What does a hybrid-bonding patent claim, and where does it sit in the classification scheme? Hybrid bonding is the advanced-packaging technique that joins two dies or wafers by forming two kinds of bond across one flat interface simultaneously: the exposed copper pads on each side bond directly metal-to-metal, and the dielectric material surrounding those pads bonds directly dielectric-to-dielectric — all without solder bumps or adhesive. Eliminating the solder bump is what lets hybrid bonding shrink the connection pitch far below conventional microbump packaging, which is why it underpins the densest 3D stacking and chip-on-wafer integration. The patents claim the sequence and structure that produce those two bonds. In the Cooperative Patent Classification (CPC) scheme, the method is classified at H01L 24/80, which the USPTO titles "Methods for connecting semiconductor or other solid state bodies using means for bonding being attached to, or being formed on, the surface to be connected." The class is method-oriented, matching the fact that the inventive content is usually in how the bond is formed.

A 2026 grant shows the claim structure. Taiwan Semiconductor Manufacturing Company Limited holds US Patent 12,575,454 (issued March 10, 2026), titled "Bonding structures for semiconductor devices and methods of forming the same." Its independent claim 1 recites the two-anneal method:

...performing a first annealing process that forms a direct metal-to-metal bond between the first electrical bonding structure and the second electrical bonding structure; and performing a second annealing process that forms a direct dielectric-to-dielectric bond between the first dielectric material and the second dielectric material, wherein the second annealing process causes deformation of one or both of the first dielectric material and the second dielectric material thereby generating a deformed region of dielectric material that partially fills a void...— US 12,575,454 B2, claim 1, source

The limitations that carry scope are specific and ordered. First, "a first annealing process that forms a direct metal-to-metal bond" — the copper-to-copper join is formed in its own thermal step. Second, "a second annealing process that forms a direct dielectric-to-dielectric bond" — the insulator-to-insulator join is a separate, later step. Third, the second anneal "causes deformation" of the dielectric "thereby generating a deformed region of dielectric material that partially fills a void." That void-filling limitation is the crux: hybrid bonding is sensitive to gaps and voids at the interface, and this claim recites using the dielectric's own thermal deformation to close them. A method that forms both bonds in a single undifferentiated anneal, or that does not recite the dielectric deforming to fill a void, would sit outside this independent claim. For claim construction, the two-step thermal sequence plus the void-filling deformation is what distinguishes this claim from a generic "bond the dies" recitation.

Where the dependent claims set the numbers

The dependents quantify the materials physics that makes the void-filling work. Claim 2 recites that the dielectric is formed with a coefficient of thermal expansion "greater than" the metal's, so that on heating the dielectric expands more than the copper. Claim 3 puts numbers on it: the metal CTE "within a first range from 10 ppm/K to 20 ppm/K" and the dielectric CTE "within a second range from 40 ppm/K to 60 ppm/K." Claim 4 names candidate dielectrics — "polyimide, polybenzoxazole, polymethyl methacrylate, or benzocyclobutene." Claim 11 recites decreasing void size "by a difference that is greater than or equal to 1 nm," and the alternative independent claim 20 recites a deformed region "width in a range between 0 nm and 1 nm." These numeric limitations are where a freedom-to-operate analysis lives: a competing process that uses a different CTE relationship, or a different dielectric, or that does not achieve the recited void reduction, may fall outside the narrower dependents even if it performs hybrid bonding in the general sense.

The protruding-pad and gap limitations

A second set of dependents recites the surface geometry that makes the dual bond mechanically work. Claim 5 recites forming each electrical bonding structure so that it "comprises a first portion that protrudes from a first surface of the first dielectric material" — the copper pads stand slightly proud of the dielectric. Claim 6 then recites placing those protruding portions in contact "such that a gap is formed between the first surface of the first dielectric material and the second surface of the second dielectric material," and claim 7 recites "applying a pressure to compress" the components so the gap shrinks. Claim 9 recites closing that gap through a "first thermal expansion of the first dielectric material and the second dielectric material that is larger than a second thermal expansion of the first electrical bonding structure and the second electrical bonding structure." This is the physical mechanism behind the CTE numbers in claim 3: the dielectric is engineered to expand more than the copper on heating, so the initially proud copper pads make contact first and the surrounding dielectric then swells to close the gap and seal the interface. Reading these dependents shows that the claimed invention is a coordinated thermal-mechanical sequence, and a process that bonded flush surfaces, or that relied on copper rather than dielectric expansion to close gaps, would engage a different scope analysis.

How the class organizes the thicket

For mapping the hybrid-bonding landscape, H01L 24/80 is the method anchor, and it is densely populated — the same search surfaces large portfolios from foundry and packaging assignees, reflecting that bonding method is a contested area of advanced packaging IP. Because the class is defined by the act of connecting solid-state bodies by surface-formed bonding means, it captures both wafer-to-wafer and die-to-die hybrid bonding, and it sits alongside structure-oriented codes such as H01L 24/08 and stacking codes such as H01L 25/0657 that appear on the same grants. A portfolio reader uses H01L 24/80 to find the process claims and the structure codes to find the resulting interface claims; a single hybrid-bonding patent often carries both.

What the record shows is that hybrid bonding, at the claim level, is a controlled two-bond sequence. US 12,575,454 claim 1 recites a first anneal forming a direct metal-to-metal bond and a second anneal forming a direct dielectric-to-dielectric bond, with the dielectric deforming to fill interface voids — and the dependents fix the CTE ranges, the dielectric chemistries, and the void-reduction magnitudes that make the method work. CPC H01L 24/80 classifies the connecting method itself. The general notion of solderless die bonding is not what the claim protects; the ordered thermal steps and the void-filling deformation are. How far any such claim reaches against a particular bonding process depends on that claim's specific limitations and its file history.