Foundational Technology Export Controls & Microelectronics
Foundational Technology Export Controls & Microelectronics
How to balance US and Chinese chip industry interdependence
Welcome to Semi-Literate, a guide to the chip industry through the lens of public policy. Feel free to forward to others you think may be interested.
BLUF: The Commerce Department is considering restrictions on sales of certain “foundational technologies” to China under the auspices of the 2018 Export Control Reform Act (ECRA). Public comments submitted by the US semiconductor industry in response to this proposal convey their extreme alarm at the prospect. These controls would potentially limit US microelectronics firm sales to their largest and fastest-growing export market, damaging China’s industrial policy goals in the short term and harming the US industry’s competitiveness in the long term. Broad controls could result in lost sales, reduced innovation, & trade diversion. New controls should target China’s state-of-the-art efforts, offsetting losses to US firms via CHIPS Act funds.
Introduction:
The Problem and Solution
In 2018 Congress simultaneously reformed the US investment screening system via the Foreign Investment Risk Review and Modernization Act (FIRRMA) and the US export control system via the Export Control Reform Act (ECRA). Congress was concerned about (1) Chinese ownership, control, and/or influence over US technology companies and (2) US technology company sales of their products to China. This was big-deal once-per-decade legislation designed to address a specific problem:
How does the US respond to massively-activist Chinese industrial policies designed to dominate the technologies that will confer the greatest economic and national security benefit in the coming decades?
The US export control system regulates dual-use exports (goods and technology that have both civilian and military uses). ECRA updated the export control system to address concerns about China’s attempts to seek global civilian and military leadership in advanced and emerging technologies through coordinated industrial policies. The Commerce Department is now taking ECRA’s intent and codifying it in regulations.
The Trade Offs
FIRRMA and ECRA have generated US business concerns because they tighten technology trade and investment with China, which is a growing market for many firms and a source of easy capital for others. This tightening is good in the eyes of a US government increasingly concerned by Chinese purchases of certain US advanced technologies. This is bad in the eyes of US technology companies who derive a large part of their cap table or profits from sales to/investments from China.
China Needs the US
One advanced technology the US is particularly good at making is chips. Part of China’s semiconductor industrial plans call for acquiring companies and technologies that make chips. This has resulted in well-documented tensions over Chinese attempts to buy “crown jewel” US chip companies to accelerate its industry’s progress. Traditional mergers and acquisitions by Chinese investors of US chip companies have been halted by the US government for the foreseeable future via the investment screening reforms in FIRRMA. But, Chinese purchases of chip technologies from US firms remain largely permitted, pending updated controls called for by ECRA.
As it attempts to ramp up domestic production of chips, Chinese industrial policies have increased demand for tools that enable chip design and fabrication. From 2015-19 China tripled its number of advanced chip factories and in 2019 HiSilicon (a Huawei chip design subsidiary) became the first Chinese fabless firm to rank among the top-10 semiconductor sales leaders worldwide. These advances were achieved using US chips technologies, resulting in a stark interdependence between the two industries:
in the past five years, the market for semiconductor manufacturing equipment (SME) in China has grown from 5% of the world’s total spending to over 25%. It is impossible to divorce the health of the U.S. semiconductor equipment industry from access to Chinese markets.
The US Needs China
Though Chinese semiconductor industrial policies have been modestly successful at best, the growth of the Chinese chip market has been great news for US SME and electronic design automation (EDA) firms eager to sell and license their tools to China’s new factories and chip design firms. As US firms sell their products to China, they reinvest the profits derived from these sales in to R&D, using their well-funded budgets to innovate and thereby maintain their technological leadership. China is basically funding the US chip industry’s innovation budget, ensuring it is even harder for Chinese chip firms to catch up with the industry leaders.
If the Commerce Department decides SME & EDA are “foundational” technologies, US firms could be prohibited from selling some of their products to key customers. This would be devastating for China’s technological ambitions in the short term and damaging for the US semiconductor industry’s competitiveness in the long term.
Summary of Foundational Tech Comments:
Are Chips “Foundational”?
Intuitively, most people know chips are foundational to life in 2021. The computer, laptop, tablet, or phone that you read this blog on is powered by chips. This blog is hosted by servers composed of chips. The software on which Substack functions was designed around the hardware capabilities of chips.
However there are a few technologies more foundational than chips: the software (EDA) and machines (SME) used to design and make the chips that go in the electronic devices that power daily life are more foundational than the chips themselves. As a result, when Commerce solicited comments on new export controls on foundational technologies, the US chip industry, especially SME companies, responded in force.
Background: Public Comments Process
Regulations.gov allows anyone to comment when a US government agency publishes an Advanced Notice of Proposed Rule Making (ANPRM) and those comments are then made public. These comments are a fantastic resource for policymakers and the level of detail in some submissions is remarkable. The only other time companies of this size publicly reveal so much about their own business practices and competition is in SEC filings, and the level of detail exceeds that of a 10-K in some respects.
