Cowp Cap
CoWP capping and post-CMP clean sit at the same point in the copper interconnect flow — right after electroplated copper is polished flat — but they answer different questions: post-CMP clean simply removes what polishing left behind, while CoWP cap tries to replace the dielectric barrier itself with a metal that bonds selectively to exposed copper. After copper electroplating fills the damascene trenches and vias, chemical-mechanical planarization strips the overburden back to the dielectric surface, and every wafer then needs a wet clean to remove slurry abrasive particles, copper slurry residue, and the thin native copper oxide that forms the instant polished copper meets air or an oxidizing clean chemistry; that clean is mandatory on every line regardless of what cap is used next. The optional step that follows is the cap itself: either a standard dielectric barrier film such as silicon nitride (SiN) or carbon-doped silicon nitride (SiCN) deposited blanket across the whole wafer by PECVD, or a selective electroless cobalt-tungsten-phosphide (CoWP) layer that plates only onto the exposed copper lines, leaving the surrounding low-k dielectric untouched.
| CoWP selective cap | Dielectric barrier cap (SiN/SiCN) | Post-CMP clean only, no cap | |
|---|---|---|---|
| What it is | Electroless Co-W-P alloy, plated selectively onto exposed Cu | Blanket PECVD insulating film over the whole wafer | Wet clean removes slurry/oxide, next dielectric deposited directly on bare Cu |
| Typical thickness | ~2-10nm | ~20-50nm | n/a — no cap layer |
| Electromigration benefit | Large — IBM reported 5-50x EM lifetime improvement by eliminating the weak Cu/dielectric diffusion interface | Moderate — interface remains the dominant EM path, lifetime set by barrier adhesion quality | None — bare Cu/dielectric interface is the fastest EM path and also oxidizes/corrodes further |
| Leakage / TDDB risk | Real — selectivity loss causes sideways overplating that bridges adjacent lines, raising inter-line leakage and breakdown risk | Low — uniform insulating film, well-characterized leakage behavior | Unacceptable in production — unprotected Cu corrodes and diffuses into the dielectric |
| Process complexity/cost | High — added electroless bath, Pd activation step, tight process-window control to avoid overplating | Low — single additional PECVD deposition, mature and well-controlled | Lowest — but not a real option, some cap is always required |
| Adoption status | IBM research/early production ~2004-2008, essentially abandoned industry-wide afterward over yield and reliability concerns | Mainstream industry default today across foundries and IDMs | Never shipped alone — included only as the non-viable baseline for comparison |
IBM did the original production-scale CoWP work, reporting electromigration lifetime improvements commonly cited in the 5x to 50x range over a standard dielectric-capped copper line, because replacing the weak copper/dielectric interface with a bonded metal-metal interface removes the single fastest diffusion path for electromigration in a damascene copper line. The catch is that electroless CoWP plating is only as good as its selectivity: the process uses a palladium activator to seed cobalt-tungsten-phosphide growth on exposed copper while nominally leaving the surrounding low-k dielectric bare, but any loss of that selectivity lets the plating grow sideways over the dielectric and bridge adjacent lines, and as interconnect pitch shrank through the 2000s that overplating margin shrank with it, producing inter-line leakage and time-dependent dielectric breakdown (TDDB) failures that showed up in yield long before they showed up as an electromigration win. Industry response was to retreat to the safer, lower-complexity alternative: a blanket dielectric barrier cap of SiN or SiCN deposited by PECVD across the whole wafer, which never offers CoWP's dramatic electromigration lifetime gain but is far more mechanically and electrically predictable, cheaper to run at volume, and does not carry any risk of shorting adjacent lines together — which is why SiN/SiCN barrier caps, not CoWP, became and remain the mainstream choice across foundries and IDMs, with CoWP surviving mainly as a case study in a reliability fix that solved one failure mode by reintroducing a worse one.
Read against the rest of this BEOL chain, CoWP cap and post-CMP clean sit directly downstream of copper electroplating and directly upstream of the next intermetal dielectric: electroplating fills the trench with copper, CMP and its mandatory post-clean leave a flat, residue-free copper surface, and the cap decision — CoWP's stronger but riskier metal bond versus the dielectric barrier's safer, lower-performance insulating film — determines how that copper surface behaves for the rest of the interconnect's life. That cap sits in the same low-k-and-barrier-engineering lineage already built out in fsg-process, osg-process, and airgap-process — each of those pushed the surrounding dielectric's own properties (fluorine doping, carbon doping, then removing the dielectric altogether) to fight capacitance and leakage, while CoWP instead tried to fix the interconnect's weakest reliability point by changing what sits directly on top of the copper, a complementary lever on the same BEOL reliability-versus-performance tradeoff rather than a competing one.