timing

Chinese fabs were late to the game, have always been behind technically, and as mentioned above, stand at the back of the line when it comes to purchasing equipment. It doesn’t work in every sector, but here it really has been first mover advantage. SMIC came into the industry 13 years behind TSMC, and 19 years later it is still behind technically, and far behind commercially.

For example, no Chinese fab was able to manufacture FinFET designs until this year. Without going into details, FinFET is the current highest end process, which can fit more transistors into a certain area using a 3D structure. TSMC has been producing FinFET since the early 2010s. After several delays, some of which were due to US pressure on ASML, SMIC finally announced that a 14nm FinFET production line was up and running a couple months ago, just starting low volume runs, and claims to be ramping up to mass production as we speak. Meanwhile, TSMC and Samsung are mature at 7nm, will have 5nm next year, and have begun construction of a 3nm fab.

Prospects?

China’s Semiconductor Big Fund invested in pure play fabs and memory fabs with little noticeable result in these two areas thus far, at least commercially. But watch this space. Big Fund Mark II, which raised RMB 200 billion (about $28 billion) in July, will likely continue to invest heavily in these areas, but there is only so much throwing money at the problem can do. Industry king makers like ASML are part owned by TSMC and Intel, making it difficult for Chinese fabs to get ahead of the queue for advanced equipment, and given the sensitivity of the technology Chinese investment or acquisition seems unlikely.

The only way out of this predicament is for Big Fund Mark II actually to invest its money into creating Chinese capital equipment players. It seems like this is a focus, especially plasma etching equipment. I imagine mainland plasma etching equipment maker Naura, will be receiving some of this money to push past its current 14nm limit, but there is still no sign of a strong Chinese player in high-end photolithography. Without one Chinese fabs will still rely on foreign technology. We all saw how Fujian Jinhua faired after foreign equipment suppliers all left.

While other countries are similarly dependent, China’s IC supply is at risk due to geopolitical reasons, and even with all the money in the world it will not have its own replacement equipment any time soon.

I wrote last week that electronic design automation (EDA) tools are an Achilles’ heel in China’s bid for integrated circuit (IC) autonomy. While representing only a relatively modest $10 billion of the global IC market, these tools—dominated by a Big Three of US, or US-linked, firms—are critical to all IC design.

This EDA reliance poses a problem for not just Huawei and IC design subsidiary HiSilicon, not just for the other companies on the US entities list who have hopes to design their own chips, not just to all semiconductor companies in China, but to the Chinese government itself. Not to sound hyperbolic. All of China’s “self-made” chips are designed, verified, validated, etc. using foreign—mainly US—EDA tools. Chinese chips may grab the headlines, but the government knows China isn’t as self-reliant as they’d like it to be.

While-domestically focused companies can get away with pirated tools, this strategy doesn’t work if you want to be globally competitive. China needs to develop its own EDA tools. Local companies, and even Huawei, have been developing EDA tools, but we have not seen any notable achievements in public. Huawei’s developments are not very public, and the maturity and functionality of other domestic Chinese companies is still lacking.

China’s EDA lag

China has been investing in EDA R & D since the mid-80s when it developed the “Panda IC Design System”, however, I haven’t come across any company using this since I have been here and can only assume it wasn’t very successful. These days the more well-known companies, at least in China, include Huada Empyrean, Xpeedic, Semitronix, Platform-da, ProPlus, Microscapes, and Arcas-da.

But most of these companies cannot provide a complete design flow. The only example I know of is Huada Empyrean’s design flow for analog ICs, and in flat panel displays it works with some of the largest manufacturers including, Samsung, CSOT, HKC, and BOE. Other customers include Ricoh, SK Hynix, Marvell, and Sandisk.

Other than Huada’s relative success in its sweet spot the Chinese EDA industry in general has struggled, but why?

There are several reasons for the gap: Chinese tools are not comprehensive enough, there aren’t enough engineers with the skills to develop such software, market entry is difficult, and Chinese companies don’t have enough access to keep up with developments in manufacturing.

What holds Chinese EDA back?

Comprehensiveness: Chinese tools are simply not comprehensive enough, especially in digital design. Most of the digital design process is dominated by Synopsys and Cadence. Even if in one or two parts of the design flow Chinese companies have technically competitive products, it is difficult to break into the market as the Big Three have the ability to support customers’ development from spec to production. Chinese companies need to create a total solution to begin competing locally on any level, but even then, it will be difficult due to other factors.