The Commerce Department has posted 60 comments submitted by individuals, companies, and associations in response to its proposed foundational technologies rule-making. Comments were submitted by 13 chip companies and associations:
Organization Name || Length of Comments || % of Sales to China
The Semiconductor Equipment Association of Japan (SEAJ) || 2 pages
Marvell Semiconductor || 4 pages || 40%
Intel || 11 pages || 27%
Ampere Computing || 14 pages || Privately held
ASML || 15 pages || 15%
IBM || 17 pages || Unknown
SEMI (association) || 22 pages
Lam Research || 24 pages || 16%
Applied Materials || 25 pages || 29%
Cadence Design Systems || 33 pages, 11 (!) attachments || 10%
Teradyne || 38 pages || 22%
KLA || 38 pages || 16%
Semiconductor Industry Association (SIA) || 38 pages
Who are these companies, what do they do, why do they care?: A metaphor
The lay reader is probably not familiar with some of these companies so it may be confusing why they would cumulatively pay for 275+ pages of very expensive lawyer hours to comment on some obscure rule. A brief metaphor might be helpful:
Think of making a chip like baking a cake. You get out a recipe book, find the right page, pre-heat the oven, mix baking powder, flour, eggs, & sugar in a bowl, pour the batter in a pan, put that pan in the oven, and then pull it out when its done. Applied Materials, ASML, and Lam Research make ovens. Marvell, Ampere, and IBM come up with recipes, but don’t bake cake. Intel comes up with recipes AND bakes cake. Teradyne and KLA provide quality control, making sure all cakes look and taste right. Cadence publishes the recipe book and numbers your pages. Combined, they make very expensive cake.
These are big, well-funded firms that stand to lose billions if they cannot sell to a market that accounts for 10-40% of their annual sales. Applied Materials, Lam Research, and KLA have a combined market cap equivalent to that of Exxon. ASML is currently worth more than Volkswagen, the largest automaker in the world. The length of their comments loosely correlates to the level of concern they are feeling.
What the Comments Say:
All of these comments reflect company-specific equities, but its worth summarizing some overarching concerns and recommendations these comments share.
Main Concerns:
US semiconductor firms are extremely concerned poorly structured export controls on foundational technologies will:
Harm profits: reduce sales to their largest and fastest-growing market (China).
Harm innovation: reduced profits mean US firms have less money to invest in R&D to win through innovation. Every dollar US firms lose in sales is won by a competitor, which they then use to fund R&D to improve competing products.
Harm competitiveness:
Trade diversion: there is foreign availability of some SME and some EDA, so unilateral controls would advantage companies located in countries not subject to new controls. US companies may also choose to off-shore production to their overseas subsidiaries to avoid new controls.
Decreased long term viability: unilateral controls would push current customers to competitors, encourage customers to design-out US products, and undermine US firm recruiting & retaining of non-US talent.
Accelerate tech indigenization: restricting access to US SME & EDA will increase China’s sense of urgency to end reliance on the US industry.
Main Recommendations:
US microelectronics firms also realize the status quo is not tenable. In addition to explaining policies they are against, they share several recommendations:
Plurilateral controls (partner with allies), not unilateral controls: The US should partner with other countries that produce SME and EDA (especially Japan and in Europe) to minimize trade diversion associated with unilateral controls. The unilateral controls targeting Huawei in 2020 were damaging to US firms.
Narrowly define “foundational technologies”: Google’s proposal seems balanced
End-use/end-user controls, not technology-based controls
Use OFAC, not BIS, to target China (financial sanctions > export controls)
Focus on innovation (promote) > export controls (protect). Run faster to win.
A Case Study in Trade Offs: Cadence Design Systems
Introduction
Cadence is a real-deal crown jewel of the US semiconductor industry. Founded in the late 1980s in Silicon Valley, it has ~9,000 employees, annual revenue of $2 billion (investing 40% of that in R&D), and is one part of a three-company oligopoly that provides EDA tools for designing integrated circuits. Controls on exports of EDA to China were called out when targeting Huawei in 2020, so Cadence is rightly concerned by this new proposal and their extremely detailed comments reflect that concern.
What Cadence Does
If you are reading this blog, there is a 100% chance that you are doing so on a device that contains some Cadence IP. Their 10-K, which is understated, begins:
We offer software, hardware, services and reusable IC [integrated circuit] design blocks, which are commonly referred to as intellectual property (“IP”). Customers use our offerings to develop and integrate software that is key to the functionality of their products, as well as to design their ICs… Using our systems, customers deliver devices such as smartphones, laptops, gaming systems, automobiles, autonomous driving systems, cloud datacenter infrastructure, AI systems & medical equipment.
In plain English: if you want to make a chip you need to license IP blocks. Cadence is one of three companies that makes these IP blocks. Most electronic devices made since 2010 probably contain some Cadence IP. “Ubiquitous” is an understatement.
