Capped aluminum fluoride low-k dielectric nanolaminate for redistribution layers

Aluminum fluoride (AlF3) has a static permittivity near 5 and a bandgap of 7.6 eV — both numbers that make it genuinely attractive as a redistribution-layer dielectric for high-frequency packaging. At RF and microwave frequencies, dielectric loss scales with both permittivity and the imaginary part of the susceptibility, and the wide gap of AlF3 suppresses that loss in a way that polymer RDL dielectrics and SiCOH cannot match. The catch, and the invention, is that bare AlF3 cannot be deployed on copper without a cap. A calculated fluorine-vacancy migration barrier of 0.693 eV is low enough that fluorine diffusion into copper conductors would be a reliability failure in real package conditions. The invention is the capped nanolaminate: AlF3 paired with a fluoride cap layer to suppress vacancy migration and render the stack pinhole-tolerant and copper-compatible. The claim protects not just AlF3 but a genus of metal-fluoride low-k dielectrics — AlF3, MgF2, CaF2, BaF2, SrF2, and K2SiF6 — all validated computationally, all deployable in the same capped-nanolaminate architecture. This breadth is meaningful: competitors cannot simply substitute a different fluoride and walk around the protection. Two independent machine-learning interatomic potentials (MACE and CHGNet) both confirm the dynamic stability of the fluoride family, and density-functional perturbation theory puts the static permittivity at 4.76, consistent with the ~5 target. This is a natural fit for the glass-core advanced-packaging substrates portfolio, which pursues dielectric and integration innovations that enable next-generation packaging at the interface of chiplet architecture and RF systems. HBM and chiplet RDL are reaching frequencies where insertion loss through the inter-line dielectric is no longer negligible, and panel-scale adoption of a low-loss fluoride nanolaminate — if qualified — represents a meaningful materials substitution.

Each candidate is validated by multiple independent machine-learning interatomic potentials. A material advances only when the engines agree on phonon (dynamic) stability — disagreement is surfaced, not hidden.

AlF3 crystallizes in the R-3c space group. Its static permittivity, computed by density-functional perturbation theory using Quantum ESPRESSO, is 4.76 — consistent with the design target of approximately 5. Its bandgap of 7.6 eV is unusually wide for a candidate RDL dielectric, and that width is the origin of the low-loss advantage: at microwave and RF frequencies, dielectric loss is tied to the gap between the operating frequency and the nearest electronic transitions, so a 7.6 eV gap leaves an enormous margin above any packaging frequency of interest. Polymer RDL dielectrics typically have bandgaps below 4 eV and permittivities above 3 but with substantially higher loss tangents at GHz frequencies. SiCOH reaches permittivities around 2.5–3 but involves porous integration challenges and higher-frequency loss that does not scale cleanly with miniaturization. The critical reliability physics is the fluorine-vacancy migration barrier, computed using the nudged elastic band (NEB) method. The calculated barrier is 0.693 eV. That value is low enough that fluorine migration from uncapped AlF3 into adjacent copper would pose a real reliability risk under damp-heat conditions — which is the industry-standard accelerated-aging test for packaging. The invention addresses this directly: by depositing AlF3 as a nanolaminate with a capping layer, fluorine migration is physically blocked. Uncapped AlF3 on copper is excluded from the claims for exactly this reason, and that exclusion is substantive, not merely a legal formality. Dynamic stability of the fluoride family was assessed using two independent machine-learning interatomic potentials — MACE and CHGNet — applied as a consensus screen. Both potentials agree that the structures are dynamically stable, showing no imaginary phonon modes. This two-engine consensus is the platform's standard for advancing a material from candidate to validated; a single potential showing stability is treated as insufficient, and the agreement between MACE and CHGNet here provides meaningful confidence. The fluoride family validation extended this screen across AlF3, MgF2, CaF2, BaF2, SrF2, and K2SiF6, confirming that the genus is computationally coherent, not just the lead compound. The nanolaminate architecture is also functionally important beyond reliability. Nanolaminates suppress pinholes that would develop in a thick single-layer fluoride film deposited by ALD or CVD. Pinhole tolerance matters in RDL because even a single conductive defect through a thin inter-line dielectric causes a short. The alternating-layer structure distributes any deposition non-uniformity and prevents pinholes from propagating vertically through the stack.

