The Narrow Scope for Cooperation among Techno-Democracies in Semiconductor Export Controls

There’s been a fair amount of anticipatory commentary predicting that the US will rally its allies to gang-up on China in chips. As nice as that sounds (to some), we should recognize that the willingness of American allies to go along with this endeavor is directly proportional to their own economic interests. The wider the scope of such endeavors the more they impinge on allies’ economic interests so expect allies to embrace only a narrow scope for such controls.

In light of the last four years of Trump, there’s a tendency to valorize any multilateral approach as being effective.  History suggests otherwise. The export control regime of the Wassenaar Arrangement since its inception in 1996 has not been binding.  Moreover, the United States has rather consistently taken a stricter stance on the controls than its allies.  To give a China-related example from the early years of Wassenaar, many American companies complained that they lost out in taking part in China’s 909 Project, the government-sponsored joint-venture fab–Huahong NEC, in the late 1990s because the Chinese perceived the Japanese government as taking a much more relaxed approach to export controls than the US.

Fast forward to today, EUV lithography equipment is controlled under the Wassenaar Arrangement, but that didn’t stop the Netherlands’ ASML from agreeing to sell such equipment to China’s SMIC until US pressure forced the Dutch government to put those exports on hold.  And notice, the Dutch hasn’t approved nor rejected the export license for the equipment, hardly a definitive stance on blocking all EUV sales to China.  Of course, this lack of a definitive answer on the license may just be a ploy to avoid angering China.

In any event, ASML sales of EUV lithography equipment are in rude health.  What can’t be sold to China’s SMIC can be sold elsewhere given ASML’s large backlog of orders, which is approximately equivalent to one year’s worth of production.  With Taiwan and Korea still doing what they do (i.e., investing in advanced fabrication) and the strong possibility that the US and EU will follow suit, not selling some EUV machines to SMIC really does not hurt ASML’s EUV sales any time soon. 

Turning to Japan, one must acknowledge the masterful spin behind the “breaking news” that Canon is going along with banning EUV lithography sales to China.  This sacrifice in the name of an alliance of techno-democracies sounds impressive. Industrial realties suggests otherwise. Given that Canon does not currently have any EUV equipment to offer and won’t have any for the foreseeable future, this sacrifice of sales of imaginary equipment is hardly a high price to pay.

Similarly, it is perfectly reasonable and sensible for the US to discuss resolving supply chain bottlenecks with American allies as highlighted in President Biden’s February 24 executive order.  Indeed, talking to “productivist” East Asian allies about the need to have supply chains in their own countries rather than in China is music to their ears, especially Taiwan.

None of these moves indicate a strong willingness on the part of America’s allies to bear the burden of much stricter controls on China.  Happily, thus far it does not look like the Biden administration is moving to enact wide and broad controls that affect a wide swathe of Chinese firms beyond Huawei and, to a lesser extent (sales above 10 nm are still allowed), SMIC so the strength of this alliance of the techno-democracies will probably not be pushed to the breaking point. Unhappily, rumors still abound about taking a hardline on the proposed Foundational Technologies policy regarding semiconductors.  Reports along these lines seem to be jumping the gun since the policies as of now are under-specified and the personnel to enact them are not in place. As unlikely as such policies are to occur, their cost to American industry if implemented could be grave. Moreover, the benefits in terms of the ability to keep a wide array of chip technology out of China over time is questionable in large part due to the likely defection of allies. After all, any such moves to weaponize the silicon supply chain against China Inc. as a whole would likely meet resistance from allies along the lines of the resistance to America’s stricter enforcement of Wassenaar Arrangement that goes all the way back to the beginnings of those multilateral export controls.

Brief Note on a Biden Appointment and US Semiconductor Controls

According to a well-placed source, the Biden administration has decided to appoint someone very close to the American semiconductor industry as the head of the Department of Commerce’s Bureau of Industry and Security (BIS). This position is critical because BIS is in charge of the Entity List. As I have interacted with this future head of BIS, I am as confident as the semiconductor industry source that gave me this information that this appointment portends radical scaling back of the semiconductor controls aimed at Huawei and other companies. With such an appointment, the 20 percent jump in the stock prices of the big American semiconductor capital equipment firms last week has proven to be more than irrational exuberance.

