Boron oxide glass network low-permittivity underfill and inter-layer dielectric

The packaging industry is in the middle of a forced architectural transition: high-bandwidth memory stacks, glass-core substrates, and chiplet redistribution layers are pushing traditional polymer-based dielectrics past their limits. The problem is not merely dielectric constant — it is the combination of low permittivity, thermal reflow compatibility, and long-term stability in the presence of high-power thermal-interface materials. Polymer RDL dielectrics absorb moisture, creep under cyclic thermal loading, and often cannot survive the 300-400 °C reflows demanded by advanced flip-chip and glass-core assembly flows. Silica-rich glasses handle the temperature but carry permittivities well above 4.5. This invention occupies a narrow but commercially critical window: a boron-oxide glass-network underfill and inter-layer dielectric that achieves a dielectric constant as low as 3.8 at 1 MHz while remaining processable at temperatures the rest of the package already sees, making it the lowest-permittivity reflow-compatible inorganic dielectric identified across the surveyed material space. The timing matters because the HBM3 and HBM4 generations are actively driving co-packaged interconnect densities where parasitic capacitance in the redistribution layer directly limits signal bandwidth and memory bandwidth ceiling. Every 0.1 reduction in Dk in a dielectric layer that sits adjacent to thousands of RDL lines compounds across the entire memory bus. The market window for qualifying a new underfill or inter-layer dielectric into an HBM or glass-core package is three to five years ahead of volume production, which means the qualification clock is already running for the generation now being designed. This asset sits within the high-power thermal-interface materials portfolio as a composition-plus-device-use filing covering the B2O3-primary-network-former glass family — with and without SiO2, Al2O3, and ZnO modifiers — and optionally reinforced with wide-bandgap fluoride phases. It is a lead asset in its family, not a defensive backup, because no prior art in the surveyed patent landscape covers this composition in the package-integrated underfill or inter-metal-dielectric context adjacent to a thermal-interface material stack.

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.

The anchor composition of this family is approximately 90 mol% B2O3, a well-understood glass-network former whose open trigonal [BO3] network structure is fundamentally responsible for both its anomalously low dielectric constant and its relatively low glass-transition temperature. Boron oxide glass is one of the few inorganic materials that combines a high degree of structural disorder — preventing polarization pathways that would raise permittivity — with a softening point accessible to practical semiconductor assembly processes. The anchor composition yields a measured-equivalent dielectric constant of approximately 4.10 at 1 MHz and a glass-transition temperature of approximately 331 °C, placing it at the lowest-permittivity end of the reflow-compatible inorganic glass space surveyed. Modifier oxides — SiO2 (network stiffener, raises Tg), Al2O3 (aluminate bridging units, narrows hysteresis), and ZnO (intermediate oxide, controls thermal expansion) — allow the glass-forming window, viscosity profile, and coefficient of thermal expansion to be tuned without materially sacrificing the dielectric advantage, giving formulators latitude to target specific package CTE requirements. The optional fluoride reinforcement phases extend the family in a distinct direction: wide-bandgap inorganic fluoride salts (including alkali borofluoride and cryolite-class fluoride members) serve as secondary phases or matrix dopants that preserve the low-loss character while raising chemical durability and potentially widening the optical bandgap. Two fluoride anchors have been computationally characterized: an alkali borofluoride phase with a dielectric constant of 4.70 and a bandgap of 7.85 eV, and a heavier alkali hexafluorophosphate phase with a dielectric constant of 4.36 and a bandgap of 7.14 eV. These values are notable because wide-bandgap dielectrics suppress leakage and dielectric loss at elevated fields — a relevant consideration for RDL structures operating at high signal frequencies where loss tangent matters as much as Dk. Computational validation for the crystalline B2O3 reference structure was performed using density functional perturbation theory (DFPT) calculations for the dielectric tensor and phonon stability, with the low-k glass network database anchor confirming the permittivity estimate for the amorphous phase. The crystalline phase was assessed for phonon (dynamic) stability using a machine-learning interatomic potential from the MACE family, which reports a lowest phonon frequency of +1.174 THz — a positive value indicating the absence of imaginary modes and confirming that the structure sits in a genuine local energy minimum rather than a saddle point on the potential-energy surface. This is a single-potential assessment; the n_potentials count of one reflects that this is a well-characterized oxide composition for which MACE provides reliable coverage, and the DFT sources provide an independent cross-check on the dielectric constant specifically. The fluoride reinforcement phases were validated separately via DFPT calculations that established both the dielectric tensor and the bandgap, ensuring the wide-gap claim is computationally grounded rather than inferred from analogy. What makes this material family distinctive at a systems level is the combination of properties that normally trade off against each other: inorganic chemical identity (moisture resistance, no polymer creep, dimensional stability), reflow processability at temperatures already present in advanced packaging flows, sub-4.5 permittivity at device-relevant frequencies, and a glass-forming composition space wide enough to accommodate CTE and adhesion engineering without losing the dielectric advantage. The family covers the full underfill geometry — gap-filling material beneath a bumped die — as well as the inter-layer dielectric role in redistribution stacks, which makes it applicable both to HBM memory stacks (where underfill compatibility with the TIM stack is critical) and to glass-core packages where the inter-layer dielectric between RDL copper layers is a primary capacitance contributor.