Talent: Most of China’s EDA tool development engineers actually work for the Big Three: of the 1,500+ such engineers in China, only 300 (in Chinese) work for domestic companies. To put things further in perspective, Synopsys on its own globally has over 5,000 such engineers. Would-be EDA entrants also have to compete for talent with more lucrative industries. Application level software development at Alibaba, Tencent, etc. pays much better than a struggling Chinese EDA company.

Market Entry: With 95% of the domestic market, belonging to the Big Three, it is a highly difficult market to enter. Even if a full set of tools could be developed, in the short-term it will be difficult for any fourth company to gain any significant market share. Companies are used to certain design flows and engineers have used tools from the Big Three since university. These difficulties have made the domestic EDA industry a less attractive target for investors and, in turn, limited development.

Integration with Advanced Process Nodes: The link between design and process is a key part of an EDA flow. The Big Three work with the world’s leading wafer plants and foundries to develop a strong understanding of their processes, whereas domestic companies often only have access after a new process is developed and even then, not necessarily complete access. This makes it difficult for domestic companies to design and improve their software to compete with the Big Three.

Piracy: As mentioned above, EDA tool piracy is rife in China. These tools aren’t cheap. Silicon IP can’t be “cracked,” but tools can be. Any domestically focused company looking to save money will save it here. This also means the government may see EDA tool investment as a lower priority, as it can still have access to the tools for military chip design for example, even if bans are in place.

State-backed EDA

The government is beginning to support EDA tool development to some extent, and I expect support to increase over the coming years. Such companies can now claim back 30% of their development costs from the government, capped at RMB 30 million (about $4.3 million).

The government has also helped individual companies. For example, the Guowei Group has been granted RMB 400 million from the central and Shenzhen government for EDA development work. Also, Huada Emperyan has received hundreds of millions in funding over the past couple of years, not just from VCs but also from the state-run “Big Fund.”

While such government help is obviously welcome and is of some assistance it is nothing compared to the Big Three’s internal R & D investments, and if China really wants to become independent in this field much more needs to be done. The recently announced new Chinese government $29 billion semiconductor fund, or “Big Fund Mark Two” as I will call it, may go some way to help, but it remains to be seen how much of this will be invested into EDA. I suspect a small amount compared to how much is invested into memory, foundry capital equipment, and traditional fabless design.

Conclusion

China’s current predicament opens up opportunities for domestic companies. Government investment, coupled with a large domestic market, means they potentially have the environment to grow and improve. China needs to do this in a gradual way though, and not let such companies rely too much on government support. Switching everything to a Chinese equivalent (if one ever exists) could mean slower time to market and worse end products. Adopting a national procurement policy across the board is risky and could discourage innovation. Only once a domestic tool or entire design flow is on a more or less level playing field should they switch, and support should be based on certain milestones to avoid creating SOE-like inefficient operations.

I can see a future where domestic companies compete domestically within China for certain chips, e.g. analog designs or simpler IoT designs. Globally this will be more difficult though, and without access to the most advanced technologies from foreign wafer companies and foundries domestic EDA companies will always be at a disadvantage. China can reduce its dependency but at least for now, has no way of being completely independent in this space.

The first of two articles in a series on electronic design automation tools in China’s semiconductor sector. Read the second part here.

Can China achieve independence in integrated circuits (IC)? It’s a question I get asked a lot, and over the next few months I plan to shed light on where China is doing well and where it is not. It’s a complicated picture, but let’s start by saying this: it’s not going to be independent anytime soon.

One big reason for this is electronic design automation tools (EDA), a critical layer of IC design currently dominated by companies that are either US-owned or at least subject to US export controls. There are home-grown alternatives for a few specialized applications and more are on the horizon, but this category is an Achilles’ heel for efforts at chip independence, especially for Chinese companies with global ambitions.

What are EDA tools?

EDA tools are the software tools used to design ICs and printed circuit boards. Despite only accounting for around $10 billion of the $450-500 billion global chip industry they are essential to the design and creation of semiconductors.

Different EDA tools are required for different tasks and work in a design flow that all chip designers use to not only design but also analyze, verify, and debug semiconductor chips. This is not something that can be done manually, especially given the complexity of modern designs, which can contain tens of billions of transistors. Tools can be broken down into four or five main subcategories: design, simulation, verification, manufacturing prep, and functional safety. It is feasible to mix and match, but people tend to stick to one flow to keep things simpler, faster, and cheaper

Who are the main players?