Sprouts of Silicon Supply Chain Grandeur? China’s Chipmaking Equipment Industry

For the last three decades, China has very actively encouraged fabrication production through a variety of industrial policies, from the 908 Project in 1991 through the second installment of the Big Fund in 2019. For the last two decades, these policies have succeeded in increasing China’s share of global fabrication capacity from less than 1 percent of total fabrication in 2000 to 12 percent in 2018, even if much of that capacity remains foreign-owned. In 2019, China’s semiconductor capital equipment spending reached 18 billion USD, 20 percent of total global capital equipment expenditure.

Despite China’s burgeoning market for chipmaking equipment, the concentration of resources has been primarily on fabrication and, secondarily, design, with the capital equipment sector given relatively little policy support. This relative neglect of the capital equipment sector may come as a surprise to some of those among of us who have been subjected to countless pitches, lectures and even 江泽民-style serenades (courtesy of ambitious officials and interested third parties) about the importance of developing the whole silicon chanyelian/supply chain in China. The policy sequencing with an initial relative de-emphasis on capital equipment actually makes eminent sense because a new industry often tries to serve its domestic market first before selling overseas. Even established multinational firms often suffer from what international business scholars call the liability of foreignness. Thus, China has chosen to foster first domestic fabrication capacity, which in turn can serve as the initial market for China’s emerging domestic capital equipment vendors.

There were already signs that the relative neglect of the capital equipment industry by industrial policymakers was changing before the last five-plus months of tightening of American export controls. Under the first installment of the Big Fund, semiconductor material and capital equipment firms received only 4.2 percent of the total installment. Furthermore, these funds were spread among fourteen firms: seven materials firms and seven equipment makers. However, under the second tranche of the Big Fund kicking off in March of this year, the government had already aimed to channel more funds to materials and capital equipment firms and target the development of what was deemed the most “core” equipment, CMP and lithography. The government also already intended to push harder for firms to verify and purchase more local equipment.  Undoubtedly, the export controls targeting SMIC on top of those targeting Huawei will now spur the Chinese government to pour money into chipmaking equipment.  In this post, we will assess the state of the Chinese industry as it is today rather than speculate about the future impact of the avalanche of funds that is sure to come chipmaking equipment’s way.

Currently, Chinese producers are producing “noncritical” equipment for fabs. In other words, they are not competing with the likes of Applied Materials and LAM. They have been able to sell more equipment outside of IC fabrication for solar panel, IC A&T, and flat-panel display manufacturing. Estimates place domestic producers as accounting for only five to ten percent of the total semiconductor equipment expenditure in China in 2018.

Most of the domestic producers only produce one or two types of semiconductor equipment (see Table 1 from which I’ve excluded many small, one-product firms). Only NAURA and AMEC make a substantial range of equipment. The three firms that American competitors and equity analysts alike have identified as most promising are not surprisingly NAURA, AMEC, and the Chinese–American hybrid firm (its headquarters is still in the US) ACM.

Table 1 Major Chinese Capital Equipment Vendors and Their Products     

Type of EquipmentNAURAAMECAMCSMEE
Oxidation/Diffusion Furnaceyes   
Lithography   yes
Coater    
Silicon Etcheryesyes  
Metal Etcheryesyes  
Plasma Etcheryesyes  
Ion Implanter    
PVDyes   
CVDyes   
MOCVD yes  
ALDyes   
CMP    
Cleaning Equipmentyes yes 
Test Equipment    

NAURA emerged via the mergers of a number of state-owned semiconductor capital equipment firms as well as the acquisition of an American wafer cleaning equipment firm, Akrion. Its controlling shareholder is the Beijing municipal government. In 2018, the firm reported less than 300 million USD from sales of semiconductor equipment. In comparison, Applied Materials had over 17 billion USD in revenue in 2018. In other words, China’s largest capital equipment maker is still a minnow in this sector.

