Cobalt Contact Fill Process
Cobalt took over the contact plug at the 10nm/7nm-class logic nodes because tungsten's own resistivity stops behaving like a bulk metal once the hole it fills shrinks below about 20 nanometers, and that single physical fact, not any dislike of tungsten's chemistry, is the entire reason this module exists. Bulk tungsten resistivity sits around 5.6 microhm-centimeters, a perfectly respectable number, but inside a contact that narrow the electrons spend as much time scattering off grain boundaries and the hole's own sidewalls as they do traveling freely, and that surface/grain-boundary scattering effect pushes the effective resistivity of a sub-20nm tungsten plug well above its bulk value, into the 40 to 80 microhm-centimeter range depending on diameter and grain size. Cobalt's bulk resistivity is actually higher than tungsten's, around 6.2 microhm-centimeters, so on paper it looks like the wrong direction to move, but cobalt's electron mean free path is shorter than tungsten's to begin with, which means the same size-effect penalty that cripples tungsten at small dimensions barely touches cobalt until the hole gets much narrower still, and that crossover is what makes cobalt the practical winner in exactly the diameter range advanced contacts now live in. Intel's 10nm process and TSMC's 7nm-class nodes both moved local interconnect and contact fill toward cobalt publicly for this reason, and the same scaling argument is why cobalt can sometimes be deposited with a dramatically thinner barrier than tungsten needs, or in some integration schemes closer to no separate barrier at all, recovering contact volume that would otherwise be lost to liner thickness as the hole keeps shrinking.
The reason this is a genuine crossover and not just a marginal improvement is that tungsten and cobalt respond to shrinking contact diameter in opposite directions relative to each other, even though cobalt's bulk resistivity is the worse number of the two. At large diameters, where surface and grain-boundary scattering barely matter, tungsten's lower bulk resistivity wins outright and there is no reason to switch metals at all. As the hole narrows, tungsten's effective resistivity climbs steeply because its longer electron mean free path means more of its electrons are scattering off boundaries they would otherwise have avoided in a wider plug, while cobalt's shorter mean free path means it was already absorbing most of that penalty at larger sizes and has comparatively little room left to get worse. Below the crossover diameter, cobalt's effective resistivity curve drops under tungsten's, and that crossover point, not any fixed resistivity number, is the actual engineering target that determines at which node and which contact tier a fab switches metals.
| Tungsten plug | Cobalt plug | Ruthenium (brief) | |
|---|---|---|---|
| Bulk resistivity | ~5.6 microhm-cm | ~6.2 microhm-cm | ~7.1 microhm-cm |
| Effective resistivity at small contact size | Rises sharply below ~20 nm, 40-80 microhm-cm range at advanced-node diameters | Stays comparatively flat into smaller diameters, overtakes W's advantage below the crossover size | Has its own, even shorter mean free path; outperforms both Co and W at the very smallest dimensions, covered as its own item |
| Barrier requirement | Full Ti liner plus TiN barrier, ~5-10 nm each, mandatory to block WF6 fluorine attack | Much thinner barrier, or liner-free direct deposition in some integration schemes | Can often be deposited with minimal or no separate barrier, even thinner than cobalt's needs |
| Fill method | CVD nucleation plus bulk WF6/H2 fill | CVD or ALD seed plus CVD bulk fill, or PVD deposition with reflow anneal | CVD or ALD, similar seed-then-fill logic |
| Adoption node/era | Established baseline since deep-submicron-era CMOS, still used in wider vias and legacy contacts | Public roadmap adoption around 10nm/7nm-class logic nodes (Intel, TSMC) | Explored for the most advanced nodes below where cobalt's own advantage starts to run out |
| Primary driver for switching | N/A, historical default | Resistivity-at-scale and barrier-volume recovery as contacts shrink past tungsten's scattering wall | Same physics pushed one step further, for the next size reduction after cobalt |
Cobalt contact fill is one point on a continuum, not an isolated fix: it is the direct successor to tungsten plug contact fill, inherits the exact same etch-liner-fill-CMP skeleton, and exists only because tungsten's resistivity size effect turned into a scaling wall that the previous module could not get past at advanced-node contact diameters. The liner-thinning and barrier-shrinking theme that makes cobalt work is the same theme that motivates ruthenium's adoption one size tier further down, where ruthenium's even shorter electron mean free path repeats the same crossover trick again at smaller dimensions than cobalt can reach, and it is the same theme running through the molybdenum contact and local-interconnect work explored as another candidate on this continuum, each metal picked for the same underlying reason: minimize the combination of bulk resistivity, size-effect scattering penalty, and barrier volume loss at whatever contact or line dimension a given node actually needs to fill. Tungsten, cobalt, ruthenium, and molybdenum are not competing replacements chasing a single winner; they are successive answers to the same scaling physics, each one valid for the diameter range where its own resistivity curve happens to sit below the others, and cobalt's place in that lineup is the first metal to make the crossover trade explicitly in production at the 10nm/7nm-class node.