Insight into the U.S. Semiconductor Export Controls Update

The Bureau of Industry and Security released a new set of rules this week aimed at updating and complementing its regulations from October 7, 2022. Last year’s rules were intended to target the People’s Republic of China’s (PRC) military artificial intelligence (AI) development by curbing the country’s access to advanced semiconductors and the tools used to manufacture them. This week’s update aims to curb workarounds to the previous regulations, filling in critical gaps in U.S. policy that will enable the Biden administration to hinder China’s access to AI chips more effectively.

Q1: What was the content of the original October 7 rules?

A1: The October 2022 controls targeted equipment and technology for advanced semiconductors, drawing specific lines in the sand in an effort to curb China’s access. The new rules added new license requirements for items to be used in supercomputers, the production of chip manufacturing equipment, and destined to production facilities in China meant for logic chips with nonplanar transistor architectures, or with a production technology node of 16–14 nanometers or below, dynamic random-access memory (DRAM) chips of 18-nanometer (nm) half-pitch or less; or NAND flash memory chips with 128 layers or more. To that end, BIS made four new Export Control Classification Numbers (ECCNs) on the Commerce Control List (CCL):

  1. 3A090, which concerns some advanced ICs that can have transfer rates of 600 gigabytes or more
  2. 3B090, which concerns semiconductor manufacturing equipment and related items
  3. 4A090, which concerns computers, assemblies, and components that include integrated circuits (ICs) over the limit delineated in 3A090a
  4. 4D090, which concerns software tailored to developing items controlled under 4A090.

In addition to these new ECCNs, BIS released some revisions related to chip software and technology and tightened the use of license exceptions for some exports, re-exports, or in-country transfers regarding the PRC. These restrictions were complemented by three new foreign direct product rules (FDPRs). Lastly, BIS released licensing requirements based on end-use for semiconductor manufacturing and supercomputers, as well as restrictions on U.S. individuals’ participation in the production of advanced ICs in China.

Q2: What loopholes in the 2022 rules did Chinese firms find?

A2: As a result of the U.S. controls, the Chinese semiconductor ecosystem faced serious headwinds. After the October 7 controls were announced, Apple dropped Yangtze Memory Technologies as a supplier and, two months later, the Chinese homegrown champion began laying off 10 percent of its workforce. Likewise, SMIC saw some of its expansion plans hindered by U.S. export controls: co-CEO Zhao Haijun announced that operations had been stalled due to difficulties acquiring “bottleneck equipment” in a post-earnings meeting. during a post-earnings meetings that operations had been stalled du to difficulties acquiring “bottleneck equipment”. Nevertheless, PRC firms reliant on high-tech chips found workarounds. Several Chinese artificial intelligence companies were found to be circumventing the rules by accessing chips via intermediary companies, posing questions for the United States’ long-term ability to enforce its own export controls. CSIS’s previous work has noted iFlyTek’s workarounds, which used cloud services providers and rent access through third parties to access Nvidia’s A100 chips. Another Chinese company, SenseTime, has also been accused of leveraging intermediaries to purchase banned AI components. According to the Center for a New American Security (CNAS), other blacklisted entities, such as China’s top nuclear weapons lab, have been gaining access to restricted chips though smuggling—in addition to renting through the cloud. Smaller Chinese businesses also used smuggling to acquire restricted chips, and about 40,000 to 50,000 of these chips are already in China.

Q3: How does this update complement the October 7 rules?

A3: Broadly speaking, the October 17, 2023, update builds on the October 7 rules by setting tighter restrictions on AI chips, strengthening restrictions to semiconductor manufacturing equipment, and adding firms to the Entity List.

In the first new rule, which focuses on advanced computing and semiconductors, BIS has reshaped how ICs are restricted. BIS said that the original controls allowed China access to chips that could provide AI model training very similar to the cutting edge. The standard for export compliance has been changed to mitigate for this; the controls are now based on two metrics called total processing performance, or TPP, and performance density. A tiered system has been written, breaking chips into two categories:

Tier 1:

  • 4,800 TPP and above
  • 1,600 TPP and above and a performance density of 5.92 or more

Tier 2:

  • 2,400–4,800 TPP and a performance density between 1.6–2.92
  • 1600 TPP and above and a performance density of 3.2–5.92

Each of these tiers is controlled, but chips in the lesser tier will be offered a license exception by BIS for certain countries. This rule update ultimately gives BIS a much larger remit over AI chips, and it will cut China’s access to these chips dramatically. Importantly, chips that fall outside of these tiers could fall into a chip “gray zone.” These chips, which are still relevant to national security, will still require firms to notify the U.S. government of their exports. Chips specifically meant for non-military end uses will, for the most part, not require a license. Licensing requirements have also been expanded to over 40 countries that BIS has flagged for national security or missile technology purposes as well as countries with which the U.S. has arms embargoes, making transshipment to China significantly more difficult.

In the second update, BIS strengthened restrictions on semiconductor manufacturing equipment (SME). Additional technologies that aid in the production of advanced SME were added to the list, and the number of restricted countries was increased to 23, applying to countries with which the United States has an arms embargo. These added SME products are not multilaterally controlled, but BIS claimed that the rules update was immediately necessary. Additionally, the FDPR has seen an expansion. As noted by Thomas Krueger at CNAS, the Commerce Department has revised FDPRs to include countries of concern beside China to counter the risk of diversion. This FDPR update also give the U.S. greater extraterritorial control over certain high-end tools. BIS has stipulated that this licensing policy will be a “presumption of approval” for countries not subject to an arms embargo.