AMEC was founded by returnees who left Applied Materials and other firms. Now this firm’s dominant shareholder is the Shanghai municipal government. The firm’s main business is selling deposition equipment for the production of LEDs, not IC fabrication. In LED equipment, it is the third largest supplier given its price-competitive products in China, which is the largest producer of LEDs. However, the market is only 500 million USD in sales.[1] In the etching market, AMEC has 1 percent market share or 80 million USD in revenue as of 2018.

ACM was founded in the United States by David Wang in 1998. In 2006, he returned to China to set up a subsidiary, ACM Research. The firm focuses on cleaning tools and wafer-level packaging for A&T. Wang was the largest shareholder until new investments reduced his holdings from 25 percent to a little over 2 percent. The dominant shareholders are investment vehicles of the Shanghai government. Ostensibly, this move was done in preparation for listing the firm on Shanghai’s answer to NASDAQ, the STAR board, but it is a sign of the state encroaching on all the promising private firms in this strategic industry. ACM has 2 percent of the global cleaning market to NAURA’s 1 percent, but this firm generates the smallest sales revenues of these three promising Chinese firms at just under 100 million USD in 2018.

 Last and very much least, the spin about SMEE is that this firm is the next ASML.  There’s next to no evidence to back this hype.  I would have ignored this firm, but the hype has been so relentless a few words are needed.  What lithography equipment SMEE offers is very much of the lab rather than for the fab.  The throughput of its equipment is only one tenth of the machines sold by ASML, Canon and Nikon. On top of that, SMEE’s lab also serves as its only production facility.  Consequently, SMEE’s lithography equipment is far from mass production-ready both in terms of its own output and fab customers. The final strike against SMEE’s lithography equipment is that it did not offer argon fluoride (ArF) laser technology2 so could not be used for high-k metal gate (HKMG) tech typical of reasonably advanced process nodes, such as 28 nm, until recently.


[1] The two largest producers, Germany’s Aixtron and America’s Veeco, have respectively left and been hammered in the LED equipment business. The margins in this China-centered business are simply too low. Even AMEC, the homegrown champion, had to allow its margins to suffer to take market share. 

[2] Correction: The original text stated that SMEE still did not have the ArF laser. Until recently, SMEE did not have an ArF laser. They have now obtained one from another spin-off of the Institute of Microelectronics at Chinese Academy of Sciences. Thanks to one reader for pointing out SMEE’s ArF and follow-up with a source. Does not solve the problem of SMEE’s equipment being for the lab instead of fab though.

Undermining American Industrial Excellence: Controls on Wafer Fab Equipment and China

This post will examine the implications of Trump’s actions against Huawei and now SMIC for America’s wafer fab equipment industry.  With the August 17 Final Rule on Huawei, American technology embodied in capital equipment cannot be sold to foundries serving Huawei (assuming no licenses are forthcoming from the Department of Commerce).  With American dominance across a large swathe of critical capital equipment, one might think that this move would hurt Huawei with little cost to American industry. Unfortunately, actual planning already underway by American industry and other plausible scenarios tell a more pessimistic story pointing to the potentially high costs of this policy for America.

American Dominance in Chipmaking Equipment?

In wafer fab equipment, American firms have a very strong position. Three of the five largest firms, comprising 71 percent of 2018 revenue, are American: Applied Materials, Lam Research, and KLA. Applied Materials is the largest firm with 18 percent of global revenue. Nevertheless, going forward, ASML of the Netherlands is likely to be the largest firm because it monopolizes high-end lithography, which is called extreme ultraviolet (EUV) lithography, and dominates lithography generally. America withdrew from the lithography market in 2001 when ASML acquired Silicon Valley Group (SVG), the last American lithography equipment producer.

America’s predominance in this industry, even excluding lithography, has been neither continuous nor dependent on America’s own share of fabrication capacity. In the 1980s, Japanese firms looked poised to wipe out the American capital equipment makers along with much of the rest of the American industry. Between 1983 and 1990, America’s share of global IC capital equipment revenue declined from 66 percent to 44 as Japanese suppliers pulled ahead. What worried so many American semiconductor firms at the time was the perception that the big Japanese equipment producers prioritized their Japanese customers leaving American chips producers at a disadvantage.  What worried American semiconductor equipment makers was the perception that Japanese chipmakers favored their homegrown capital equipment providers.  Many inside the American chip industry thought it was their end days. Sematech was in no small part what rescued the American industry, including American lithography production, from oblivion by organizing vertical and horizontal cooperation among the American industrial participants and conducting a number of successful projects to improve product and process technologies (see Browning and Shetler’s history of Sematech). In the wake of this successful public–private partnership to revive American industry, the one major error was, arguably, the approval of the sale of the last American photolithography maker, SVG, to ASML in 2000. At the time, SVG was ahead of ASML in EUV lithography research.