Although there are many EDA tool companies out there, the industry is dominated by three main players—Synopsys, Cadence, and Mentor Graphics. These were all US companies until recently, when Mentor was acquired by Siemens. It is still based in the US and is a US company in every other way. Together they account for approximately 60 to 70 percent of the global EDA market, with Synopsys alone accounting for over one third.

Personally, I have not met a single fabless chip design company in China that has not said they use either Synopsys or Cadence design flows, or tools from both. There are no local alternatives, and while some may supplement parts of the Cadence or Synopsys flows with tools from other companies, these are usually for niche situations and do not play a main role in their design process.

The back up my anecdotal experience, showing that around 95 percent of EDA sales in China are divided amongst these three companies.

What does this mean for China?

Speaking to my clients, and Chinese in the industry, suggests Synopsys and Cadence are no longer able to work with Huawei, or any other company on the BIS entity list. This was further confirmed during the launch of the Huawei Ascend 910 AI chip when Huawei rotating chairman said Synopsys and Cadence could no longer work with Huawei.

This isn’t a disaster right away. Companies like HiSilicon will no longer receive support from their suppliers, but they already have access to the tools, and they know how to use them. I wouldn’t be surprised if they continue bringing out chips through 2020.

Suppose this becomes the new norm though, and that HiSilicon loses access indefinitely. This would mean competitors have access to support, receive all the latest patches, updates, and improvements. HiSilicon doesn’t. Perhaps current licenses run out and they cannot renew them.

What tools can they use then? How quickly can they re-train engineers who have relied on US tools since their university days to use a whole new set of tools? Designs will come out slower and fall behind competitors.

What about alternatives?

A purely domestic company may be able to secretly use unlicensed/pirated tools. Indeed, this is extremely common in China. Many pure domestic companies I speak with will use pirated tools somewhere in their design flow.

Listed companies, or international players like Huawei, wouldn’t be able to get away with this though. A team of lawyers would be waiting for them.

Perhaps they could use third party design services? This would get around the problem, but would mean outsourcing the design. It would also mean a significant number of employees were no longer needed. For HiSilicon, the jewel in China’s IC design crown, it would be impossible to admit it was no longer designing chips, at least not the whole design. To get around this problem, China will need its own EDA solutions. In the next installment, we’ll see why this is not as easy it sounds.

It was over three months ago when the US decided to put Huawei on the entity list. After some lobbying from US suppliers, this was delayed for three months. More Huawei entities were added to the list and this 3-month reprieve is supposed to only serve existing customers. Warranted or not, Huawei’s faces the prospect of losing access to key US suppliers.

Huawei is still highly reliant on US tech and I’m not talking only about Qualcomm: Huawei still needs US tech to design and test processors while their memory chips incorporate US tech.

So, the three months is up, and the US government has given Huawei another 3-month reprieve of sorts, but this story isn’t any closer to ending.

There are some big questions that need answering before we can really say how this will all turn out:

• If Huawei was finally cut off from US suppliers exactly how much trouble would the company be in?
• What are its — and the Chinese government’s — options?
• Could there be big opportunities for European and Asian tech firms to step in to fill the void left by banned US technologies?

Huawei’s problem

Whilst Huawei founder Ren Zhengfei has put on a brave face in public, and first-quarter profits and growth have been impressive, in reality, the company has a very difficult road ahead with no end in sight if this ban is enforced.

Although there’s nothing so specific the US can point to this time, Huawei — like ZTE before it — has been in the country’s crosshairs for some time: concerns of IP theft, security back doors, and connections with the Chinese Communist Party.

Not only will the ban cut off Huawei’s US suppliers, but it can also spread to non-US vendors whose technologies incorporate American-made components.

We’ve seen plenty of headlines about Google ceasing Android support for Huawei and the recent release of HarmonyOS shows Huawei is taking this seriously. For now though, the company will still use Android in handsets but has its OS ready as a backup and for use in other devices. This is important, but it’s not the whole story by any means.

If it were the only issue, perhaps HarmonyOS could fill the gap. Granted, the company would face a drop in handset sales outside China. People don’t like moving to another operating system, and they’d face an absence of Google services and a smaller ecosystem of apps. Microsoft tried this and failed. But it might be a barrier Huawei could possibly overcome, especially as Android apps are supposed to work on HarmonyOS.

Huawei’s Hongmeng may not replace Android on smartphones after all

Hardware issues

An equally critical issue for Huawei, however, centers on its hardware. Much has been made of the company’s ability to design its own chips. ZTE had been reliant on Qualcomm, so when it faced a US export ban, it didn’t have anything to fall back on, nor sufficient internal capabilities to design its own, no matter how hard its subsidiary Sanechips had tried.

