Self-aligned ruthenium cap on copper interconnects for electromigration lifetime extension

Advanced semiconductor packaging is undergoing a generational shift away from organic laminates toward glass-core substrates. Glass offers dramatically lower dielectric loss, superior dimensional stability, and the ability to etch tight-pitch through-vias — all of which are prerequisites for the chiplet-dense, high-bandwidth architectures that AI accelerators and next-generation RF front-ends demand. The problem is that copper, the conductor of choice in those through-glass vias and redistribution layers, degrades faster in this new configuration than it did in legacy silicon back-end-of-line (BEOL) stacks. Glass-core packaging strips away the SiN or CoWP capping layers that process engineers have relied on for electromigration management, leaving the top copper surface exposed to the dominant failure mode: surface and interface electromigration. When current densities climb — as they must at finer pitches — voids nucleate at that bare interface and grow until a line opens. The mean-time-to-failure penalty is severe and, without intervention, sets a hard reliability ceiling on glass-core roadmaps. This asset addresses that ceiling directly. A self-aligned metal cap — primarily ruthenium, with validated alternates — is deposited selectively on exposed copper in the through-glass via and redistribution-layer stack. Self-alignment means the cap forms only where copper is exposed, requiring no lithographic mask step and adding minimal process complexity. Ruthenium was chosen because it combines exceptionally high electromigration resistance with chemical stability under the bias conditions and wet-process chemistries encountered in advanced packaging fabs. The cap interrupts surface electromigration pathways and simultaneously suppresses copper corrosion under electrical bias, addressing two interconnected failure modes with a single deposition step. The novelty is not simply the cap material in isolation — ruthenium caps are known in planar BEOL — but the ordered, self-aligned placement within a glass-core through-via and redistribution-layer architecture, where the geometry, thermal environment, and chemical exposure differ materially from silicon BEOL. The timing of this filing is deliberate. Glass-core substrate adoption is at the qualification-and-ramp stage, with leading packaging houses and IDMs actively building pilot lines. Reliability standards for copper in this configuration are still being written — JESD61 coupon testing is the emerging evaluation vehicle, and the qualification data generated against that standard will become the reference point for the industry. A composition-plus-device-use claim filed now, and computationally screened, before that standard solidifies around incumbent CoWP-based solutions could, if granted and if the configuration is adopted, support licensing to glass-core copper interconnect lines that use it.

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.

Ruthenium crystallizes in the hexagonal close-packed structure (space group P6₃/mmc) and is one of only a handful of platinum-group metals that can be deposited by chemical vapor deposition or atomic layer deposition at temperatures compatible with back-end-of-line thermal budgets. Its combination of high melting point, low bulk resistivity relative to refractory alternatives, and inertness to standard copper wet-etch chemistries makes it a practical cap material without the resistivity penalty associated with tungsten or the adhesion issues associated with thin iridium films. In the self-aligned cap scheme, the ruthenium selectively nucleates on exposed copper surfaces during a CVD or electroless step; on dielectric surfaces (glass, SiO₂, or low-k) nucleation is suppressed by the surface-termination chemistry, producing a cap that is inherently registered to the copper feature with no mask. The primary computational validation uses two independent machine-learning interatomic potentials to evaluate the phonon spectrum of bulk hexagonal ruthenium. Both potentials — returning minimum phonon frequencies of 0.349 THz and 0.376 THz respectively — find no imaginary modes anywhere in the Brillouin zone, confirming that the structure is dynamically stable at the level of the interatomic force constants. Two independent DFT reference calculations corroborate this result. This level of consensus is the threshold the pipeline requires before a material is advanced: a structure where even one potential finds imaginary phonon modes is held back regardless of other metrics. For a well-characterized elemental metal like ruthenium this stability result is expected, but it establishes the computational provenance of the material entry and ensures that alloy and compound alternates in the family are evaluated against a validated reference baseline. Beyond bulk stability, the simulation suite addresses the specific failure physics of the application. The electromigration lifetime analysis uses Black's-equation fitting against the relevant current-density and temperature conditions for glass-core via dimensions. The cap geometry is modeled in the context of the full via stack, including the barrier layer (a Group A barrier metal such as TaN or TiN), because the barrier-cap-copper interface is the site where void nucleation either is arrested or allowed to propagate. A database of ten curated reference structures (the ten-entry calibration set) anchors the cap-metal surface-energy and adhesion-energy calculations, which in turn feed the surface-electromigration pathway analysis: by raising the surface diffusion barrier at the copper-cap interface, the ruthenium cap is modelled to shift the dominant diffusion pathway away from the surface toward the slower grain-boundary route, which would extend mean-time-to-failure under Black's law; no MTTF gain has been measured. The alternate cap compositions — NiMoP, CoWP, CoWB, RuCo, and Mn-silicate — are claimed as proposed fallbacks not yet evaluated on this page (only Ru phonon results are shown, and several are amorphous alloys where phonon screening does not apply), for process nodes where ruthenium deposition chemistry is not yet established or where cost is a constraint. NiMoP in particular is recorded in our screen as cross-engine stable on two potential evaluations, although only Ru phonon results are shown on this page. The Mn-silicate arm is compositionally distinct and extends protection to the copper-dielectric interface through a self-forming barrier mechanism, rounding out a family that covers surface, grain-boundary, and interface electromigration mitigation in a single claim family.

