Read the claim, not the title. US20260223625A1, published 30 July 2026 and assigned to Tokyo Electron Limited, is titled as a metrology filing, and its first two independent claims read that way — measure a wafer's topography, let a process model generate the bonding recipe, bond. The third independent claim is a different kind of thing.

Claim 18 recites a method of bonding wafers using a plurality of wafer bonding systems. A first system measures a wafer, generates a recipe, bonds, and then transmits at least one of the measured physical parameter or the recipe to a second bonding system — which bonds its own wafers in accordance with what it received. Claim 21 keeps the loop open afterwards, updating the recipe used by either system based on a later measurement.

The method of claim 18, wherein transmitting comprises transmitting both the physical parameter and the wafer bonding recipe.— Integrated Metrology for Process Controls in Wafer Bonding System, US20260223625A1

Why a transmission step is the commercial one

Bonding tools are capital equipment bought in multiples. A fab running hybrid bonding at volume operates a set of them, and the practical problem is not whether any single tool can be tuned — it is that each tool drifts on its own and each has to be qualified, matched and re-matched against the others. Tool-to-tool matching is where a large share of the process-engineering effort in a bonding module actually goes.

A claim that covers moving the measurement and the derived recipe between systems is a claim on the mechanism that makes a fleet behave as one instrument. Note what the claim does not require: it does not recite a network, a fab host, a manufacturing execution system, a data format, or a common vendor. It requires only that the parameter or the recipe travel from the first system to the second. Read plainly, that reaches the transfer irrespective of how it is carried.

For an equipment supplier, this is coverage at the layer where installed base compounds. The value of the second tool rises because the first one has already characterised the process, and the derived recipe is the artefact carrying that value. It is worth being explicit, though, that the record says nothing about products, licensing, tool counts or commercial arrangements. That reading is an inference from the claim structure, not a disclosure.

The two independents underneath establish where the transmitted knowledge comes from. Claim 2 is the feedforward path: measure a physical parameter of the incoming wafer — at least one of overlay, in-plane distortion, slope, curvature, force or stress — and have a controller holding a model of the bonding process generate the recipe before bonding. Claim 10 is the feedback path: bond, measure the post-bond wafer, generate a second recipe from that result.

The instrument lists are specific and slightly different between the two, which is a real detail rather than a drafting accident. Both recite a dual Fizeau wafer interferometer, a Shack-Hartmann sensor, a PhotoStress analysis system, a wavefront phase sensor, a multi-beam reflectometry system and capacitance sensors; claim 10 adds a scanning interferometric sensor, tied by claim 15 to measuring topography of the post-bond wafer. The feedback path gets an instrument suited to a bonded stack rather than a free wafer.

Claim 3 shows where the model originates, and it is notable that the answer is simulation rather than accumulated production history — measure two wafers, simulate their bonding under given process conditions, estimate the post-bond result, and build the model from that. A supplier can therefore ship a usable model with the tool instead of requiring a customer to generate one from scrapped lots, which shortens the qualification period a customer has to fund before a new bonder earns anything. That is the part of an equipment sale that is hardest to compress and least visible in a price list.

Claim 7 adds a lot-level frame: the first wafer and a third wafer belong to a first wafer lot, the second and fourth to a second lot, and the two lots are processed together in a semiconductor fabrication process flow. That is the claim acknowledging the unit a fab actually schedules, and it matters for claim 18 because a recipe derived from one lot is being applied by a second tool to wafers that may not share that lot's history. The claim does not require them to.

Limits on the read

This is a published application, not a granted patent. It confers nothing enforceable in this form, has not been examined on the public record, and the independents — claim 18 in particular — are broad enough that narrowing should be expected. The published set also shows clear amendment traces: claim 1 is (canceled), the wafer ordinals skip (claim 2 goes from a second wafer to a fifth and sixth, producing a third post-bond wafer with no second one introduced), and claim 4 depends from claim 2 while reciting terms that appear only in claim 3.

The same week gives a second data point. Tokyo Electron had a bonding patent issue on 28 July — US12696826B2, on using a laser liftoff layer to move a die between carriers before bonding it to a wafer carrying a second die. A granted position on performing the bond and a pending one on governing it, two days apart, is a supplier filing across both the mechanism and its control layer. For a buyer evaluating a bonding module, those two records describe different things to negotiate over: one is a process step, the other is the data the step produces.