Today, the most prominent American firms have large market shares across a number of categories of wafer fab equipment (see Table 1). American firms as of 2018 monopolized production of four product areas: optical mask-making lithography (not IC lithography and not included in Table 1), bevel edge removal, gate stack tools, and ultra-high-dose doping equipment. In other areas, such as etch, metrology, and inspection, American firms maintain a monopolistic position in certain high-end products. This sounds like a very dominant position, but interviewees have stressed to me that they believe that these monopolies are not secure ones. In their view, Japanese firms and others can make every one of these pieces of equipment if given some time, and one type of equipment, bevel edge removal tools, is an optional tool in fabrication rather than a necessary one.

Table 1 2018 American Share in Wafer Fab Equipment Markets (%)

 Applied MaterialsLAMKLAOther US FirmsTotal US Share
Plasma Chemical Vapor Deposition (CVD)52.135.5  88
Sputtering74.4   74
Electrochemical Deposition16.176.5  93
Bevel Edge Removal 100  100
Conductor Etch32.753.2  86
Chemical-Mechanical Planarization (CMP) and post-CMP cleaning70.3   70
Gate Stack Tools100   100
Doping Equipment68.2  17.886
Ultra-High-Dose Doping Equipment100   100
Process Control11.5 51.17.870
Thin-Film Metrology  40.630.771
Optical Metrology  46.430.377
Macro Defect Inspection  68.516.585
Unpatterned Wafer Inspection  96.3 96
Patterned Wafer Inspection15.8 68.2 84
Optical Patterned Wafer Inspection6 87.5 94

To give an example of how latent or trailing competitors can come back, we can look at the experience of American companies. In 2010, Applied Materials lagged far behind in conductor etch products. By devoting R&D resources and working closely with key customers, Applied Materials was, within five years, again able to become an industry leader in selling this type of equipment for advanced process nodes. Huawei does not have five years to wait if cut off from American equipment-laden fabs, but industry insiders think the Japanese firms only need two years to fill the gaps left by eschewing American equipment. And this assumes market-based competition. If concerned countries are willing to throw money at inducing this transition, the catch-up of foreign capital equipment could be even faster.

What Tokyo Electron did in the wake of the United States placing Fujian Jinhua on the Entity List illustrates the capabilities of competitors of American firms. As American capital equipment makers pulled out of Fujian Jinhua, Tokyo Electron made a big show of staying put in Fujian Jinhua’s fab, meaningless though this show of support was given that there were few other tools available to see fabrication through to the end. Building on this publicity stunt, Tokyo Electron went around to other fabs in China saying that American equipment could not be trusted because it carried political risk. For Tokyo Electron’s American competitors, the problem with this sales pitch is that it is the best kind—the truth. Consequently, American equipment vendors have claimed that they lost sales to Japanese vendors in the wake of Fujian Jinhua. To be clear, these sales were in products where Tokyo Electron already had competing products, but it is a large and capable firm with many such products.

Corporate Strategies to De-Americanize Fabs: Two Timelines

There has been a private industry study, which must remain anonymous, involving active industry players that has emphasized a somewhat slower replacement rate than what interviewees reported. By this estimate, to ready equipment for a cutting-edge de-Americanized fab would take four to six years. The major hurdles for replacing American equipment are high-end inspection, process control, and etching equipment, so this study suggested that rather than replacing American firms, the faster approach would be to replace American content. All the major American vendors have some production overseas, principally in Southeast Asia, and the idea would be to reorganize production of these American MNCs to remove American content. Such a move would result in the capital equipment vendors remaining compliant with the Entity List while still being able to provide equipment legally to foundries serving Entity List–designated firms. Executives from KLA, the leading supplier of metrology, inspection, and process control tools, admitted to considering using the firm’s manufacturing sites outside of the US to maintain business as usual.[1]

This scenario assumed that the foundry for this de-Americanized line would be TSMC. This solves two problems.  First, TSMC is unlikely to be cut off from ASML’s EUV equipment. Second, TSMC is at the cutting edge of fabrication so it is the best foundry partner with which to set up an advanced line.

