Silicide Formation Process

Every contact plug this batch has built so far, tungsten and cobalt alike, lands on a surface that is not actually silicon, and that surface is what makes the plug's resistance low instead of enormous: a thin self-aligned layer of metal silicide grown directly on the source/drain and gate before the plug ever shows up. Bare crystalline silicon forms a high, non-ohmic Schottky-like barrier against almost any metal pressed against it, so a tungsten or cobalt plug landing straight on exposed Si would carry a parasitic contact resistance easily in the thousands of ohms per contact, swamping everything downstream of it. Silicide fixes this by reacting a deposited metal with the silicon itself, consuming a controlled few nanometers of the top of the source/drain and gate and converting it into a metal-silicon compound whose Fermi-level pinning and work function give a genuinely low, ohmic-like contact to the device underneath, typically in the tens-of-ohms range per contact instead. The entire process is self-aligned, the "sal" in "salicide," meaning no separate lithography mask is needed to define where the silicide forms: metal is deposited everywhere, reacts only where it touches exposed silicon, and is stripped away everywhere else, so the silicide pattern lines itself up automatically with whatever active-area and gate pattern already exists on the wafer.

Gate + spacer complete
Source/drain implants activated, spacer oxide/nitride defines exposed Si vs. protected sidewall
Pre-clean
Dilute HF or plasma clean strips native oxide from exposed Si; any residual oxide blocks the metal-Si reaction entirely
Blanket metal deposition
Sputtered Ti, Co, or Ni (era-dependent) coats the entire wafer: exposed Si, spacer sidewalls, and field oxide alike
First RTA (metal-rich phase)
Ni2Si or C49-TiSi2 forms
Low-temperature anneal, ~300-500 C, reacts metal with Si only where directly in contact; spacer/oxide-covered metal stays unreacted
Selective wet strip
Wet etch (e.g. H2SO4/H2O2 or SC-1 class chemistry) removes all unreacted metal from oxide/nitride; reacted silicide is untouched. This is the self-aligned step itself.
Second RTA (low-resistance phase)
NiSi or C54-TiSi2 forms
Higher-temperature anneal converts metal-rich phase to the stable, low-sheet-resistance monosilicide phase
Silicided S/D + gate, ready for contact
Low-resistance ohmic landing surface now in place; contact etch and W or Co plug fill proceed on top of this layer

The two-anneal structure is not a quirk of process engineering, it is the entire selectivity mechanism, and skipping either RTA or doing them in the wrong order breaks the self-alignment that gives the process its name. The first, lower-temperature RTA is deliberately kept below the temperature at which the metal would react with the oxide or nitride spacer material, so only the Si-metal interface reacts while the spacer-covered and field-oxide-covered metal sits inert on top of material it cannot bond with; this is what makes the wet strip able to cleanly distinguish reacted silicide from unreacted metal, since the strip chemistry is chosen to dissolve the metal but leave the already-formed silicide and the underlying oxide/nitride alone. The second, higher-temperature RTA then only needs to act on material that already is silicide, converting a high-resistance, silicon-rich or metal-rich intermediate phase into the stable low-resistance monosilicide, with no further selectivity problem to solve because the unreacted metal that could have caused a short or bridging defect was already removed in the strip step before this anneal ever runs.

