Export controls as policy: instruments and mechanics

An export control is, in legal mechanics, a license requirement, not a categorical ban. A controlled item (a machine, a chip, a piece of software, a service) may be exported to a destination only after the relevant authority grants a license, or after the exporter establishes that no license is required under the regulation's exceptions.

The three policy variables a regulator can dial are:

• Coverage — which items, which technologies, which services fall under control.
• Destination — which countries, end-users, or end-uses require a license.
• License policy — whether the default is approval, presumption of denial, or hard denial.

These variables can be set independently. A control can be coverage-narrow but destination-broad, or coverage-broad but destination-narrow. The political debates often conflate these axes; the legal instruments do not.

The rest of this lesson examines the specific instruments used by the United States, the Netherlands, and Japan, and the mechanics by which they apply to semiconductor manufacturing.

The US Department of Commerce's Bureau of Industry and Security (BIS) operates the Export Administration Regulations (EAR). Three instruments within the EAR do most of the work on semiconductors.

• The Entity List. A published list of specific firms and organizations for which exports of controlled items require a license, typically under a presumption of denial. Adding a firm to the Entity List is the most visible and discrete tool.
• Parameter-based controls (the Commerce Control List, CCL). Items are described by technical parameters — gate length, density, transistor count, clock frequency, bandwidth — and a control applies if the item exceeds the threshold. This is how 'advanced' chips and equipment are defined in writing rather than by name.
• End-use and end-user catch-all controls. Controls that apply not because of the item's parameters but because of its end-use (e.g., supercomputing, semiconductor manufacturing in certain destinations) or end-user (e.g., a military intelligence end-user).

The three instruments often stack on the same transaction. A given export may need to clear all three independently.

Parameter-based controls let a regulator define 'advanced' without naming a firm. For semiconductors, the published thresholds use measurable properties.

Illustrative examples drawn from the structure of the regulations (specific numbers move as the regulations update):

• Logic chips: non-planar transistor architecture with a node identifier of 16/14 nm or less, or with a gate length below a published threshold.
• Memory chips: NAND with 128 or more layers; DRAM with half-pitch below a published threshold.
• AI accelerators: total processing performance (TPP) and performance density above published thresholds, often in TOPS at specified numeric precisions.
• Manufacturing equipment: lithography systems with critical dimension below a threshold; etch systems for high-aspect-ratio features above a threshold.

The parameter approach has structural implications. It is technology-agnostic (a firm that achieves the parameter by any process route still falls under the control) and legally precise (a license officer can determine in/out by reading specs). It is also self-updating only when the regulation is amended, so thresholds can drift relative to industry capability between updates.

The Foreign Direct Product Rule (FDPR) extends US export jurisdiction extraterritorially to items produced outside the US that are the 'direct product' of certain US-origin technology or software.

The mechanics: if a chip is manufactured at a non-US fab, but the fab uses US-origin equipment, or US-origin design tools, or runs a process developed with US-origin technology, the resulting chip can fall under US export jurisdiction even though no physical US export occurred.

The rule's reach depends on which specific FDPR variant the regulation invokes. Variants used in semiconductor policy include:

• A general FDPR for items destined to specific Entity-List end-users.
• An 'Advanced Computing FDPR' covering chips above the parameter thresholds and specific destinations.
• A 'Supercomputer FDPR' covering items used in supercomputing end-uses.

The FDPR is structurally consequential: because so many fabs globally use US-origin equipment and software, the rule effectively gives the US a license-veto over much non-US chip production destined for covered end-users, even when no US person or US territory is involved in the transaction.

The US controls do not operate alone. The Netherlands and Japan added complementary measures starting in 2023, each under its own legal basis.