The immediate commercial target is the inter-layer dielectric and underfill market segment within advanced semiconductor packaging — specifically the HBM redistribution layer and glass-core substrate segments where signal integrity at high data rates is now a primary design constraint. Industry analysts estimate the advanced packaging dielectric materials market (covering underfill, RDL dielectric, and inter-metal dielectric materials for chiplet and memory packages) in the range of one to three billion dollars annually, growing with the HBM and heterogeneous integration build-out driven by AI accelerator demand. This estimate should be understood as a rough order-of-magnitude for the addressable segment — the actual realizable market for a new material depends heavily on whether it can achieve qualification at one or more major HBM or glass-core package manufacturers, after which per-wafer material costs are relatively modest but the volume scales rapidly with production. The buying structure for this type of material follows a tiered path: chemical or glass material suppliers (Corning, AGC, Nippon Electric Glass, Ferro/Vibrantz for glass frits; Shin-Etsu, Henkel, Namics for underfill formulations) license or acquire novel glass compositions and formulate them into dispensable or printable underfill or dielectric pastes; these are then qualified by OSAT or IDM packaging engineers (Samsung, SK Hynix, Micron for HBM; Intel, TSMC advanced packaging lines; ASE, Amkor for glass-core). The royalty or licensing model would most naturally sit at the glass composition or formulated material level — either an exclusive license to a material supplier for a defined application space (HBM RDL underfill, glass-core ILD) or an acquisition of the patent family as a blocking position to control entry into this composition space. Given the narrow but well-defined dielectric performance window, a composition-plus-device-use claim structure creates meaningful leverage over any glass-based low-k underfill that enters the market in this Dk range with this Tg range. The timing of market relevance is governed by the HBM qualification cycle. HBM4 and co-packaged optics packages being designed now will enter volume production in the 2027-2029 window, and material qualification typically requires two to three years of reliability testing ahead of production release. This means a supplier or licensee that begins coupon-level characterization in 2026 is on the right schedule to contribute to that generation's material set. The absence of a qualified inorganic low-k underfill in the current supplier ecosystem — where the default remains polymer-based materials that carry moisture sensitivity and creep liabilities — creates a genuine substitution opportunity rather than a displacement of an entrenched incumbent at the same performance tier.

wide-gap low-loss RDL underfill/IMD with the lowest-eps (4.10) reflow-compatible glass surveyed

package-integrated underfill/IMD adjacent to TIM; particulate/redistribution-layer context distinct from wafer-scale 2D-transistor fluoride gate-dielectric

The most natural acquirers or licensees are advanced packaging material suppliers seeking to differentiate their underfill or ILD product lines for HBM and glass-core customers. Glass and specialty ceramic companies with existing packaging-material divisions — Corning (OLED and display glass, with growing packaging exposure), AGC (glass substrates for advanced packaging), and specialty frit suppliers (Ferro/Vibrantz, Heraeus) — have the glass formulation expertise to take this composition from computational anchor to formulated, dispensable underfill paste, and they have existing customer relationships with the HBM and glass-core packaging houses. Electronic materials companies with underfill portfolios (Henkel, Namics/Showa Denko, Shin-Etsu Chemical) are a second tier: they formulate rather than manufacture glass, but they have deep customer qualification relationships and could in-license the composition family to develop a new inorganic-glass underfill product line that their polymer-only portfolios currently lack. On the semiconductor company side, IDMs and memory manufacturers with in-house packaging R&D (Samsung Electro-Mechanics, SK Hynix, Micron) are potential direct licensees if they are pursuing proprietary underfill or ILD materials for next-generation HBM stacks — licensing the composition family would give them exclusivity in a space where no competitor has yet planted a flag. Intel's advanced packaging group (pursuing glass-core at scale) and TSMC's CoWoS and SoIC advanced packaging lines represent additional strategic fits. The asset is also relevant to fabless companies that specify packaging materials in their supply agreements and want to influence the material roadmap for the packages their AI accelerators will occupy.

The primary technical risk is the gap between computational prediction and measured coupon performance. The dielectric constant values of 3.8-4.5 are computationally anchored but not yet experimentally confirmed on a fabricated glass sample in a package-representative geometry. B2O3 glass is hygroscopic — it absorbs moisture from ambient air — and moisture uptake raises both the dielectric constant and the loss tangent, which could narrow or eliminate the advantage over incumbent polymer dielectrics in humid operating environments unless the glass composition is modified with moisture-resistant modifier oxides or fluoride phases, or the application environment is controlled. This is a known challenge with boron-rich glasses and must be addressed explicitly in the coupon characterization program. A second risk is processing: boron oxide glass has a tendency to volatilize boron trioxide at elevated temperatures, which can contaminate adjacent package structures and create compositional drift during reflow. Demonstrating stable, void-free underfill deposition through a 300-400 °C reflow profile without B2O3 volatilization or crystallization-induced cracking is a non-trivial process engineering challenge that the current computational work has not addressed. The roadmap to de-risk these concerns is well-defined: the open validation gate — a measured Dk/loss plus reflow-compatibility coupon test — is a standard materials qualification protocol that any glass supplier with packaging experience can execute. Modifier oxide selection (higher ZnO or Al2O3 content to reduce hygroscopicity, SiO2 to stabilize against volatilization) is an established lever in glass science and does not require new chemistry. The computational foundation provides the composition targeting to enter that experimental program at the right starting point rather than searching blind across glass formulation space, which is the core value of the existing work. Patent risk from the fluoride-gate-dielectric prior art is managed by the application-context distinction already embedded in the prosecution strategy.