Lastly, in the third rule update, 13 Chinese firms were added to the Entity List. These firms were identified by BIS as being “involved in the development of advanced computing integrated circuits.” Additionally, the added entities were given a “footnote 4 designation,” meaning they are subject to the restrictions of the FDPR.


Q4: How has U.S. industry been affected by previous controls?

A4: The October 7, 2022, controls have affected U.S. industry in two major ways: (1) loss of Chinese market demand for leading-edge chips and their associated technologies and (2) retaliatory controls and sanctions.

In the weeks following the October 7 controls, U.S. companies began to announce the negative impacts in terms of chip offerings in China. Nvidia announced that its A100 chip, a leading AI processor, was not compliant with the new rules. Advanced Micro Devices said that it would be unable to sell its NI250 chip in the Chinese market. Lam Research, a U.S.-based provider of chipmaking technology, warned that it could lose up to $2.5 billion in revenue shortly after the previous controls were announced.

Nvidia, however, changed its AI chip offering in China after October 7. The A100 chip, a high-performance AI chip, was above the limit set by previous controls. Nvidia decided to produce new chips, the A800 and H800, which had performance characteristics just under the limit of the U.S. regulations. Their cut-down chip had a transistor size of 7 nm, is made by TSMC, and is exported to Chinese companies. The A800, while compliant with October 7 controls, is still capable of performing complex AI tasks, and was a top seller in China before the updated October 17, 2023, controls. The new rules, however, also closed this loophole and cover these chips.

The Chinese government, in response to the October 7 controls, took retaliatory measures. In July 2023, China put export restrictions on critical raw materials for chipmaking such as gallium and germanium products. Micron, an Idaho-based semiconductor firm, was subject to a “cybersecurity review” by Chinese officials. The move threatened to evict Micron from the Chinese market, costing them $3.3 billion—however, the review seemed to only be a scare tactic, as the end result was a limited product ban.

Additionally, the Chinese government has been working to convince domestic technology firms to source their inputs domestically rather than from U.S. suppliers. This threat of designing-out U.S. critical technological inputs threatens U.S. industry en masse.

Q5: How has U.S. industry been reacting to the expectation of new controls?

A5: In recent months, as the government has been communicating their desire to strengthen controls, U.S. semiconductor companies have urged caution.

CEOs for both Intel and Nvidia have made recent trips to China, where they have publicly argued that additional restrictions would hamper their ability to access the Chinese market. “We have become unable to sell more advanced semiconductor chips to one of our largest markets,” Nvidia CEO Jensen Huang said.

In July 2023, the Semiconductor Industry Association (SIA)—which represents U.S. semiconductor companies—called on the United States and China to “ease tensions and seek solutions through dialogue, not escalation” in response to rumors of additional export controls. The statement added, “Allowing the industry to have continued access to the China market, the world’s largest commercial market for commodity semiconductors, is important to avoid undermining the positive impact of this effort.”

In the immediate aftermath of the updated October 17 export controls, the SIA released an additional statement saying “Overly broad, unilateral controls risk harming the U.S. semiconductor ecosystem without advancing national security as they encourage overseas customers to look elsewhere. Accordingly, we urge the administration to strengthen coordination with allies to ensure a level playing field for all companies.”

Q6: What has been left out of the controls?

A6: The controls, which were mostly targeted at AI chips, are missing a key workaround component: cloud computing. A gap in export control policy allows Chinese firms to access highly advanced computers physically located in different countries through the cloud. In the updated rules document, BIS states that they are “also concerned regarding the potential for China to use [infrastructure as a service] solutions to undermine the effectiveness of the October 7 [interim final rules] controls and continues to evaluate how it may approach this through a regulatory response.” However, they do not directly address the issue with a policy change—instead they call for public comment on the issue. It was reported in July that BIS was considering a rule change, so the solicitation for public comment indicates that an additional rules update is forthcoming.

Another solicitation for public comment refers to a fundamental problem in semiconductor exports: how to control products that leverage techniques allowing non-cutting-edge AI chips to be used to train small- or medium-scale AI models. Controlling products that utilize multiple integrated circuits is offered as a possible solution, but BIS ultimately asks for technical proposals from stakeholders.

Conclusion

The Commerce Department’s expansion includes more products and countries and adds provisions to curb Chinese circumvention on supercomputing and artificial intelligence controls. The three updates, respectively, expand the breadth of controls on manufacturing equipment, add restrictions on advanced chips, and place 13 new entities to the Entity List. Secretary of Commerce Gina Raimondo said in a statement that these expansions would “increase effectiveness of our controls and further shut off pathways to evade our restrictions.” The Commerce Department seems to work towards that goal by broadening the geographic scope of the U.S. controls and significantly expanding licensing requirements to additional countries. That expansion, in addition to the fact the administration seems to intend to update the rules annually, renders effective enforcement even more critical than before. It also highlights the need to ramp up the United States’ capabilities in the semiconductor sector by spurring innovation in its homegrown ecosystem and diversifying supply chains of legacy chips in critical sectors.

William Reinsch holds the Scholl Chair in International Business at the Center for Strategic and International Studies (CSIS) in Washington, D.C. Matthew Schleich is an intern with the Scholl Chair in International Business at CSIS. Thibault Denamiel is a research associate with the Scholl Chair in International Business at CSIS.

Matthew Schleich

Research Intern, Scholl Chair in International Business