Huawei, on the face of it, has its own chip design subsidiary in HiSilicon, and all high-end Huawei handsets use its Kirin application processors. Huawei’s main business though is telecoms equipment — cellular base stations, routers, switches and the like — and these use HiSilicon chips, too.

No Qualcomm, no problem – right? Wrong.

Firstly, to design and verify its chips, HiSilicon relies on Electronic Design Automation (EDA) tools from US companies like Synopsys and Cadence, as well as US-based Mentor Graphics (recently acquired by Siemens). There are no other companies in the world which can replace these tools. Empyrean, a Chinese company, may be able to handle a few of their tasks.

No chip design tools = no chips.

The second irreplaceable area is Field Programmable Gate Arrays (FPGAs), where the two big players are US firms Xilinx and Intel (Altera).

Their FPGAs are sometimes used in end devices if it’s deemed ASIC is not required, often when volumes are low, but they’re also used in the semiconductor design process for prototyping. Support for these and any future products would be lost, putting Huawei at a great competitive disadvantage.

Chinese companies such as GoWin and Unisoc have FPGA products and are trying to catch up, but they’re not there yet. I have yet to meet a Chinese company actually using their products in this way, although I see increased government investment in this specific area.

Huawei’s smartphone supply chain reveals reliance on US technology

Memory lapse

The third irreplaceable area is memory. Many of Huawei’s products use Micron memory right now. It could potentially switch from this US supplier to Korean vendors such as SK Hynix or Samsung. But, at the time of writing, I’m wondering if these companies could be affected by the ban because their products incorporate US technologies. Toshiba seems to have already been hit for the same reason. Not to mention the problems these companies face with the ongoing Korea-Japan trade war.

China has been building up its own NAND and DRAM capabilities, but the likes of YMTC, a leading Chinese memory design and manufacturer, are still some way off, and using the company’s memory would result in inferior products to Huawei’s competitors’. The same can be said for storage, solid-state and traditional, components in their data center and laptop products: there are no viable Chinese alternatives.

Processor worries

Added to all of this, Huawei will lose access to Intel processors, potentially killing its server and storage business, as the industry is dominated by Intel x86 architecture Xeon chips.

Without these chips, it can’t effectively compete in the server market – ditto the laptop sector. Huawei’s laptops receive universal praise on virtually any review site you look at. But without Intel CPUs, what does the Chinese firm have to fall back on? How well do non-Intel CPU laptops sell, and what if they aren’t even running Windows?

What may be the final nail in Huawei’s coffin is the way the ban is already spreading from US suppliers to other international firms.

Whilst I’ve mentioned Toshiba, the true shock for Huawei is that it could potentially lose access to Arm technology.

Most – if not all – HiSilicon’s chips are Arm-based, with its Kirin chip, server chips, camera chips, and router chips all based on Arm architecture. As an Arm architectural licensee, Huawei will, as far as I understand, be able to continue using the IP it’s already paid for it. But it loses access to support and further developments, which will deeply affect its competitiveness.

Huawei is already using its self-designed 64 core Arm-based Kunpeng server chip in some products, but its ecosystem is still small, and if it loses access to Arm, Huawei’s competitors would have access to the latest Armv9 architecture, which may come out in the next couple of years, while the telecommunications giant would be stuck with Armv8 and would likely have to build its own variant from that.

Decisions, decisions

Huawei and the Chinese government have some tough decisions to make.

Some may argue Huawei could innovate its way out of this situation. Perhaps it could develop RISC-V based designs. But that’s a long shot without the tools required. And, although RISC-V is maturing and there’s a growing ecosystem around it, it’s still not Arm or x86. Nor will we see a RISC-V high-performance computing (HPC) server replacing Intel in the medium term or replacing the application processor within a handset for that matter. Some RISC-V based processors can run Linux now, but I don’t believe any are running Android yet. The community is moving fast, though, and some have even predicted the downfall of Arm at the hands of RISC-V.

If the “ban” is actually enforced for a long period of time, then possibly — through intensive Chinese government support — Huawei could be kept alive while it works out how to replace all this US technology with other options. Indeed, whether the ban is lifted or not, it’s another wake-up call for China.

I suspect investments in the industry will increase further from the $118 billion already planned. But, in the meantime, the company’s short-term success depends on how much of this technology it has in stock and whether the ban is actually enforced or if there is some workaround, like using design services. If it is enforced and there are no workarounds the company may need to lay off thousands of people until it, China, or non-US companies come up with suitable replacement technologies.