The addressable market for advanced-package inter-line dielectrics spans HBM memory packaging, chiplet RDL integration, and RF/microwave package substrates. A reasonable estimate for the total serviceable segment is $1–5 billion, reflecting the dielectric materials and deposition steps for advanced packaging at high volume. This estimate should be treated as a directional range — actual capturable share depends on qualification timelines and adoption decisions by leading packaging houses. The economic argument for a fluoride nanolaminate is strongest in the RF and microwave segment, where reduced insertion loss has a quantifiable system-level value that can support pricing above commodity dielectric materials. The customers are HBM RDL integrators and RF package makers. HBM and chiplet packaging are driven by bandwidth-density economics: more signal lines in less space at higher frequencies. As RDL line pitches shrink below 2 microns and operating frequencies climb into the multi-GHz range, the loss tangent of the inter-line dielectric begins to matter to signal integrity budgets. A capped fluoride nanolaminate with permittivity near 5 and a wide-gap low-loss character addresses that need directly. RF and microwave package makers face an even sharper loss constraint, since their systems are specified against insertion-loss budgets at defined frequencies, and a lower-loss dielectric is a direct product differentiator. Licensing logic for a deposited film module naturally points to a per-wafer or per-panel running royalty applied to packages incorporating the capped nanolaminate process. Because the cap is required — and because that requirement is built into the claims — a licensee cannot use AlF3 as a bare dielectric and avoid the license; the only commercially viable implementation is the capped stack. That structural feature makes enforcement straightforward and the licensed unit unambiguous. Field-of-use licensing that separates RF and HBM/chiplet applications is also feasible and may maximize total capture if adoption rates differ significantly between segments.

wide-gap low-loss RDL dielectric at RF/microwave

package-integrated capped RDL; uncapped + simmonsite cryolite excluded

The primary acquirer or licensee profile is an RF and microwave package maker whose products are differentiated by insertion-loss performance. For that buyer, a capped fluoride nanolaminate with a 7.6 eV gap and permittivity near 5 addresses a real product constraint, and the economics of a per-panel running royalty are easily justified by the system-level value of reduced RF loss. A field-of-use exclusive license in the RF packaging segment would command a meaningful premium given the direct product-differentiation benefit. HBM and chiplet RDL integrators are the second buyer category, motivated by signal-integrity improvement in fine-pitch, high-bandwidth interconnects rather than RF loss per se. For these buyers, a non-exclusive license to the nanolaminate process is likely the appropriate structure, since the RDL dielectric market at HBM scale involves multiple competing integrators and exclusivity in that field may be difficult to command. Semiconductor substrate manufacturers and advanced-packaging foundries with deposition capability (ALD or CVD tools) are also natural licensees, since they can integrate the nanolaminate module into existing RDL process flows without requiring their customers to change chip designs.

The central technical risk is that the cap suppresses fluorine migration in modeled conditions but has not yet been measured. A damp-heat coupon test with ToF-SIMS is the defined next step, and until it is completed, the reliability case rests on the 0.693 eV NEB barrier and the physical plausibility of cap-layer blocking — not on measured data. If the cap does not suppress migration adequately in hardware, the entire approach would require either a different cap material or a substantially thicker stack, both of which affect integration economics. The permittivity is similarly unconfirmed by measurement on integrated films; the DFPT value of 4.76 is credible, but film stress, stoichiometry variations, and deposition conditions can shift the measured value. Process integration is a secondary but real risk. Fluoride dielectrics are not standard back-end-of-line materials, and introducing them into copper RDL flows requires demonstrating compatibility with CMP, wet cleans, and barrier layers that are optimized for oxide and SiCOH chemistry. Fluorine outgassing during subsequent thermal steps is a related concern. The roadmap to de-risking is straightforward in sequence — ToF-SIMS damp-heat coupon first, then ellipsometry and capacitance measurements on integrated test structures, then a full RDL process-compatibility screen — but each step requires a buyer with deposition infrastructure and packaging qualification capability, which is why the natural buyer is a packaging foundry or an RF package maker with in-house advanced deposition tools.