Metal fluoride low-k dielectric family for capped redistribution layer applications

The glass-core advanced-packaging substrates portfolio targets a possible shift in advanced packaging dielectrics: moving from polymer interlayer materials toward inorganic fluoride compounds proposed for their thermal stability and moisture resistance, used with a barrier cap in the copper redistribution-layer (RDL) metallization schemes of chiplet integration. This asset broadens that strategic position by specifying a family of metal fluoride dielectrics — MgF2, CaF2, BaF2, SrF2, and K2SiF6 — as candidate alternatives to the portfolio's lead compound, screened as dynamically stable by machine-learning potentials and proposed for use within capped RDL architectures. The timing of this filing reflects a forced-substitution dynamic already underway. As RDL pitch shrinks below 2 µm and signal speeds push into the millimeter-wave regime, the dielectric constant of conventional polymer low-k materials (typically k ~ 3.0–3.5) becomes a genuine bottleneck. Crystalline fluorides do not sit below that range — CaF2 and MgF2 have bulk dielectric constants of approximately 6.8 and 5.4 respectively at low frequencies, higher than polymer low-k; their case rests on other properties such as very low mid-IR absorption, and whether thin films fare better at RF is untested. More importantly, inorganic fluorides eliminate the moisture-absorption penalty that degrades polymer low-k performance under thermal cycling, a critical reliability argument for heterogeneous integration substrates. The fluoride family filing functions as a coverage-broadening arm within the portfolio. Rather than claiming a single champion material, it establishes that multiple crystallographically distinct fluoride structures — from the rutile-type (MgF2), the fluorite-type (CaF2, BaF2, SrF2), to the hexafluorosilicate (K2SiF6) — can serve the same device function when deployed with an appropriate barrier cap. This breadth is the core strategic value: it makes it substantially harder for a competitor to design around any single compound while staying within the fluoride low-k space.

Candidates are screened for imaginary phonon modes with the machine-learning interatomic potentials named above. This is a computed harmonic check, not a measurement; where engines disagree, that is shown.

The materials in this family span three crystal structure types, which is itself significant. MgF2 adopts the rutile (P4₂/mnm) structure — a tetragonal framework with octahedrally coordinated Mg²⁺ centers that provides strong Mg–F bonds and correspondingly high fluorine vacancy migration barriers (computed at 1.05 eV). CaF2 and its isostructural congeners BaF2 and SrF2 all adopt the fluorite (Fm-3m) cubic structure, characterized by eight-coordinate cations and a three-dimensional F⁻ sublattice. BaF2 shows an F-vacancy barrier of 1.13 eV despite its more open structure, likely reflecting the high polarizability of Ba²⁺ that stiffens the potential energy surface. CaF2, by contrast, shows a notably lower barrier of 0.27 eV — the lowest in the group — which directly informs the mandatory-cap classification for CaF2 in direct contact with Cu metallization. K2SiF6 is a hexafluorosilicate with a distinct layered octahedral anion (SiF6²⁻) and introduces compositional diversity into the family that could open separate claim space. Dynamic stability is assessed using two independent machine-learning interatomic potentials. Both potentials agree that the representative structures in this family are dynamically stable: the phonon density of states shows no imaginary (negative-frequency) modes, with the lowest-frequency acoustic branch bottoming out near 1.03 THz on one potential and 0.87 THz on the other. The convergence of two independent potential energy surfaces on the same stability verdict — using models trained on entirely different DFT datasets and with different architectural priors — is the key computational validation gate for advancing a material to the claims stage. Disagreement between potentials flags a metastable or force-field-artifact candidate and stops it from advancing; agreement here gives confidence that the stability is a genuine feature of the electronic structure, not a potential artifact. Two independent DFT reference calculations back up the machine-learning results. The F-migration ladder simulation is the workhorse analysis that differentiates the cap requirement across the family. For each compound, a fluorine vacancy is introduced into the relaxed supercell and the migration barrier is computed along the lowest-energy hopping path — effectively a nudged-elastic-band (NEB) calculation reduced to the key saddle point. The resulting barrier hierarchy (BaF2 1.13 eV ≈ MgF2 1.05 eV >> CaF2 0.27 eV) directly maps to the cap-required versus cap-recommended designation: CaF2 requires a diffusion barrier cap at all process conditions, while MgF2 and BaF2 are classified as cap-recommended given their higher barriers but are still not recommended uncapped on copper given the thermodynamic driving force for CuF2 formation at elevated temperature. This is a concrete, quantitative basis for the negative limitation ("uncapped on Cu excluded") that appears in the claims. The dielectric tensor (DFPT/linear-response calculation) provides the frequency-dependent dielectric response for each compound, distinguishing the electronic (optical) contribution from the ionic contribution. For fluorides, the ionic contribution is large and structure-sensitive, making first-principles DFPT the appropriate validation tool rather than empirical estimates. The simulations are reported to place all five materials as candidates for RDL interlayer dielectric use, though no computed values are shown here, the low-k label is not established against polymer dielectrics, and the exact thin-film effective dielectric constant depends on deposition conditions, crystallographic texture, and thickness — open variables that physical vapor deposition (PVD) or atomic layer deposition (ALD) process development would need to pin down.