That would be embarrassing for Beijing, given how it has talked up its tech superiority to its domestic audience in recent years.

Alternatively, China could strike a deal with the US. This would likely mean concessions on China’s side, which wouldn’t go down well domestically, given the country’s history of unfair treaties with western powers.

So, China’s in a tight spot: let Huawei potentially lose its global status, or give further concessions to the US — and, either way, manage the message as best as it can. In both scenarios, a loss of face is inevitable. It must hope the tech lobby in the US is more powerful than Washington’s desire to “win” the trade war or destroy Huawei. Given the bipartisan support for a harder line against China, waiting for a new US president isn’t going to help Huawei get the parts they need.

Baidu, Alibaba, and Tencent (BAT) are the top three investors in AI startups among large Chinese companies. Most of such funding goes to those focused on application layers or algorithms. While the tech giant trio has only made strategic investments so far, the field has also seen a stream of government and VC funding in recent times. Baidu and Alibaba have also moved into chip design themselves.

These new chip firms generally fit into three main categories—pure fabless semiconductor startups (which outsource manufacturing), algorithm startups that are moving into chip design, and internet firms that are exploring design too.

In total, I can think of at least 20 Chinese companies that fit into these categories, all of which are designing AI chips. I am sure there are a lot more of them out there in the vast industrial landscape. Traditional fabless design companies in China are also increasingly designing their own AI chips, making the market ever-more competitive.

It isn’t just China though where we see non-traditional chip companies like Baidu and Alibaba moving into design. In the US, countless listed giants, including Tesla, Google, Amazon, Microsoft, Facebook, and Apple, are designing their own AI chips, specific to their required applications. How can AI chip startups from both China and the US alike expect to live long when they face competition from behemoths that boast such technical and economic strength?

This vertical integration is not just a threat to startups but also to traditional chip designers that previously considered internet companies their customers. This dynamic will shape the industry in both regions for years to come.

These startups have moved quickly from burning VC cash to coming up with architectural innovations, and now they are tasked with actually finding customers in a very competitive market. Industry voices say it is becoming increasingly difficult for them to differentiate themselves and any start-up late to the party will struggle to bring anything unique, new, or different to the table.

Those that really do innovate and pursue emerging technologies like analog computing, in-memory computing, and neuromorphic computing warrant attention but have to deal with a much longer path to get a commercial product to market.

‘Chinese’ chips

From a hardware design perspective, these Chinese AI chips aren’t necessarily all that Chinese. As most of the companies work to tight deadlines, they tend to license an IP rather than develop in-house. The majority with which I have spoken will gain authorization to use networks that link design parts together, known as on-chip interconnect. They also rely on development and debug tools from companies like Lauterbach. That’s not to mention that the higher-end chips mostly use TSMC for fabrication. Of course, a large proportion of designs also use Arm cores. Some I have spoken with have developed their own GPU or custom neural network core, but the majority have neither the time nor luxury to do so. The bottom line is VCs want to see returns fast!

Returns come much faster in other industries. With the typical design process for a semiconductor taking 18 months and requiring tens of millions of dollars to finance just the first few designs, there is a real need to seek out customers from the get-go.

Some more experienced chip-focused VCs in the country understand this and are more patient, but we can expect to see some of these startups die off over the next couple of years as the pressure from VCs builds. The sector will consolidate for other reasons as well though. Industry mainstays are expected to snap up those new firms that look the most promising. The trend has been seen already when America’s Xilinx, the world’s leading designer, and supplier of programmable logic devices, acquired Beijing chip unicorn DeePhi last year. Other players will resort to acquisitions as a means to enter the market, as was seen when Alibaba took over C-Sky.

At this point, I think it’s safe to say the necessity to have dedicated AI hardware is here to stay. Most chips have some kind of ‘AI’ functionality, and this will only increase with time. Sure, some design teams will have to decide how much effort they dedicate to AI, but given that almost every major player is now involved, it is difficult to see such functions not becoming critical features of most future designs.

For China’s AI chip startups, it is a balancing act of getting to market first, being innovative, and having patient investors. The most successful will be acquired or secure more money, and the lame will die. The flipside for them is that AI and semiconductors are two areas of focus for the Chinese government in its push for technological independence. Those able to combine these verticals will garner a lot of attention. Despite this, the market is still approaching saturation, and only those that strike the right balance will survive.

