Tungsten Plug Contact Fill
Before copper ever reaches a chip, tungsten gets there first: a CVD tungsten plug, wrapped in a thin titanium/titanium-nitride liner, is what actually lands the contact on silicon or silicide and carries current up to the first copper metal level, because at the depth and diameter of a modern contact hole, tungsten's deposition chemistry and mechanical properties do a job copper cannot. The sequence starts the moment the contact or via opening has been etched down through the interlayer dielectric to a landing surface — silicide over a transistor's source/drain or gate, or a lower metal line for a via — and it runs through five tightly coupled steps before the wafer is ready for the next metal level: a titanium liner that reacts with the landing surface to form a low-resistance silicide, a titanium-nitride barrier layer on top of that, a thin tungsten nucleation film grown from WF6 chemistry, a bulk tungsten CVD fill that plugs the entire hole, and a chemical-mechanical polish or etch-back that strips the blanket tungsten overburden back to the dielectric surface. Each layer exists to solve a problem the one before it created or could not solve alone: the liner fixes contact resistance and adhesion, the barrier stops fluorine from reaching silicon, and the two-step nucleation-then-bulk-fill CVD sequence is what lets tungsten fill a hole many times deeper than it is wide without leaving a void at the center.
| Contact W-plug (FEOL/MOL) | Via W-plug (lower-level vias) | Cu damascene (BEOL metal lines) | |
|---|---|---|---|
| What it connects | Transistor source/drain silicide or gate to first metal level | One metal level to the next, in processes that still use W for lower vias | Metal line to metal line across most of the BEOL stack |
| Typical aspect ratio | Very high, often 5:1-10:1 at advanced nodes, narrow and deep | High, though usually shallower than a deep contact | Moderate, trenches are wider relative to their depth than a contact hole |
| Fill method | CVD nucleation + bulk WF6/H2 fill, grows inward from a conformal seed | Same CVD sequence as contact fill | Bottom-up electroplating onto a PVD or ALD seed layer, not CVD |
| Why this metal | Tungsten's melting point and grain structure resist electromigration at contact scale, and CVD conformality fills deep narrow holes that PVD or plating alone would void or pinch off | Same reasoning as contacts: reliability and fill conformality matter more than bulk resistivity this low in the stack | Copper's lower bulk resistivity (~1.7 µΩ·cm vs ~5.6 µΩ·cm for W) wins once lines are wide enough that resistance, not fill conformality, dominates performance |
| Key failure mode | Fluorine attack ("wormholing") into silicon if the Ti/TiN barrier is too thin or discontinuous | Same fluorine risk, plus seam voids if nucleation is incomplete | Electromigration and barrier/seed discontinuity rather than fluorine attack, since no WF6 chemistry is involved |
The fluorine chemistry that makes tungsten CVD possible is also the single biggest reliability risk in this entire module: WF6 is a powerful fluorinating agent, and if the TiN barrier over the Ti liner is too thin, discontinuous, or damaged, fluorine from the WF6 gas can punch through it and attack the silicon or silicide underneath, etching narrow filamentary channels — commonly called "wormholes" — directly into the substrate. This happens because the same WF6 + H2 reduction reaction that deposits tungsten on a properly nucleated surface will instead attack bare silicon directly if the fluorine reaches it before a continuous tungsten or TiN film has formed, consuming silicon and leaving behind voided, high-resistance, and sometimes electrically shorted contacts that can pass initial wafer test and fail later in the field as the damage propagates under thermal and electrical stress; this is why the nucleation step exists at all — a slower, more controlled tungsten seed deposition using silane or diborane reduction chemistry is used first, specifically because it is gentler on the underlying barrier than full WF6/H2 bulk-fill flux, and only after that seed layer is continuous does the process switch to the faster, more aggressive bulk fill chemistry. Barrier thickness, uniformity, and step coverage at the bottom corner of the contact — exactly the geometry most prone to thinning — are therefore treated as hard process-control limits, not tuning parameters, because a barrier that looks adequate across most of a wafer can still be locally thin enough at a handful of contacts to let wormholing start.
Tungsten plug formation is the hinge between the front end of the line and the copper back end of the line that sits directly above it: everything described here — contact/via etch, Ti liner silicidation, TiN barrier, WF6 nucleation and bulk fill, and W CMP — exists to land a transistor's source, drain, and gate reliably at the first metal level, a job tungsten does because its melting point and grain structure make it immune to electromigration at these tiny, high-current-density dimensions, and because CVD conformality can fill a hole far deeper than it is wide without the voids that would plague a sputtered or plated fill at that aspect ratio. Once the tungsten plug has been polished flush with the dielectric, the module this process feeds directly into takes over: via-etch-stop-dual-damascene-integration and post-etch-via-trench-ash-strip-clean open and clean the next level's geometry, pvd-barrier-seed-redeposition lines it with a Ta/TaN barrier and copper seed, and copper electroplating fills it bottom-up, with cobalt-tungsten or cobalt caps sealing each line before the next level starts — a completely different metal, deposited by a completely different mechanism, chosen because copper's lower bulk resistivity wins once a feature is wide enough that resistance matters more than fill conformality. The two metals coexist in the same chip, in the same interconnect stack, precisely because neither one is better everywhere: tungsten owns the narrow, deep, reliability-critical contacts and lower vias where CVD fill and electromigration resistance decide yield, and copper owns the wider BEOL metal lines where lower resistivity decides performance, and the handoff between them at M1 is the exact boundary this process and the Cu damascene chain above it were built to meet.