A revised US semiconductor export control rule from the Bureau of Industry and Security (BIS) at the Department of Commerce took effect on November 19, 2026. That same day, Reuters reported, citing two people familiar with the matter, that TSMC had told staff to suspend some orders of 16nm-and-below chips from Chinese AI chip designers until US export licenses are obtained. Under the revision, when a company that makes or assembles semiconductors uses transistor counts to prove a chip falls outside export controls, conditions on the location of the designer and the headquarters location of its ultimate parent company now also apply. Even for chips made with the same manufacturing process, the available method of proof now varies depending on the designer.
In announcing the revision on November 16, BIS explained in Section V of the rule's preamble that it had tightened the conditions under which manufacturers and assemblers can overturn the existing presumption in item classification. TSMC told Reuters by email that it would comply with the new rule, but declined to comment on market rumors.
Why are 16nm/14nm-class chips presumed to be controlled?
The mechanism that presumes 16nm/14nm-class logic ICs to be advanced products was introduced in a rule published and effective on January 16, 2025. It covers logic ICs that use a 16nm/14nm-class or smaller manufacturing process, or that adopt a non-planar transistor structure such as FinFET. When front-end foundries that form circuits on semiconductor wafers, or outsourced semiconductor assembly and test companies (OSATs), export, reexport, or transfer in-country covered products, those products are presumed to fall under Export Control Classification Number (ECCN) 3A090.a of the US Export Administration Regulations (EAR) and to be designed and marketed for data centers. The compliance deadline for that provision was January 31, 2025.
The technical thresholds of 3A090.a are a total processing performance (TPP) of 4800 or more, or a TPP of 1600 or more combined with a performance density of 5.92 or more. Performance density is TPP divided by the applicable die area. The manufacturing process and transistor structure are used for the presumption of classification under Note 1, while TPP and performance density are the criteria for judging whether a product itself falls under 3A090.a. If a front-end foundry or OSAT cannot overturn the presumption, it must follow the license requirements that apply to 3A090.a.
New location conditions added to transistor-count proof
The revision adds conditions on the designer's location and the headquarters location of the ultimate parent company when a front-end foundry or a BIS-approved OSAT uses approximate transistor counts to overturn the presumption that a chip falls under 3A090.a. Meanwhile, the method by which a BIS-approved designer, or a designer meeting the rule's recognition conditions, proves non-applicability using performance data remains in place.
The specific changes become clear by comparing paragraphs a, b, and c of Note 1 as published in the Federal Register of January 16, 2025 (90 FR 5314) and the Federal Register of November 19, 2026 (91 FR 74987).
| Note 1 method of proof | January 2025 rule | November 2026 revision |
|---|---|---|
| a: Performance proof by the designer | A BIS-approved designer, or one meeting the rule's recognition conditions, proves non-applicability of 3A090.a using TPP and performance density data | Existing method of proof retained |
| b: Proof by the front-end foundry | A front-end foundry that did its own packaging outside Macau and Country Group D:5 certifies the total approximate transistor count | In addition to the existing conditions, the designer's location and the ultimate parent's headquarters must be outside the specified countries and regions |
| c: Proof by an approved OSAT | A BIS-approved OSAT that handled the packaging certifies the total approximate transistor count | In addition to the existing conditions, the same designer and ultimate-parent location conditions as in b apply |
Under methods b and c, the approximate transistor counts of the covered logic dies in the final package are summed. The basic threshold is fewer than 30 billion, but for products without high-bandwidth memory (HBM), thresholds of fewer than 35 billion for covered transactions completed in 2027 and fewer than 40 billion for transactions completed in 2029 and later can also be used.
The new location conditions cover Country Groups D:1, D:4, and D:5, plus five countries specified in EAR §742.6(a)(6)(iii)(A)(1): Indonesia, Malaysia, Singapore, Brazil, and Thailand. For example, even if the same manufacturer packages a chip at the same facility and the transistor count is below the threshold, methods b and c cannot be used if the designer's ultimate parent is headquartered in a specified country.
Export licensing for individual transactions is governed by the newly established §744.23(a)(3)(i)(C) (91 FR 74977–74978). For specified advanced logic ICs subject to the EAR and falling under 3A090.a, if they were designed by a company that has neither BIS approval nor recognition under the rule, a license is required regardless of destination when a front-end foundry or OSAT exports, exports from outside the US, reexports, or transfers them in-country. The data-center remote access restrictions included in the same rulemaking are set out separately in §744.24.
Does this affect 16nm automotive chips too?
According to Reuters, a person who had reviewed TSMC's guidance said that affected Chinese designers would effectively be limited to older process generations such as 28nm and 40nm. That applies, however, to products that do not use HBM and do not require advanced packaging. Scrutiny is also said to be tightening not only on customers' locations and parent companies but also on back-end subcontractors and similar parties.
Nikkei Asia also reports, citing multiple industry executives, that the rule affects chips for mobile and consumer devices and designers outside China.
16nm logic ICs are used not only in AI products but also in cars. According to materials announced by TSMC and NXP on June 2, 2021, the two companies began volume production of the S32G2 automotive network processor and the S32R294 automotive radar processor, both using TSMC's 16nm FinFET process. The S32G2 is used in gateways that connect in-vehicle networks, and the S32R294 handles signal processing for radar that detects surrounding objects.
If a front-end foundry were to export the S32G2, the 16nm FinFET specification in the joint announcement shows only that the manufacturing process and transistor structure meet the Note 1 presumption conditions. To determine whether the presumption can be overturned, one must check the designer's location, the ultimate parent's headquarters location, and the status of BIS approval or recognition. Then, depending on the method of proof used, one needs either the TPP and performance density shown by the designer, or the location of the company and facility that handled packaging along with the approximate transistor count of the covered dies. The applicable transistor-count threshold also differs depending on whether HBM is included and the year of the transaction. Whether a license is actually required would be decided by confirming the scope of the EAR and then applying the relevant provisions to the item classification, the exporter, and the type of transaction.
The Reuters report and TSMC's response alone do not reveal which companies or product part numbers the measure reached, how long any suspension lasted, or whether individual export licenses were obtained. Nor has it been confirmed whether the two NXP automotive chips require licenses under the new rule or how their shipments are affected. To gauge the scale of the supply impact, it will be necessary to distinguish, among suspended orders, those where shipments actually stopped from those where deliveries later resumed.