Having been briefed on this private group’s report but not privy to the detailed contents of the report, I would like to add a note of caution before accepting the report’s pessimistic conclusions forecasting limited participation by non-American capital equipment vendors in this de-Americanization process. The report seems to approach de-Americanization with the historically reasonable assumption that firms are still generally on a commercial footing even with government subsidies playing a role as they always do in fabrication. With Sino-American technological rivalry heating up, this assumption may not hold any longer. When pressed into a corner, China might provide such large subsidies that less inefficient and thus more costly alternative equipment might become a viable alternative. The countries providing such equipment might also try to seize the opportunity by providing their own subsidies as well. 

How long would it take TSMC or another fab, such as Samsung’s foundry services, to create a fab line that designed out American equipment? Assuming the foundry would not build a whole new fab building on a greenfield site,[2] the amount of time needed to move in new equipment and get the new fab line up and running is one year to fifteen months.[3] Of course, for any equipment for which an immediate off-the-shelf replacement from a non-American company is not yet available, the time to develop the equipment would have to be added to the fifteen months needed to move in and ramp up production. Thus, the minimum amount of time would be more than three years if the estimates of interviewees concerning their non-American rivals’ capabilities are accurate. 

This timeline means that there is unlikely to be any de-Americanized fab savior in time for Huawei because Huawei is very unlikely to have existing chip inventories that can last so long.[4]  However, the expanding scope of controls give firms incentives to pursue this whether or not Huawei is the end customer.

Notice that I have not brought up the possibility of Chinese capital equipment saving the day.  I’ll write a post about that soon, but the short answer is Chinese capital equipment firms will not play any but minor roles in de-Americanization of wafer fabs in the next five years.  In order to de-Americanize fabs, one of three things has to happen: American firms will have to de-Americanize their supply chains, Japanese and other vendors will have to plug the gaps or a mix of these non-American vendors and American offshoring will have to occur.

Conclusion

The American government’s moves against Huawei and SMIC provide perverse incentives for American firms to accelerate the offshoring of research and production of what is the final bastion of American-made machine tool excellence. On top of that, these moves open up opportunities for American competitors to supplant American semiconductor capital equipment firms even in areas of their greatest competitive strength.  Flipping MAGA on its head, Trump has stumbled upon the precise policy to undermine one of the remaining areas of American industrial “greatness”.


[1] “KLA Corporation (KLAC) CEO Rick Wallace on Q1 2020 Results – Earnings Call Transcript.”

[2] There are a lot of fab shells (fab facilities without any equipment) sitting empty in East Asia, so it is a fair assumption that the foundry undertaking this project would not have to build a greenfield fab.

[3] This estimate assumes six months for equipment move-in and six to nine months for testing and pilot production (or approximately twelve months to go from equipment installation to ten thousand wafers per month), which is moving at the rapid foundry-in-Asia speed. Former and current fab managers and/or top executives from Asian foundries provided these estimates. TSMC’s Nanjing fab was able to move in equipment in under six months. 

[4] For some chips, Huawei may have built chip inventories that can last eighteen months or more.

Huawei, EDA Vendors and the Entity List: How the Chokepoint Strategy Could Backfire

            In the last post, I covered the rise of the EDA tool industry and how an American oligopoly has dominated the industry from early on.  In this post, I want to examine what the effects of the Entity List controls placed on Huawei will have on both Huawei and American EDA tool firms. I argue that the dominance of the Big Three EDA tool vendors (Synopsys, Cadence and Mentor Graphics­–yes, I know Mentor Graphics is now owned by German Siemens but the bulk of the chip EDA tools’ tech and activity resides in the US) has never really been tested and may prove fragile when other players in the chip industry, principally foundries, are incentivized to work closely with other EDA tools beyond the Big Three. Before going into the details of that argument, I first want to put the damper on any notions that Chinese EDA tool vendors will be the ones to break up the current cozy oligopoly.