The addressable market is the advanced packaging substrate and redistribution-layer reliability segment, estimated at $0.5–2 billion across process licensing, materials supply, and equipment qualification. That range reflects genuine uncertainty about how quickly glass-core substrates displace organic laminates in high-volume production — the low end assumes glass-core remains confined to high-performance AI and RF applications through the end of the decade, while the high end incorporates mainstream adoption in automotive and datacenter compute packages. Both scenarios produce meaningful licensing opportunity because electromigration reliability is not optional: every copper via in a production package must meet the qualification lifetime, and a cap deposition step that enables that lifetime is a per-unit cost that every packaging customer must bear. The buyers of this technology are the packaging substrate manufacturers, OSATs (outsourced semiconductor assembly and test houses), and IDMs that operate BEOL and redistribution-layer process lines. These organizations currently rely on incumbent CoWP cap processes qualified on silicon BEOL, and they are actively searching for analogs validated specifically for glass-core geometry. Licensing takes two forms: a process license to a packaging house that wants to offer a ruthenium-cap RDL flow to its customers, or a materials supply agreement with the deposition chemistry provider (CVD precursor or electroless bath supplier) who bundles the IP into a qualified chemistry kit. Royalty logic could follow per-wafer or per-substrate fee structures seen in advanced packaging process IP; as an illustrative assumption, $0.10–$0.50 per substrate on volumes of hundreds of millions of units per year would give revenue in the tens to hundreds of millions of dollars annually, if claims are granted and adopted at scale. The reliability standard trajectory is an important commercial accelerant. As JESD61-based qualification becomes the industry norm for glass-core copper reliability, every new packaging line will need a documented cap solution that passes the coupon test. Being the reference solution when that standard is written — rather than a late entrant qualifying against an already-locked-in competitor — is a durable competitive advantage that is difficult to replicate even with substantial R&D investment after the fact.

EM + corrosion reliability in glass-core config

ordered placement in glass-core TGV/RDL with Group A barrier

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

The most direct acquirers or licensees are advanced packaging substrate manufacturers and OSATs that operate through-glass via and redistribution-layer process lines — organizations such as those building next-generation glass-core interposer capacity for AI accelerator and high-bandwidth-memory stacking applications. These buyers need a qualified copper reliability solution that is specific to the glass-core geometry, and a licensed process with computational provenance and a clear JESD61 qualification roadmap is a more attractive starting point than an internal development program starting from scratch. A second category of buyer is the deposition chemistry supplier — the company that sells electroless bath formulations or CVD precursors for copper cap metals — who would license the IP to bundle with a chemistry kit sold to packaging customers, effectively converting the IP into a recurring consumables revenue stream. IDMs with captive advanced packaging operations (particularly those developing glass-core interposers for AI and automotive radar applications) represent a third buyer category. These organizations are evaluating reliability solutions now, before their glass-core lines reach volume production, because the qualification timeline for a new cap process is typically 18–36 months. Acquiring or exclusively licensing the IP during that window, before the JESD61 coupon data is public, provides a meaningful head start over competitors who must license a non-exclusive position or develop an alternative. The family's breadth across six compositional alternates also gives an acquirer flexibility to match the cap chemistry to their specific deposition infrastructure without forfeiting IP coverage.

The central technical risk is that the JESD61 coupon validation gate remains open. The computational work is sound and the material choice is well-motivated, but packaging qualification teams operate on measured data, not simulations, and the lifetime improvement predicted by Black's-equation fitting must be confirmed on physical samples before any major process adoption decision is made. If the coupon data shows a smaller improvement than modeled — for example, because void nucleation at the cap-barrier interface is more significant than the surface-electromigration model assumes — the claim of lifetime extension would be weakened, though not eliminated, since even a modest improvement in a glass-core configuration currently lacking any cap is commercially meaningful. A secondary risk is the competitive response from CoWP incumbents who may accelerate their own glass-core qualification programs specifically to establish prior use or to design around the configuration claims. The mitigant here is the claim structure: the ordered self-aligned placement in a TGV/RDL with a Group A barrier is a specific combination that a CoWP process optimized for silicon BEOL cannot simply assert as equivalent without running its own qualification program, and that program itself generates the timeline advantage described above. The roadmap to de-risk is straightforward: fund JESD61 coupon fabrication and testing at a packaging-qualified fab, generate measured Black's-law parameters for the ruthenium cap configuration, and publish or present that data in a format that packaging qualification teams can use to initiate their own adoption evaluations.