MAIN CONTENT
A manufacturing route narrows
Reuters reported that TSMC had directed affected customers to suspend manufacturing orders at 16 nanometres and below unless a United States licence is obtained. Its sources said that 28- and 40-nanometre production remains a possible route for some Chinese chip designers when products do not use high-bandwidth memory or require advanced packaging. These are reported company instructions, not evidence that every Chinese customer faces identical restrictions. The distinction matters because restrictions depend on chip characteristics, counterparties and end uses as well as the nominal manufacturing node.
The figures describing a chipmaking process are often treated as a simple ladder of technological progress. In practice, the smaller process generations generally offer higher transistor density and can support better power efficiency, but a processor's usefulness also depends on its architecture, software, memory and packaging. Leading AI accelerators commonly rely on advanced production nodes as well as high-bandwidth memory, which transfers large quantities of data to computing units. Removing access to any critical part of that ecosystem can slow the delivery of a commercially usable product.
A mature process is not useless: it continues to power automotive electronics, industrial controllers and numerous everyday devices. Yet making a competitive large-scale AI accelerator at those older nodes can involve significant penalties in power use, size and cost. The economic effect of the new restrictions therefore extends beyond individual purchase orders to research schedules, cloud computing budgets and the decisions investors make about domestic semiconductor projects.