Silicide Formation: Process Cross-Section Sequence One continuous self-aligned flow: gate + spacer through silicided source/drain, ready for contact 1 Gate + spacer 2 Pre-clean 3 Metal deposition 4 First RTA 5 Selective strip 6 Second RTA 7 Ready for contact What each stage does poly/gate + spacer; S/D implants activated native oxide stripped; HF or plasma clean blanket Ti/Co/Ni coats entire wafer metal-rich phase forms, ~300-500°C, Si only unreacted metal removed; wet etch, self-aligned step low-R monosilicide forms; higher-temp conversion ohmic landing surface; W/Co plug lands next Si (source/drain or gate) Blanket deposited metal (Ti/Co/Ni) Metal-rich silicide (1st RTA) Monosilicide, low-R phase (2nd RTA) The two-RTA sandwich around the selective strip is the entire self-aligned mechanism: react low, strip clean, then convert Cross-sections are illustrative of the documented process flow, not a specific fab's measured dimensions
Why It Is Self-Aligned: Selective Reaction, Then Selective Strip No separate mask ever defines where silicide forms; the exposed-Si pattern defines it automatically Before strip: metal everywhere exposed Si (S/D) spacer field oxide reacts here: Si-metal silicide forms does not react: metal just sits on oxide no reaction on spacer sidewall After strip: silicide only Si (S/D) silicide stays field oxide: bare, unchanged spacer: bare, unchanged Unreacted metal (removed by strip) Reacted metal-rich silicide (first RTA) Final low-R silicide Selectivity comes from chemistry (metal-Si reaction) plus strip-etch contrast, not from a lithography mask: this is the "self" in self-aligned
TiSi2CoSi2NiSi
Sheet resistivity (low-R phase)~13-16 microhm-cm (C54 phase, wide lines)~17-20 microhm-cm~14-20 microhm-cm
Si consumed per nm of metalHigh, roughly 2.2 nm Si per nm TiModerate, roughly 3.6 nm Si per nm CoLower, roughly 1.8 nm Si per nm Ni; gentler on shallow junctions
Anneal temperatureHigh, C54 conversion needs ~700-900 CModerate, ~700-800 C for CoSi2Low, NiSi forms around 400-600 C, easing thermal budget
Narrow-line behaviorDegrades badly: C49-to-C54 phase transformation is nucleation-limited and stalls on narrow polysilicon lines, a well-documented industry problemMuch better narrow-line behavior than TiSi2, one reason it displaced TiGood at the line widths it was adopted for, but prone to NiSi2 agglomeration/piping if overheated
Thermal stability / failure modeC54 is stable once formed, but getting there reliably on narrow lines is the failure mode itselfAgglomeration risk at high temperature, generally more robust than TiSi2Lower thermal budget is an asset, but narrow process window risks NiSi2 formation and silicon piping defects
Adoption era/nodeDominant through roughly the 0.35-0.25 um generationTook over around 0.18-0.13 um as line-width degradation made TiSi2 unworkableIndustry-wide standard from roughly the 90nm node onward

The TiSi2-to-CoSi2-to-NiSi progression tracks the same scaling pressure as the contact-metal story one level up: each material was the right answer only until shrinking geometry exposed a new failure mode the previous one could not avoid. Titanium disilicide's problem was never its resistivity in isolation, which was respectable in the wide C54 phase, but the fact that forming C54 at all requires nucleating it from an intermediate C49 phase, and that nucleation becomes progressively less reliable as polysilicon line width shrinks, so narrow gates on advanced nodes were stuck with the high-resistance C49 phase never fully converting, a documented industry-wide scaling wall rather than a single fab's defect. Cobalt disilicide sidestepped that specific nucleation problem and extended usable linewidths further down, but it does not solve thermal budget or junction-depth pressure, and as junctions got shallower the silicon consumption of either Ti or Co silicide started eating dangerously close to the junction itself. Nickel monosilicide answered that generation of problems directly: it consumes less silicon per nanometer of metal, forms at a meaningfully lower anneal temperature, and does not carry a narrow-line nucleation penalty the way TiSi2 did, which is why it became the industry default from roughly the 90nm node onward even though its own failure mode, NiSi2 agglomeration and piping under excess thermal exposure, demands its own tight process control.

Silicide formation is the step underneath the entire contact stack this batch has been assembling, not a side detail: tungsten-plug-contact-fill-process and cobalt-contact-fill-process both describe metal filling a hole, but neither one, on its own, explains how that metal actually makes a low-resistance electrical connection to silicon, because a bare metal-to-Si interface simply does not give one. The self-aligned silicide layer built here is what turns the source/drain and gate surfaces into something a tungsten or cobalt plug can land on and get an ohmic contact from, and the plug-fill process that follows is, in a real sense, just the wiring that connects to a junction this silicide layer already made usable. Every resistivity number and barrier-volume tradeoff discussed for the tungsten and cobalt plugs assumes this silicide is already there; take it away and the plug stack, however well optimized, is contacting a Schottky barrier instead of an ohmic one, which is why silicide formation is best understood as the missing link beneath the whole contact chain rather than a separate, optional module.

Take silicide formation process further

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