• The Netherlands. Acts through its dual-use export-control regime, with specific national rules on advanced lithography equipment. Because ASML is based in the Netherlands, Dutch controls on EUV and on certain advanced DUV immersion scanners directly affect what equipment can ship to which destinations.
• Japan. Acts through its Foreign Exchange and Foreign Trade Act (FEFTA). In 2023 Japan added 23 categories of semiconductor equipment to its license-required export list, covering deposition, etch, lithography, cleaning, and inspection tools. The categories were chosen to align with US controls without naming specific destinations.

The legal independence matters: each country's measures are enforced under its own law, with its own license process and appeal mechanisms. The result is a coordinated effect — equipment from the three countries' major vendors faces aligned controls — without a single multilateral treaty.

The alignment is not automatic. It depends on continuing coordination between the three jurisdictions, and gaps can open if any one updates its rules slower or faster than the others.

Once items and destinations are defined, an exporter still needs to know whether a particular shipment requires a license. The license logic for semiconductor exports involves several layers.

• Item classification. Is the item on the CCL? If so, under which Export Control Classification Number (ECCN)?
• Destination check. Is the destination subject to a country-group restriction? Is the consignee on the Entity List or other restricted-party list?
• End-use check. Is the item destined for an end-use that triggers a catch-all control (e.g., 'semiconductor manufacturing' in covered destinations)?
• End-user check. Is the end-user a controlled entity, a military end-user, or a military intelligence end-user under the relevant rules?
• De minimis and FDPR check. Does the item's US content exceed the de minimis threshold, or does an FDPR variant apply?

For a US-origin EUV scanner exported to a covered destination, multiple layers trigger simultaneously. For an export of a finished consumer GPU above the AI-accelerator parameter thresholds to a non-covered destination, fewer layers trigger, but the exporter must still establish end-use to avoid catch-all controls. Mistakes are enforced as civil penalties, denial of export privileges, or criminal prosecution in the most serious cases.

Enforcement runs through several mechanisms.

• License applications. Exporters submit license applications to BIS. Licenses can be granted, granted with conditions, or denied. The process generates a record that becomes evidence in later proceedings.
• Compliance audits. BIS conducts post-shipment verifications and end-use checks. A vendor who ships under one license and discovers post-hoc that the items were diverted faces a different liability than one with no diversion signal.
• Re-export controls. A shipment to country A that is later re-exported to country B is treated as if the original exporter had shipped to country B. This forces vendors to track downstream destinations.
• Secondary sanctions. Distinct from the EAR, the Office of Foreign Assets Control (OFAC) administers sanctions that can apply secondary penalties to non-US persons who transact with sanctioned parties.
• Smuggling and diversion. Goods can be re-routed through third countries, repackaged, or relabeled. Enforcement against diversion depends on physical inspections, financial-flow analysis, and information-sharing between regulators.

The structural pattern: controls work in part by raising compliance cost. A purchaser who is technically able to obtain a restricted item via diversion still pays a premium, accepts longer lead times, and forfeits the support and software updates that come with an authorized channel.

Export controls on semiconductors are unusually effective compared to controls on more diffuse goods. The reason is the industry concentration described in the earlier lessons.

A control on a fungible commodity — wheat, steel, basic chemicals — leaks through arbitrage: many sellers, many ports, many substitutes. A control on EUV scanners does not, because there is one vendor with one supply chain, and that vendor sits in a jurisdiction that participates in the controls.

A control on advanced AI accelerators is harder to enforce than EUV (more units, more channels, more buyers), but easier than commodities, because design and manufacture concentrate to a few firms whose finished-goods flows can be monitored.

A control on photoresist or specialty gases is also feasible because the qualified-supplier base is small, but is rarely used as a public lever because the secondary effects on the global civilian supply chain would be large.

The lesson from a structural standpoint: the same horizontal concentration that makes chip manufacturing efficient also makes it amenable to narrow policy intervention. Whether a given intervention achieves its stated goals depends on enforcement, diversion, and the substitution responses of the firms and countries affected. Those are empirical questions for the next decade, not predictions to be made here.