Tungsten Silicide Process
The tungsten silicide process builds WSi2 as a "polycide" gate strap — deposited directly on top of polysilicon as part of the gate stack itself, patterned together with the poly during the gate etch — rather than as a self-aligned contact silicide formed afterward. That architectural difference from the titanium, cobalt, and nickel silicide processes covered elsewhere isn't a minor detail; it's the reason WSi2 still runs DRAM word lines today, long after logic moved on to salicide and then to metal gates entirely.
Polycide and salicide solve the same basic problem — polysilicon alone is too resistive for a fast gate or word line — by attaching the silicide at completely different points in the process, and that timing difference is the whole story. A salicide gate (the TiSi2/CoSi2/NiSi family) forms after the gate is already etched and spacers are already in place, reacting metal with exposed silicon wherever the earlier steps left it exposed — the self-aligned mechanism the nickel silicide process covers. A polycide gate instead gets its WSi2 strap deposited as a blanket film directly on the polysilicon before any gate patterning happens at all, and the poly and WSi2 are etched together as a single combined stack in the same lithography step. Polycide isn't self-aligned to anything — it's lithographically patterned, the same way the gate material itself always was; it just happens to be two different conductive films patterned as one.
That early timing is exactly why WSi2 polycide survives DRAM's manufacturing flow when a salicide gate wouldn't. DRAM needs a storage capacitor built after the transistor's gate and word line are already formed, and forming that capacitor — particularly a high-k or stacked capacitor structure — often demands high-temperature anneal steps considerably hotter than anything a logic process runs after its gate is complete. A salicide like NiSi, as the nickel process covers, is specifically vulnerable to exactly that kind of continued thermal exposure — it's a metastable phase that degrades into a worse one if pushed too hot for too long, and CoSi2 and TiSi2 carry their own high-temperature agglomeration risks. WSi2 polycide, by contrast, is thermally far more stable once formed, tolerating the DRAM capacitor module's anneal budget without the phase degradation or agglomeration that would compromise a salicide gate exposed to the same downstream heat.
That thermal stability is also why logic abandoned polycide for salicide once gate resistance alone stopped being the limiting concern. Logic processes don't need a high-temperature module after the gate forms the way DRAM's capacitor does, so logic was free to adopt salicide once self-alignment's lower contact resistance and simpler integration became the more valuable property — eventually moving further still, to the replacement metal gate and HKMG stacks covered in the gate-last process, which eliminate the polysilicon resistance problem altogether rather than strapping it with a silicide at all. DRAM's word line, still built around a polysilicon-plus-silicide gate rather than a metal gate, kept the older polycide architecture specifically because its process flow still has that later high-temperature step that would otherwise undo a more modern — but more thermally fragile — alternative.
Read the tungsten silicide process as proof that "which silicide and when" is as much a scheduling decision as a materials one: polycide attaches its low-resistance strap early, lithographically, and tolerates whatever heat comes afterward; salicide attaches its strap late, self-aligned, and generally can't tolerate much heat afterward at all — and DRAM still runs on the first architecture specifically because its own process flow, unlike logic's, keeps demanding serious heat long after the gate is already built.