EDA in China

            Going back to the Seventh Five-Year Plan (1986-1990), China has been trying to foster its EDA industry without much success. The IC Mega-Project started in 2014 and directed some funds toward EDA development but progress has been slow. Huada Empyrean has indeed enjoyed some success: it offers sets of tools covering the complete design flow for analog chips as well as EDA tools for LCD driver chips. No other Chinese EDA companies can offer tools covering complete design flows. Industry interviewees expressed their skepticism concerning the ability of China’s EDA firms, including Huada Empyrean, to compete with the established giants. The market data agrees with them. The total market for local EDA tools in 2019 was only 77 million USD, representing a mere 10 percent of total EDA sales in China.

China’s EDA development is not limited by just the difficulty of matching the product scope the Big Three EDA providers have created over the past few decades. The Big Three’s close links with leading foundries provide them an inside track on keeping up with changes on the manufacturing side and thus allow them to keep their software up to date with the latest process technology ahead of would-be rivals. New Chinese competitors have access only once the new process is developed, and their access will not be as wide as it is for the Big Three. 

To make matters worse for would-be Chinese challengers, the Chinese marketplace is shifting away from their strengths. China’s fabless design industry is consolidating into large firms, resulting in more of the EDA tool market being in the hands of firms that can afford the relatively expensive offerings of the Big Three. Furthermore, the design revenues are shifting to digital design areas and away from the relative strength Chinese EDA firms have in analog design.

Piracy is still an issue for these smaller vendors that are reliant on their home market. Part of this is due to the incomplete protections for IP in China, but the larger issue is that some of these smaller vendors allow their products to be downloaded rather than accessed by license keys. This type of access provides easy opportunities for the software to be copied illegally. Presumably, because these smaller firms are desperate for sales, they are more willing to provide more convenient access to their software than the Big Three, which employ the more-protective license key model. 

Additionally, the best software application engineers in China do not want to work in obscure EDA when they can make much more money working for Tencent, Alibaba, and other internet firms. Perhaps this will change with the enormous amount of state investment set to be directed towards the IC industry generally. However, as it stands now, only three hundred or so EDA tool development engineers are employed by local Chinese EDA vendors compared with more than five thousand such engineers worldwide at Synopsys. A potential advantage for China’s EDA industry is the fact that foreign EDA tool vendors employ at least 1,500 engineers in China. Offsetting that potential advantage is the fact that the global EDA engineering workforce is approximately forty-five thousand, so China’s total EDA workforce still represents a relatively small portion of the global total. In the short term, being cut off from American-origin EDA tools would be a highly problematic for Huawei. Even in the medium term of years rather than months, it is unlikely that local firms could fill the gap.

Huawei and the EDA Vendors’ Strategic Responses to the Entity List

            If the controls as of the August 17 Final Rule are fully implemented and no licenses are granted for selling EDA tools to Huawei, Huawei will not have legal access to the Big Three’s EDA tools. There are no Chinese or other foreign vendors that can fill the gap sufficiently to allow Huawei’s HiSilicon to continue to design chips legally. Because the vast majority of the Big Three’s EDA tools today are accessed via license keys that provide online access to the EDA software, the preferred method for gaining illegal access to the Big Three’s EDA tools is to hack the license keys.[1] With the right technical support, hacking these licenses is feasible as attested to by interview subjects and past cases, such as InnoGrit’s hacking of Synopsys’s license keys (link). If forced into a corner with no choice, Huawei could choose to hack license keys for Big Three’s tools.

Would the EDA vendors try to seek legal remedies? It would be difficult to do so in China given the ill will that full implementation of the export controls would generate there. From the EDA companies’ perspective, Huawei’s continued use of their tools without paying is preferable to Huawei trying to develop alternative EDA tool vendors on the off chance that these alternative EDA vendors emerge as peer competitors (as unlikely as that would be). Turning a blind eye to the hacking in hopes of a loosening of the controls in the future might very well be the smartest move for the Big Three vendors.

