Cobalt Process

The cobalt process covers two jobs that both got assigned to cobalt for the same underlying reason — it keeps working at dimensions where the materials it replaced started failing. Cobalt silicide (CoSi2) took over the source/drain and gate contact job from titanium silicide once lines got too narrow for TiSi2 to behave reliably, and plain cobalt metal later took over narrow contact and local-interconnect fill from tungsten for an analogous scaling reason: both times, cobalt won by surviving a dimension where its predecessor's resistance unexpectedly shot up.

Co sputter deposit
Blanket thin film over exposed Si
RTA silicidation
Co + Si react, form CoSi then CoSi2
Two scaling-driven roles
CoSi2 contact silicide
Replaces TiSi2 below ~0.25 µm
Co contact/local fill
Replaces W at the narrowest contacts
Selective wet strip
Removes unreacted Co, keeps silicide
Low-resistance contact
Holds up where predecessor didn't

Titanium silicide's failure mode was specific to narrow lines, not a general resistance problem, which is exactly why cobalt's advantage only showed up once geometry shrank far enough. TiSi2 forms in two crystal phases — a higher-resistance C49 phase first, which has to transform into the lower-resistance C54 phase during anneal to reach its target sheet resistance. That C49-to-C54 transformation depends on nucleation sites that become statistically scarce as line width narrows, so TiSi2's resistance climbs sharply — the "narrow line effect" — right at the dimensions advanced logic needed most. Cobalt silicide doesn't go through an equivalent high-resistance intermediate phase with the same narrow-line sensitivity, so CoSi2's sheet resistance stays essentially flat down to line widths where TiSi2 had already become impractical — which is why CoSi2 displaced TiSi2 industry-wide through roughly the 0.25 µm to 90 nm era.

Sheet Resistance vs. Line Width: Why CoSi2 Replaced TiSi2 TiSi2's narrow-line effect, and CoSi2's flat response at the same dimensions line width (decreasing →) sheet resistance TiSi2 — narrow line effect C49→C54 nucleation fails CoSi2 — stays flat dimension where TiSi2 broke down both silicides are deposited and reacted the same way — the difference is purely which crystal phase forms, and how reliably

Cobalt's second, later role — as a contact and local-interconnect fill metal displacing tungsten — comes from a volume argument rather than a phase-transformation argument. Tungsten contacts need a relatively thick Ti/TiN liner and barrier to control the WF6 fluorine-attack risk covered in the tungsten process, and that barrier occupies a meaningful fraction of an already-shrinking contact hole's cross-section, leaving less room for the low-resistance fill metal itself. Cobalt can be deposited with a much thinner liner — and in some flows nearly barrier-free directly on silicide — because cobalt's deposition chemistry doesn't carry tungsten's aggressive fluorine hazard, which means more of a shrinking contact's volume is occupied by actual conductor instead of barrier. At large contact dimensions that liner-thickness difference barely matters; once contacts scale down to a few tens of nanometers, the barrier can approach a double-digit percentage of the whole hole's area, and that's the regime where cobalt's thinner-barrier advantage turns into a real resistance win over tungsten.

The same cobalt chemistry also shows up capping copper lines for electromigration, a role the copper process covers in more depth — worth noting here only because it's the same element solving a third, unrelated problem. A thin cobalt layer selectively deposited onto exposed copper strengthens the copper-to-cap interface against the via-bottom electromigration failure mode that affects copper specifically; that use has nothing to do with either silicidation or contact fill, but it means a modern BEOL/MEOL stack can carry cobalt in three structurally different roles without it being the same material doing one unified job anywhere.

Read the cobalt process as two separate "it works at the old dimension, until it suddenly doesn't" failure stories, each solved by cobalt for a different underlying reason: titanium silicide failed because its best crystal phase stopped nucleating reliably in narrow lines, and tungsten contacts lost ground because their necessary barrier thickness stopped being a rounding error once the whole contact got small enough — cobalt happened to dodge both specific failure mechanisms, which is why it ended up filling two structurally different roles in the same shrinking transistor.

Take cobalt process further

Ask the copilot about this term, or have our engineers assess it against your process.