Tungsten Process
The tungsten process fills contacts and vias with CVD tungsten, reduced from tungsten hexafluoride (WF6) gas, and it exists because tungsten can do something aluminum and copper both struggle with: deposit conformally, bottom-up, into extremely high-aspect-ratio holes rather than just coating their tops. That gap-fill advantage comes bundled with a genuine hazard — the same WF6 that deposits tungsten so well is also aggressive enough to attack the silicon and barrier materials it's supposed to be landing on.
WF6 is reactive enough to deposit tungsten directly onto exposed silicon without any help, and that reaction is exactly the hazard the process has to control, not exploit at scale. Tungsten hexafluoride reduces spontaneously on bare silicon, consuming the silicon surface in the process — a reaction fast enough to actually damage a shallow contact if it's allowed to run uncontrolled, in the same spirit as aluminum spiking at a junction, but driven by fluorine attack rather than thermal dissolution. Left unmanaged, that same fluorine chemistry can also attack a thin or imperfect TiN barrier layer, punching tiny "wormhole" defects through it and reaching the silicon or dielectric underneath in spots where the barrier should have stopped it entirely.
Switching the reducing agent from silane to hydrogen between the nucleation step and the bulk fill is the whole trick, and it trades reaction aggressiveness for growth rate exactly when each is needed. WF6 reduced by SiH4 deposits a thin, dense, self-limiting tungsten seed layer at a controlled, relatively gentle rate, fully sealing the barrier before anything more aggressive gets near it. Once that seed layer fully covers the surface, the recipe switches to WF6 reduced by hydrogen — a much faster bulk deposition chemistry, but one that's now reacting against a continuous tungsten surface instead of against the barrier or silicon directly, so its greater reactivity has nothing vulnerable left to attack.
Tungsten's real payoff is bottom-up, aspect-ratio-blind gap fill, which neither aluminum nor copper delivers as reliably in the smallest, deepest contacts. CVD tungsten growth is close to conformal and chemically driven rather than line-of-sight like PVD aluminum, so it fills a narrow, deep contact hole completely rather than pinching off at the opening the way a sputtered or even some flowable films risk doing. That reliability is why tungsten plugs remain the standard contact and (at many nodes) local via fill material even in otherwise all-copper BEOL stacks — the two metals aren't competing for the same job, they're assigned to the geometries each handles best.
Removing the blanket tungsten overburden afterward is its own process decision, not an afterthought. Early tungsten processes etched the blanket film back with a fluorine-based plasma etch, selective to the underlying oxide; most modern flows instead polish it back with tungsten CMP, the same chemical-mechanical planarization discipline used elsewhere in the fab, chosen because CMP gives a flatter, more uniform result across a wafer with many plugs of varying density than a plasma etch-back reliably achieves.
Read the tungsten process as a case where the deposition chemistry's greatest strength and its greatest hazard come from the same molecule: WF6's reactivity is exactly what lets tungsten fill a deep, narrow contact so reliably, and exactly what would compromise the barrier underneath it if the recipe didn't first spend a gentler, self-limiting step sealing that barrier before unleashing the faster chemistry on top of it.