RISC-V, the instruction-set architecture out of UC Berkeley, has been making waves in the semiconductor sector. Some even say that it could threaten industry heavyweight ARM in the long run.

The key difference between the pair’s respective products, which basically define the way in which software talks to a processor, is that RISC-V is open-source. It is this aspect that could be of particular interest to Chinese companies as such products are not directly subjected to US sanctions.

The RISC-V Foundation, which promotes the ISA’s use, features leading global players including Microchip, Western Digital, Google, Nvidia, and Qualcomm, to name just a few. Through collaborative and independent projects, several members are working to create RISC-V based designs.

China’s growing interest

Over the past few years, and especially in 2019, I have witnessed a huge increase in Chinese interest in RISC-V. Three years ago when mentioning the ISA, most engineers would look at me puzzled. Then, two years ago, they had at least heard of it though most would mispronounce it (it’s Risk-Five by the way).

Fast-forward to today and not only does every company I meet know of it, but the majority are actively researching it. Whether they have taped out an actual RISC-V based chip or are currently designing one, the interest is clearly there.

Today the foundation includes more than 25 Chinese companies, and what’s more, as of last year China now has two of its own RISC-V industry alliances with more than 185 members. Some of the most well-known Chinese members include Huawei, Sanechips from ZTE, Bitmain, Alibaba, and Xiaomi’s wearables partner Huami.

So, what’s all the fuss about? Why are so many large global companies jumping on the bandwagon, and what does RISC-V mean for China?

RISC-V provides an open-source ISA which users can build upon. As its a frozen ISA, software designed to run on one RISC-V processor will run on any other. It also provides a processor business model similar to that of Linux. Commercial vendors can build on the open-source ISA, or open-source cores to create their own IP to license and support.

It is important to note that whilst RISC-V is open-source, any serious product is probably going to want to license a commercial RISC-V core. Alternatively, companies with the resources and expertise can design their own. Often people may misunderstand RISC-V to be free. It isn’t but it is cheaper. Some commercial core suppliers do not ask for royalties, and license fees can be low, especially as these suppliers try to gain market share.

The main barrier to entry in the RISC processor world has been less technical and more ecosystem-related. Whilst ARM has a much more mature and sizeable ecosystem, that of RISC-V is growing fast and within these short few years, there are already products based on the ISA in the market and a supporting ecosystem.

While India has adopted more of a central government approach, China’s local authorities appear to actively compete in the semiconductor space, and we are now seeing RISC-V specific investments in the form of grants from the Shanghai and Nanjing governments among others. While what the Indian government is doing is great, China and its large number of semiconductor designers and household names are much better placed to take advantage of opportunities presented by RISC-V.

Sanction-proof

An interesting aspect of RISC-V to China is that it is not covered by the US Entity list as it is open-source. This means Chinese companies can use it without any fear of losing access in the future. Even SiFive, the first commercial RISC-V core company, and from the US, can still license to Chinese players. The Chinese unit is a completely separate entity that has allowed them to circumvent any export restrictions.

Even if somehow SiFive was prevented from licensing to certain Chinese firms, there are several non-US alternatives, even some domestic players. These include Andes, PTG from Alibaba, Syntacore, and Nucleisys. There are also free open-source cores available online.

We have seen that the entity list has the potential to limit ARM’s cooperation with Huawei’s HiSilicon despite it being a UK company owned by Softbank. With RISC-V, such risks are eliminated for Chinese firms. Additionally, it provides a globally recognized open-source standard for the country’s chip designers to latch onto. This removes any trust issues that would undoubtedly arise if China were to push its own closed ISA globally to compete with ARM or Intel.

This year I expect to see several Chinese companies taping out RISC-V based IoT chips as well as AI chips which include the ISA’s cores somewhere in the design. Whilst RISC-V has started with simpler IoT designs or low-power chips like Greenwaves’ GAP8 or the Chinese Kendryte, over the coming years I expect larger, more powerful versions to emerge.

Similar to how ARM boasts a range of cores covering low-power IoT to servers, RISC-V has the potential to do the same. I understand that many RISC-V cores struggle in terms of design size compared with ARM equivalents. However, there is no denying the benefits of this open-source, more customizable ISA that is also more cost-effective. The ecosystem is growing, results are improving, and the competitors are beginning to sweat, even if it is just a little.

The Wall Street Journal wrote last week that US semiconductor company AMD gave away the “keys to the kingdom” in a Chinese joint venture. The main theses of said article, which relied mostly on US Department of Defense sources, were that AMD had given away x86 architecture to China in 2015, and that this Chinese JV deal was key to the company’s revival.

