Every chiplet roadmap eventually runs into the same wall: you can only place dies as fast as you can align them. In conventional pick-and-place for heterogeneous integration — the practice of building one package out of separately fabricated dies — a head grips a single die, a vision system finds it, the head corrects the offset, and the die comes down. The cycle is serial, and it repeats once per die. As package die counts climb, that serial alignment loop becomes the dominant term in tool throughput. A published US patent application from ASML Netherlands B.V., US20260206525A1, titled "Recess-Based Pick and Place for Heterogeneous Integration," describes an apparatus built to break that loop by moving the alignment work off the pick head entirely.

The application published on July 16, 2026. It is an A1 publication — a pending application, not a granted patent — and what it discloses is a machine architecture rather than a process recipe. Claim 1 is an apparatus claim with three recited elements. The first is a first stage carrying a population of machined pockets: a first stage comprising a plurality of recesses, the plurality of recesses configured to accept a plurality of donor dies. The second is a second stage comprising a support for one or more targets — the receiving wafer or substrate. The third is a measurement system functionally coupled to the first stage.

That measurement system is where the claim does its work. Claim 1 recites it as configured to perform four actions in sequence: obtain the locations of the donor dies sitting in the recesses; obtain the locations of the one or more targets; based at least on the obtained locations, provide output signals to adjust the die locations so they correspond to the obtained target locations; and, based on those adjusted locations, place the dies by relative movement between the first and second stages. The architectural consequence is that alignment happens once, in parallel, across every die in the recess field — and the transfer that follows is a single relative move of one stage against the other, not a die-by-die descent.

One drafting detail is worth flagging for anyone reading the record rather than the summary. The published abstract, which is descriptive text and not a statement of claim scope, does not include the target-measurement step. It moves from obtaining the donor-die locations directly to adjusting those dies "to correspond to locations of the one or more targets," without the "obtained" qualifier that appears in claim 1. The claim describes a system that measures both sides of the transfer and aligns measured-to-measured; the abstract as written describes only the die-side measurement. Where the two texts differ, the claims govern what is claimed.

How the alignment happens inside the pocket

The dependent claims fill in the mechanism, and they answer the obvious objection — if the die is parked in a fixed pocket, how do you correct its position? Claim 2 recites adjustable stages disposed within the recesses, supporting the dies, with the measurement system providing output signals to adjust die locations using those in-recess stages. Claim 3 routes those signals to one or more actuators. Claim 4 specifies piezoelectric activations as the actuation mode, which is the conventional choice where you need sub-micron correction with high stiffness and no backlash. Claim 5 recites that the adjustable stages are adjustable with respect to six degrees of freedom — meaning each pocket can correct translation in three axes plus rotation about all three, which is what you need if the die landed in the pocket with a residual yaw error.

Location-finding is claimed optically. Claim 6 recites an imaging device that obtains images of the dies in the recesses and a processor that determines die locations from those images. Claims 7 and 8 split by detector geometry. Claim 7 covers at least one photodetector obtaining two-dimensional images, with the processor detecting features including an edge of a donor die, a corner of the die, an edge of a recess, a corner of a recess, and the gap between a die edge and a recess edge. Claim 8 covers at least two photodetectors obtaining one-dimensional images and detecting edge and top features. The recess edge is doing real work in both: because the pocket is machined into a stage whose position is known, the die-to-recess gap is a directly usable error signal rather than something that has to be referenced back to a moving head.

Two further dependents address holding and release. Claim 12 recites electrostatic clamps disposed within the recesses, controllable by the measurement system, holding the dies in place and releasing at least in part as the two stages close. Claim 13 recites a dynamic stage interposed between each recess and each die, allowing translation and rotation during alignment. On the method side, claim 19 adds a voltage applied between the dies and the target locations, configured to cause alignment as the electrostatic clamping force is partially released — a self-alignment assist at the moment of contact, which is where residual placement error would otherwise be locked in.

Pipelining the metrology

The throughput argument only closes if the optical measurement of the die field does not itself become the new serial bottleneck. Claim 21, a dependent of the method claim, addresses that directly by splitting the recess field into sets:

wherein the first stage comprises multiple sets of recesses and wherein, when the first stage is in the first position, a first set of recesses is configured to accept a first plurality of donor dies and a second set of recesses is configured to allow optical measurement of the locations of a second plurality of donor dies— Claim 21, US20260206525A1

That is a pipeline: one set of pockets is being loaded while another set is being measured, so loading time and metrology time overlap instead of stacking. Claim 14 completes the picture by reciting a pick-and-place tool coupled to the measurement system and controlled by it to load dies into the recesses — the conventional pick head is still present, but demoted to a coarse loader whose placement accuracy no longer sets the accuracy of the final transfer.

Structurally, the application carries 21 numbered claims, of which claim 9 is canceled, leaving 20 operative. Claim 1 is the independent apparatus claim and claim 15 is the parallel independent method claim, tracking the same sequence in process terms: place dies into recesses, measure the die locations, measure the target locations, adjust, then place with the first stage moved to a second position relative to the second stage. The record classifies the application under CPC codes H10P 72/0446, H10P 72/0606, H10P 72/53 and H10P 72/7622 — Cooperative Patent Classification being the shared EPO/USPTO scheme used to sort filings by subject matter.

The filing sits inside a dense same-day cohort. Eight ASML Netherlands B.V. applications published on July 16, 2026, and several read as adjacent capability rather than unrelated work. US20260204466A1, on an actuation stage and electromagnet apparatus, is the closest mechanical sibling to the in-recess adjustable stages. US20260202761A1 (metrology based on multimode optical fiber imaging) and US20260202760A1 (a dispersion-engineered beam modifier for a metrology system) sit on the measurement side, while US20260202766A1 covers determining reticle deformations and US20260202767A1 covers lithographic apparatus thermal conditioning. Read together, the cluster is consistent with a company whose core competence is precision stages and optical position measurement applying both to the packaging step rather than the exposure step. None of this is settled. US20260206525A1 is a published application under examination, and claim scope can narrow before any grant issues. What the document establishes is the disclosed approach: treat die placement as a metrology problem solved once across a whole field of dies parked in pockets, rather than a motion problem solved repeatedly on a single head.