The addressable market for RDL low-k dielectrics within advanced packaging substrates is a specialized but rapidly growing segment. Industry estimates place the advanced packaging substrate market broadly in the range of several billion dollars annually, but the specific low-k dielectric materials-supply and licensing opportunity within that is narrower. The context places the directly addressable market at $0.2–0.5 billion — this is a realistic estimate for the fluoride-specific lane, recognizing that polymer low-k incumbents hold most of the installed base today and that fluoride adoption will initially be confined to the highest-performance, highest-frequency applications where the dielectric constant difference is worth the process development investment. The customers for this technology are the companies running advanced RDL flows: integrated device manufacturers (IDMs) and outsourced semiconductor assembly and test (OSAT) providers building 2.5D and 3D heterogeneous integration packages. These are not materials buyers in the traditional sense — they are process engineers at companies like major foundries, advanced packaging specialists, and memory manufacturers integrating high-bandwidth memory stacks. The licensing logic therefore runs primarily through process technology licensing or materials-supply agreements with PVD/ALD tool companies that already have fluoride deposition capability (fluoride optics is a well-developed PVD application). A secondary licensing path runs through the substrate manufacturers themselves, who increasingly hold their own dielectric material intellectual property as a competitive differentiator. Royalty framing for a composition-and-device-use patent in this space typically tracks either a per-wafer or per-substrate royalty, or a percentage of the dielectric materials bill-of-materials. Given the relatively small material volume per substrate and the high value of the finished package, even a modest royalty rate on the dielectric step produces meaningful revenue per unit for the patent holder. The strategic value, however, may exceed the direct royalty income: if granted as drafted, broad fluoride-family coverage in RDL dielectrics could give a position that competitors developing a listed fluoride dielectric for this application would need to license or design around.

fluoride-family breadth for the low-k lane

capped package-integrated form

Preliminary screen only. This is not a freedom-to-operate opinion.

The most natural acquirers or licensees for this asset are companies with active advanced packaging substrate programs who need to differentiate on dielectric performance at the materials level. This includes major packaging OSATs and substrate manufacturers expanding into glass-core or glass-adjacent interposer products, where the thermal stability of inorganic fluoride dielectrics offers an argument for the substrate-level glass/fluoride integration story. ALD and PVD equipment companies with existing fluoride deposition platforms for optical applications are a second category: they hold the process know-how to translate these materials into manufacturable thin films and could use a composition-and-device-use patent position to offer a differentiated materials process to their semiconductor customers, either through licensing or through a materials-supply arrangement bundled with the equipment. Integrated device manufacturers running their own advanced packaging lines — particularly those developing chiplet architectures at high frequency where the polymer low-k dielectric constant becomes a signal integrity constraint — represent the end-use buyer category most motivated by the technical merits rather than the IP position alone. For such buyers, this asset is valuable primarily as freedom to operate within the fluoride low-k design space and as a defensive position that competitors cannot use to block their own process development. The breadth of the claim family, covering five core compositions plus mixed fluoride extensions, is particularly valuable to a buyer who intends to run a multi-compound process development program and does not want to take the risk of optimizing one fluoride only to find it blocked while the adjacent compositions remain available.

The primary technical risk is the gap between bulk-computed fluorine vacancy migration barriers and real thin-film interface behavior. A CaF2 barrier of 0.27 eV is low enough that grain boundary diffusion paths — which are typically lower than bulk paths — could result in meaningful fluorine transport to the copper interface even with a barrier cap, particularly during high-temperature solder reflow steps (above 250°C) used in package assembly. If the physical coupon experiment reveals that even capped CaF2 shows unacceptable copper fluoride formation, that compound would need to be removed from the practiced embodiments, reducing claim coverage. The higher-barrier compounds (MgF2, BaF2) are less exposed to this risk given their computed barriers, but the same thin-film caveat applies. The commercial risk is adoption pace: fluoride deposition is not a standard semiconductor fab process, and the process development investment required to qualify a new dielectric class is substantial. The $0.2–0.5B addressable market estimate is realistic but assumes meaningful fluoride adoption in RDL flows within a credible window; if the industry settles on a different low-k approach (air gaps, ultra-low-k organosilicates) the addressable market contracts. The roadmap to de-risk both dimensions runs through the F-migration coupon experiment as the immediate gate, followed by thin-film dielectric constant measurement on representative substrates, and ultimately through a partnership with an equipment or materials company that already has production-scale fluoride deposition infrastructure.