I believe, however, that Huawei probably would not have to choose the blunt instrument of hacking license keys. The final rule issued on August 17 offers an either–or clause, where either knowledge on the part of the provider of the good or service (e.g. EDA software) to Huawei is required, or Huawei and its affiliates “touching” the product somewhere along the supply chain[2] is sufficient to make the product controlled. However, to be legally liable, a firm still has to have knowledge that it supplied Huawei or dealt with a Huawei-touched supply chain as explained to me by a DC lawyer deeply versed in the Entity List. These requirements have already encouraged creative circumvention. On July 13, a well-placed representative for small foreign EDA tool vendor had already told me that at least one Entity List company had set up a shell company with no apparent links to the Entity List–designated firm to serve as a legal front for EDA licenses. 

Other tactics may emerge if the major EDA vendors do not passively accept these onerous regulations. Some have speculated that Synopsys’s new joint venture with AMEDAC (全芯智造), formed in September 2019, is designed as a vehicle to add plausible deniability to any charges of dealing with Huawei and its Entity List–designated affiliates. Speculating as to how this exchange might work, AMEDAC could provide tools to a Huawei front company. By adding two intermediaries between Synopsys and Huawei, Synopsys could claim to not know what AMEDAC was doing, and AMEDAC could deny any knowledge of supplying Huawei. There are rumors that another of the Big Three has set up a partnership that looks suspiciously like a vehicle for circumvention. With the broader controls incorporated into the August 17 final rule, these joint venture gambits might not provide enough legal cover to be worth the risk, however.

The other big issue is that even if Huawei could somehow get its hands on EDA tools via shell companies, could Huawei actually find a foundry willing to fab its chips. It seems unlikely that Samsung or TSMC would take on a mysterious client company with suspiciously sophisticated tech as long as those foundries are dependent on American capital equipment (an issue that will be covered in the next post).  Such a high-risk-low-reward tradeoff seems unappealing.  After all, Huawei being displaced from chip design doesn’t mean the market for chips will shrink so TSMC is probably telling the truth in stating that the end of Huawei orders won’t have too much impact.  Foundries in China arguably would be more willing to take on the risk of fabbing for Huawei shell companies, but they can’t offer the cutting-edge processes that TSMC and Samsung have so they are only a partial solution.

Ultimately, Huawei can probably survive without designing its own chips. In this scenario, it would revert to its 2010 business model of building telecom equipment mainly with chips from outside vendors. To the extent that some of Huawei’s current competitive advantages are silicon-based, Huawei would lose these advantages. The other major downside is that it could not work directly with chip vendors as under this scenario Huawei would have to get chips through gray channels that the US Department of Commerce would find difficult to track.

Dangers for American EDA Dominance

            For an anti-Huawei hawk, the US government’s plan to use EDA tools as a chokepoint to strangle Huawei looks intuitively appealing given American dominance in EDA tools and the critical position of these tools in the chip value chain. However, there are great dangers in trying to weaponize this part of the chip value chain. First, we do not know how quickly one of the big EDA vendors could be displaced by new entrants because the United States has not put other countries with economic clout in the position of being forced to seek alternatives. This policy would encourage others, not just China, to consider alternatives. More critically, this policy could force foundries to question priority of positioning (i.e. which EDA vendors have priority when working on new process nodes). The Big Three EDA vendors derive a lot of their expertise at the design–foundry interface from their favored position as the firms working with the foundries early on for each new node (see my September 24 post). If the foundries face enough uncertainty, they may include a new close partner or two to hedge their bets, leading over time to eroding positional power on the part of American EDA tool vendors. If the Entity List controls are limited to Huawei and the other much smaller firms, then the likelihood is lower that the foundries will erode the Big Three’s positional advantage through hedging.  Any further expansion of the Entity List targets would just further encourage foundries to include less politically problematic EDA tool vendors as one of the favored partners working on the design-foundry interface for the newest process node. And of course, a de-Americanized fab (again, to be addressed in the next post) would beg for a non-American EDA partner to be brought into the mix.