Why Cadence is a Good and Bad Target for Foundational Controls
Cadence illustrates the trade offs US policymakers face with respect to controls on exports of "foundational technologies.” To policymakers concerned about China’s efforts to make memory chips (commercial end use) and radiation hardened integrated circuits that go in satellites (military end use), it makes sense to target the underlying US-origin technology that enable both end uses. Two birds with one stone and all that. China needs Cadence (and its two EDA competitors, which combined have 90% of the Chinese EDA market) to design chips. At first blush, restricting Cadence/US EDA exports make for an appealing policy tool to slow China’s chip ascendence:
EDA is a foundational chokepoint: China needs it, the US has it
Limited foreign availability: US EDA firms enjoy 96% market share worldwide
First mover advantage: US EDA firms benefit from high barriers to entry, a massive IP moat, and deep pockets capable of buying upstart competitors. There have been basically no serious challengers to their oligopoly in decades.
China is not as important a customer for EDA firms as it is for SME: “only” 10% of Cadence’s annual sales are derived from China. Most fabless chip design (Cadence’s customers) is done by US, European, and Asian firms.
It’s worth noting here that Cadence attempts to rebut points 1, 2, and 3 above in their linked comments. There are also a few obvious drawbacks:
96% market share is not 100%. There are Chinese firms working on EDA tools. Cutting off access to US EDA tools would incentivize accelerating that work.
EDA tools are sold via software licenses. Licenses can be copied. Some of these tools have already been licensed in China so there may be no getting them back.
There are roughly 1,500 EDA engineers working in China, though only ~300 work for Chinese firms. The US EDA workforce would take a hit if those 1,200 employees could no longer work in China and/or chose to stay in China.
This mixed-bag of trade offs is indicative: every US microelectronics firm has complicated considerations. Here’s Cadence’s (hyperbolic, yet cautionary) conclusion:
Cadence submits that strict, unilateral export control regulations on US EDA tools will have the opposite effect as intended. They will curtail the US EDA industry, while accelerating growth and innovation for Chinese and other foreign EDA companies. Such regulations will decrease US EDA revenue, suppress innovation, cut jobs, and result in US EDA companies losing global relevance. Meanwhile, they will not impede Chinese companies in their ability to design and develop advanced electronic components, instead incentivizing growth in the Chinese EDA industry… and may cause Chinese EDA companies to surpass US EDA companies.
Conclusion: Supporting the US Microelectronics Industry while Reforming Export Controls
The picture painted above is complicated: China is clearly pursuing advances in microelectronics with the goal of promoting its economic and national security. The US microelectronics industry is profiting enormously from China’s pursuit, but risks upsetting a US government increasingly concerned by China’s advances. The US government wants to slow China’s ascendence, but doing so risks the profits and innovation which have made the US chip industry a crown jewel of the tech industry.
Restrict microelectronics exports to China & minimize US firm losses
The goal of US export controls on foundational technologies should be to slow (there is no stopping) China’s indigenization and development of advanced chip technologies:
Do not apply controls on US-origin commercial chips and avoid targeting Chinese companies unless justified (don’t be myopic about Huawei)
Focus foundational controls on sales of all SME and EDA to China required to produce the most advanced chips: anything below 16/14 nanometers.
For example: (1) allow EDA firms to continue licensing EDA packaging and printed circuit board tools in China (both have widespread foreign, and in-country, availability) but (2) restrict licensing of EDA digital IC design tools (logic synthesis tools, design for test tools, and IC manufacturing tools for resolution enhancement technologies in particular).
Make new controls on SME & EDA plurilateral. Ensure US competitors in Japan and the Netherlands don’t fill the void by increasing their sales to China. This will maximize the effectiveness of the controls and minimize US firm losses.
One estimate of the value of lost sales from export controls on <16nm SME from the US, Japan, and Netherlands in 2021 is $920 million combined.
Mitigate the cost of lost sales via CHIPS Act funds, using money made available in sections 9902, 9903, and 9905 to offset reduced R&D budgets of US SME and EDA firms and US allies. (side note: Offsetting $920 million is nothing. DOD gave GlobalFoundries $400 million last November and no one noticed.)
Means test the offsets: the more a company can demonstrate lost sales to China stem from new export controls (as Teredyne and KLA did in the comments linked above) the more money they should be eligible for.
Reform Export Controls
The export control system is in desperate need of reform more generally. It’s still operating in mostly the same structure its had since The Cold War. But, export controls are technical and arcane and most of the costs borne by the export control system are B2B, so public interest in reform is limited. The most substantive recent proposal on reforming the US export control system came from CNAS.
A joint comment by Microsoft/OpenAI is by far the most creative proposal to reform the entire US export control system. Microsoft picked up Bill Chappell, the former head of DARPA’s Microsystems Technology Office, to become its Chief Technology Officer for Azure in 2019. Its ambitious proposal to embed software features, hardware roots of trust, and tamper-resistant tools in to foundational technologies echoes some of the work the MTO pursued under his leadership. However, if the BIS website is any indication of the agency’s technical acumen, this proposal is a bit beyond them. The US export control system operates in black and white, so the proposals above are going to require creativity from agencies that aren’t known for it.
Note: the views here are my own and drawn solely from the documents I cite here.