It’s an exciting story of greed and recklessness—and a very hard one for an industry insider to believe. It looks like the Journal—and perhaps their US government informants—fell for the hype of a few companies hoping to pass off imports as a breakthrough in Chinese domestic production. What AMD did is far from unique in the field and doesn’t meaningfully reduce Chinese dependence on US integrated circuits.

The key accusation is that a 2015 JV between AMD and Chinese supercomputer maker Sugon gave China control of key x86 designs. In fact, the JV was nothing special.

Over the past couple of years, I have met with many Chinese chip design companies—one of which is a little-known Shanghai company called Zhaoxin. Like Intel and AMD, it has access to an x86 license. Its partner VIA Technologies, based in Taiwan, is the third company globally with an x86 IP license. VIA founded Zhaoxin as a JV with the Shanghai government all the way back in 2013.

VIA’s design team is now essentially incorporated inside Zhaoxin—a JV set up with the Shanghai government for the sole purpose of designing x86 processors two years before AMD even established its JVs. I do not know how much access Zhaoxin has to source code, but VIA will, and if having this kind of JV is giving away “the keys to the kingdom,” China had them long before its AMD deal. That it was missing from the article suggests strongly it was written without real industry understanding.

The truth is, it looks like the JV was mostly hype—its real function seems to have been letting Sugon tell the Chinese government it was buying domestic chips while continuing to use AMD products. The first thing to look at is the setup of the Haiguang Microelectronics (HMC) and Chengdu Haiguang Integrated Circuit Design Co., Ltd (Hygon) JVs, collectively known as Tianjin Haiguang Advanced Technology Investment Co., Ltd. (THATIC).

A brilliantly detailed description of the structure can be found here, but the key take away is that China wanted a chip it could call “homegrown” and AMD needed to make sure it had control of its IP at all times and nothing was given away. HMC was 51 percent AMD owned, and hence the US government had no objections. Hygon, however, was 30 percent AMD owned, and only designed top-level architecture, relying on HMC to provide the key elements associated with basic IP. This was enough to claim the chip was “Made in China,” even though no IP really changed hands.

With no IP changing hands, there is really no way China could design or manufacture these chips itself. Even with a stolen design, it would not have access to suitable foundries to fabricate the design. Local foundries might be able to produce something, but the eventual product would most likely be made on a larger process node and full of bugs because it was not fabricated on the Global Foundries process the AMD processor was designed for. So even with a stolen design China would end up with a product far behind competitors in the market and perhaps even unusable.

AMD would not even be the only US company with a similar setup. Intel works with Shanghai-based semiconductor design company Montage on a local x86 based processor and Qualcomm, until recently, had a JV with the Guizhou government called Huanxintong Semiconductor, which works on its own Centriq Arm based processors.

From a competition point of view, the x86 processor market is in a rather sorry state of affairs, and it wouldn’t be so bad if more companies, Chinese or not, could compete in the space. In the server and High-Performance Computing (HPC) markets Intel has well over 90 percent market share. Even in the laptop market, Intel has close to a 90 percent market share.

Intel is a gorilla in the room and the industry has been crying out for AMD or others to step up and compete. Using a 7nm process, AMD’s new Epyc processor “Rome” may be well placed to cut into Intel’s market share, but even then, Intel is only predicted to fall below 90 percent market share in 2020. Perhaps it is time a third player really steps up.

Until now Zhaoxin has only designed lower-end processors on a 28nm process, but its next design is said to be 16nm, with Intel not expected to move to 10nm until 2020. Having said that, Zhaoxin is still lagging behind, and it will likely continue to for some time to come. It seems to be happy maintaining a low profile focused on the Chinese government market.

I expect further government investment in companies like Zhaoxin as the JV route China has supported continues to be attacked. Having a strong third player in the x86 processor space will be a good thing, and in the short-term though we need AMD to continue its recent progress and bring some competition back into the market.

If the Journal’s sources represent the views of the Pentagon, it mostly likely reveals either the ineptitude or a lack of semiconductor knowledge on the part of the US government.

But, if somehow myself and others are wrong about the AMD JV, and it is a serious threat to national security, the US needs to improve regulations and enforcement for a lot of other technology JVs in China.