[1] EDA tool vendors are often plagued by illegal overuse of EDA tools, e.g., a contract that specifies access for one user for one server, but the tool vendor discovers that the server is utilizing the tool 24-7. However, in the context of being completely cut off from legal access to the EDA tools, this form of IP theft is irrelevant.

[2] Huawei just has to be a purchaser, end user, intermediate consignee, or ultimate consignee. In other words, Huawei just has to somehow be involved in the relevant product’s supply chain.

The Rise of the EDA Industry

            To better understand the impact of the Entity List export controls on Huawei, we need to take a look at the critical parts of the integrated circuit (IC) value chain. In this post, I’ll discuss the development of electronic design automation (EDA) tools for chip design. The software-based EDA industry is a recent phenomenon. Today’s dominant EDA vendors were all founded in the 1980s: Mentor Graphics (1981), Cadence (1987/1988), and Synopsys (1987). Before the 1980s, computer-aided design (CAD) functions were typically sold with their requisite hardware, workstations. Much of the earlier CAD work was done inside systems companies. For example, IBM was a systems company par excellence as it made both chips and the end products the chips went into, such as mainframe computers. Today there are EDA tools to cover the whole design process, usually referred to as the design flow.

Three things happened in the 1980s to create the software-based EDA industry we know. First, technological advances to develop software tools across all the important design functions proceeded apace so one could have a suite of EDA tools to cover the entire design flow. Second, the rise of general-task, powerful workstations allowed software-only EDA firms to gain a competitive edge. Third, standardization of EDA tools allowed them to gain wider acceptance in the commercial marketplace.

In the 1990s, a fourth development that spurred the EDA industry was the maturation of pure-play foundries making chips for fabless design firms. Neither the foundries nor the fabless design firms were going to spend the money required to develop in-house EDA tools, as larger firms such as IBM had done in the past. Consequently, fabless design firms relied on EDA vendors for their design tools, and the EDA firms began to work closely with the foundries to ensure that the EDA software could create fabrication-ready designs. 

The pace of innovation in IC design is so rapid and occurs across such an array of chip products that two features of the industry have emerged. Just to attempt to keep pace, the two largest vendors, Cadence and Synopsys, routinely spend 30 percent or more of their revenues on R&D each year. Consequently, the EDA industry has not yet developed one dominant platform for design and instead has an oligopoly of three large players. Indeed, it is not uncommon for different design teams within the same firm to use EDA tools from different vendors. Some go so far as to argue that each of the “Big Three” EDA vendors offers a superior tool in one specific segment of the design flow.1 Consequently, best practice is to use tools from all three vendors. Nevertheless, switching costs for entirely replacing one firm’s EDA tools with another’s appear high, so market shares have been fairly stable across the Big Three.2

The other feature is that acquiring new technology from start-ups is very common despite the Big Three remaining the dominant firms for three decades. The increasing amount of overall semiconductor R&D spending taken up by EDA firms (15.1% to 15.9% between 2016-2018) and the ability of venture-backed start-ups to take on higher-risk projects means these new firms make attractive targets for acquisition; at the same time, they find it difficult to compete with the established triumvirate. In other words, the start-ups can offer a particular advanced tool for a certain design task, but because these start-ups typically do not produce wider sets of tools, they have a very narrow competitive position. As a result, their best strategy more often than not is simply to sell themselves to one of the dominant firms.3 The Big Three have consistently captured two-thirds of the EDA market broadly defined over time. Since all three firms are based in America (Germany’s Siemens acquired Mentor in 2017) and their EDA technology is overwhelmingly of American origin, these firms’ EDA tools fall firmly under the Entity List export controls.

In the next post, I will discuss how the Huawei sanctions will impact Huawei’s chip design, global EDA firms and the local Chinese EDA industry. 


[1] To give some examples, Synopsys is regarded as having the best compiler, Cadence the best layout tools, and Mentor the best verification tools.

[2] One reason switching costs may be high is simply the fact that designers can grow comfortable with tools from one provider and getting down the learning curve with another company’s tools may appear formidably inefficient at the team if not individual level.

[3] Henkel, Ronde, and Wagner, “Entrepreneurship as a Contest.”

Create your website with WordPress.com
Get started