Tungsten boride copper diffusion barrier on alumina-borate liner for glass-core vias

The core proposition is a bilayer barrier architecture for high-aspect-ratio through-glass vias: an ordered alumina-borate (AlBO3) chemical-adhesion liner beneath a refractory tungsten boride (B2W / WBx) copper-diffusion barrier. The combination is modelled to reach the package copper-diffusion performance threshold at film thickness below that of tantalum nitride, the current process workhorse; this has not been measured. In a technology where via geometry is a hard constraint on bandwidth density and fill yield, that thickness delta is directly monetizable. Glass-core interposers and high-bandwidth memory stacks are entering volume manufacturing at exactly the moment the industry is confronting via-scaling limits. Thinner barriers mean more copper cross-section per via, lower resistance, better fill at high aspect ratio, and greater flexibility in metal-layer pitch. The glass-core advanced-packaging substrates portfolio positions this barrier at the inflection point where the incumbent TaN/TiN chemistries cannot thin further without sacrificing diffusion integrity.

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 barrier material, B2W / WBx (with x ranging approximately 1.5 to 2.5), crystallizes in the P6_3/mmc hexagonal close-packed structure, placing it firmly in the family of refractory transition-metal borides known for high melting points, chemical inertness, and low bulk diffusivity. The critical performance figure is a copper migration barrier of 3.05 eV, computed by the nudged elastic band (NEB) method, which describes the energy penalty a copper atom must overcome to transit through the boride lattice. At 3.05 eV this barrier is high relative to standard diffusion barrier materials, and the modeling covers a spread of stoichiometries rather than a single idealized composition — a deliberate choice that reflects real deposition variability and strengthens the claim's practical scope. The AlBO3 liner underneath is not decorative: it provides chemical adhesion to the glass surface and defines a stable, ordered interface that the tungsten boride can nucleate on. Without it, a bare crystalline B2W-to-glass contact is thermodynamically unstable and decomposes. The liner-mediated stack is the performance-enabling configuration. Thermal stability was assessed through CHGNet constant-volume molecular dynamics runs spanning 873 to 1273 K, temperature ranges that bracket soldering and back-end-of-line thermal budgets relevant to package integration. The boride does not amorphize or phase-separate under these conditions in simulation, which is a necessary (though not yet sufficient) condition for a manufacturable barrier. Separately, a MACE-driven sweep across the tungsten-boride compositional space — covering B2W, WBx, BW, W2B, B12W, and boron-doped variants — was performed to map which stoichiometries carry meaningful copper-migration barriers, providing the compositional breadth the claims cover. Two independent DFT source calculations underpin the energy landscape. Dynamic stability of the B2W structure was assessed with two independent machine-learning interatomic potentials, MACE and CHGNet. Both independently confirm the structure is dynamically stable: phonon calculations with each potential show no imaginary (negative-frequency) modes, meaning the crystal sits in a true energy minimum and will not spontaneously distort or amorphize at rest. Consensus between two ML potentials trained on distinct datasets and architectures is materially stronger evidence than a single-engine result, and it is what the Lattice Graph validation protocol requires before a material advances. This consensus, backed by two DFT references, gives the thermal-stability picture genuine computational credibility.

The advanced semiconductor packaging market, covering OSATs, glass-core interposers, chiplet integration, and high-bandwidth memory substrates, is broadly estimated above $50 billion and growing as AI accelerator demand pulls chipmakers toward heterogeneous integration. The specific via-barrier layer — the addressable slice for this technology — is a low-single-digit-billion-dollar subset, primarily driven by the volume of glass-core substrates and fine-line redistribution layers that require a reliable copper-diffusion barrier at each via. We estimate the serviceable addressable market for via-barrier chemistries in glass-core and advanced-organic packaging at several billion dollars annually once the technology generation is in full production volume, though that estimate is appropriately contingent on glass-core yield ramp timelines. All figures are estimates; no committed volumes or supply agreements are reflected. The royalty and licensing logic is well-suited to a process module. Because the tungsten boride barrier slots into existing atomic-layer deposition, physical vapor deposition, high-power impulse magnetron sputtering, or CVD flows without redesigning the package architecture, it can be licensed as a process recipe and composition right rather than as a hardware product. The natural monetization structure is a per-wafer or per-panel running royalty tied to glass-core substrate output, capturing value in proportion to production volume. The per-via framing also works where via count is a known process parameter, since the thickness advantage scales directly with via density. Because the barrier is a horizontal module usable by every OSAT running glass-core lines, non-exclusive licensing likely maximizes total royalty capture across the customer base.

Cu-barrier at sub-TaN thickness frees via geometric budget

ordered glass/AlBO3/B2W config; bare crystalline contact excluded

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

The most direct buyers are OSATs running or developing glass-core interposer and advanced-packaging lines, where via-barrier thickness is a process constraint with measurable yield and density implications. An OSAT that can qualify a thinner barrier while holding copper-diffusion performance gains a geometric advantage over competitors using standard TaN without needing to redesign via architecture. The value proposition is concrete and quantifiable in their yield models. Barrier-ALD and deposition-tool suppliers are equally natural licensees, since they can incorporate the tungsten boride process recipe into multi-customer tool flows and collect royalties across every OSAT adopting the technology. This creates a licensing tier structure: a non-exclusive process license to tool and chemistry suppliers for broad volume, with the option of a time-limited or field-restricted exclusive to a lead OSAT willing to fund the SIMS validation and early process qualification. Memory manufacturers building on glass-core substrates for high-bandwidth memory stacks represent a secondary buyer category with strong alignment, since via resistance and fill uniformity at high aspect ratio are critical to their bandwidth and power targets.

The primary technical risk is the gap between the modeled 3.05 eV copper-migration barrier and measured diffusion performance in a deposited film. NEB calculations capture the ideal lattice; real ALD and PVD films have grain boundaries, interface roughness, and stoichiometry gradients that can create fast-diffusion pathways the bulk calculation does not see. The acknowledged spread in migration-barrier values across the WBx compositional range means no single number can be asserted as the guaranteed outcome, and the SIMS coupon experiment is the only path to resolving this uncertainty. Until that data exists, the copper-diffusion advantage remains a modeled claim rather than a measured one. The second risk is interface integration. The bare crystalline B2W contact with AlBO3 is known to decompose, which means the performance-enabling ordered configuration must be reliably achieved in manufacturing. This places a process-control requirement on the liner deposition step that does not exist with TaN. The qualification pathway to an OSAT barrier flow requires demonstrating that the ordered stack is reproducible, not merely that a single ideal sample performs well. Addressing both risks — SIMS validation and process-window characterization of the liner-mediated stack — is the logical scope of a funded development phase, and the cost of that work is low relative to the potential royalty value if the results confirm the model.