5G was the focus of the Mobile World Congress Shanghai this year, and I was amazed by how many 5G equipment companies I saw showing off networking equipment. The usual suspects, Huawei and ZTE were there, but they were joined by lesser known tier two and tier three companies—Baicells, Comba, Certusnet, Innogence, Skynetworks, H3C, Ruijie Networks, Raisecom, among others—who brought prototype 5G small cell and open radio access network (O-RAN) products, core cellular network components that will be sold in vast quantities over the next few years as 5G networks are deployed.

It’s easy to get the idea that China is racing ahead in 5G. Headlines in mainstream media often tout the country’s 5G prowess, saying that the US and the west are falling behind. But of course, the reality is much more nuanced: the first 5G base stations on the market may have Chinese brand names, but they’re multinational products that rely heavily on US components. This means Washington has leverage over China’s base station manufacturing—but also that US companies are earning a lot of the profit in the base station market even if Chinese companies get the glory.

We all know Huawei and ZTE, and perhaps to a lesser extent Datang. While Huawei designs its own chips for its various wireless product lines, it is by no means completely independent, as the recent US ban has shown. It is, however, more independent and capable than other Chinese companies, the majority of which would likely completely cease production of 5G equipment, at least in the short-to-medium term if they were to suddenly face similar bans. While Huawei and ZTE’s reliance on foreign EDA tools, semiconductor IP and software have been well documented less has been said for this whole other area within the 5G ecosystem.

At least 15 companies, by my count, displayed small cells. Small cells are a key component in any cellular wireless network. They help fill coverage gaps, especially indoors where macro cells are not suitable.

But what was obvious at the show is that the companies making these small cells have limited internal R&D capabilities. In general, at least for the time being, they do not have chip design capabilities. Their initial 5G products will all be based on FPGAs (field programmable gate arrays) from either Intel or Xilinx—in fact, not a single company was using anything but Intel or Xilinx. FPGAs are programmable chips, often used over hardwired application-specific circuits in base stations because they can be upgraded in the field with better algorithms, have lower entry costs, and enable companies to get to market faster. Companies may move to ASICs later once the volumes make sense, but in the beginning FPGA is usually the only choice.

There are domestic companies designing FPGA chips: Tsinghua Unigroup has two subsidiaries doing this, and there are others such as GoWin Semi, but they are still far away from providing a product which is viable for 5G applications.

The CEO of a small cell maker told me they were too reliant on US suppliers and was actively looking for Chinese alternatives. Their customer, perhaps the government, was demanding a fully domestic solution. He/she was optimistic, but other small cell providers laughed off the idea, saying they thought it would be at least five years before any domestic FPGA company could possibly have a competing product and that they were happy with using Intel or Xilinx.

This is great news for Intel and Xilinx. As we know, 5G frequencies are much higher than 4G so base station density will be much higher. More base stations means small cell revenues will be even larger for 5G than they were for 4G. Since the companies making them rely on US technology, the idea that the US is not involved in 5G isn’t exactly true—and US companies are in a great position to profit from the success of Chinese 5G equipment companies as they roll out their equipment in China and around the world.

On the flip side, small cell components are another angle of attack for the current US administration. While the last Huawei ban seems to have been lifted during the G20, if the US wants to slow down or even destroy China’s 5G plans, then export bans on these products to Chinese companies could be a quick way to do it. Most of the small cell  companies I saw at MWC consider themselves too small to be on the US government’s radar, but China and 5G are not, and who knows what the current President may do.

Of course, banning IC exports isn’t simple. I would expect strong push back from US suppliers given the potential revenues involved, and such an action would set back global 5G development significantly. It would also speed up Chinese development in this specific area even further.

Since the ZTE debacle last year, followed by the recent Huawei ban, the race to become semiconductor independent has sped up, and I have recently seen this manifest itself in the FPGA chip sector too. I know of at least a couple of domestic FPGA companies which are now investing further into creating 5G-capable products. This is on top of well-established players like HiSilicon and Unisoc, which already have 5G-capable ASICs in the market for handsets and base stations.

The US may not have a large macro cell brand like Huawei, ZTE, Ericsson, or Nokia but it does have companies integral to the 5G ecosystem and these companies are well placed to make huge profits from global 5G roll outs. China is aware of its reliance on these products though and recent US policy has hastened the development of Chinese equivalents. The likelihood of US FPGA chips being replaced over the next five years with Chinese equivalents is low, and even 5G ASICs from HiSilicon and others rely to some extent on foreign architectures, tools, and IP. But Pandora’s box has been opened, and sooner or later China will catch up in many of these areas, even if the end of the Huawei ban reduces the immediate pressure.