The Geopolitics of Chips — Export Controls and Supply Chain Rewiring, Explained Technically

Why chokepoints form exactly where they do, from three conditions: physics, fixed cost, and tacit knowledge. Covers the technical reason controls are written as numeric thresholds, the common-cause formula that kills dual sourcing, and the learning curve that sets how fast a substitute can arrive.

Why some parts have no substitute

Break a bolt and you can buy the same bolt anywhere on Earth. But a semiconductor line contains steps where, if the supply stops, no replacement appears for years. These knots are called chokepoints.

The intuitive explanation is that the equipment is complicated and expensive. That can't be the whole story: a jet engine is also complicated and expensive, and several companies build them. What decides whether a chokepoint forms is not price or complexity but the shape of the constraints the step sits under.

Think of the difference between a recipe and a cook. A recipe — a drawing, a spec sheet — fits on paper, so you can copy it and hand it over. Hand over the same recipe and you still don't get the same dish. The heat adjustments, the quirks of one particular pan, the compensation for a bad batch of ingredients: none of that fits on the page, and it lives only inside someone who has been cooking for years. Semiconductor manufacturing has steps where the share of "doesn't fit on paper" is extreme.

This article explains where the knots are from the technical reasons that put them there. For the map of who holds what, see The Semiconductor Supply Chain, End to End; here the question is why a given step becomes a knot, and how long it takes to untie one.

Three conditions that make a chokepoint

Wherever a knot forms, all three of these are present.

Condition 1: physics removes design freedom. When you print fine patterns with light, the thinnest line you can draw is set by wavelength and numerical aperture. To go finer you shorten the wavelength — but shorten it far enough and glass absorbs the light, so lenses stop working. The way past that dead end was 13.5 nm extreme ultraviolet, and from that choice everything else follows: vacuum, mirrors instead of lenses, and a light source built for that one wavelength (see EUV Lithography). Where the design tree has few branches, the field collapses to one winner — because no alternative solution physically exists.

Condition 2: fixed cost has too few units to spread across. If development costs $F$ and lifetime shipments are $N$, each unit carries $F/N$. When a tool sells a few dozen units a year and $F$ is enormous, a second entrant has no path to recovering the same investment. The fewer buyers a machine has, the more the sellers converge to one — a force that runs opposite to intuition.

Condition 3: some knowledge never reaches the drawing. A process recipe looks like a set of numbers — this temperature, that many seconds — but it is really the residue of years of tuning against the quirks of specific tools in a specific building. Buy the same tool, dial in the same numbers, and you do not get the same result. Moving a process therefore requires not just the equipment and the recipe but the people who were running it.

Swap one input, and verification explodes

"Just switch to a domestic material" is easy to say, but process steps are not independent. Change the resist and it interacts with the surface preparation before it and the develop and etch steps after it.

For $n$ factors, a full factorial experiment that catches every interaction takes $2^n$ runs. In practice you cut that down with fractional factorial designs — but what you cut becomes the set of combinations you never looked at, and those can surface after you are in volume production.

A tool is not a one-time purchase

There is a second dependency that is hard to see from outside: buying a production tool does not turn it into an independent asset. It runs on a stream of spare parts, periodic calibration, and control-software updates, and every time you shift process conditions you rebuild the tool-side parameters too.

That has a practical consequence: when supply stops, the effect arrives gradually rather than on the day it happens. Uptime holds while the spares last, and more tools fall out of service as the shelf empties. "It's still running, so we're fine" is a judgment that misses the lag.