Deep Submicron CMOS Scaling 1990s Integrate the Self Aligned Salicide Process
# Deep Submicron CMOS Scaling (1990s): Step 8 — Integrate the Self-Aligned Salicide (SALICIDE) Process
## 1. Siliciding the Gate and the Diffusions Without Extra Masks
Step 7 established why both the polysilicon gate and the source/drain diffusions need a silicide layer to control sheet resistance. Doing this with separate masked silicidation steps for the gate and for each diffusion region would add lithography steps the deep submicron die's already-growing mask count (a cost the `bicmos_logic_1980s` series' step 5 already flagged as a general integration tax) could not comfortably absorb. The self-aligned silicide — salicide — process instead forms silicide simultaneously on the gate and on the source/drain regions using the same blanket metal deposition and anneal, with no additional patterning mask required for either.
## 2. How Self-Alignment Works Here
After the sidewall spacers from Step 4 are in place (themselves already serving the LDD implant), a blanket layer of titanium or cobalt is deposited over the entire wafer surface. A first anneal reacts the metal with any exposed silicon or polysilicon it directly contacts — the gate top and the exposed diffusion surfaces — while the metal sitting on top of the sidewall spacer's dielectric and on top of the field-oxide or trench-isolation regions does not react, because there is no silicon beneath it to react with. A selective wet etch then strips the unreacted metal everywhere it did not react, leaving silicide only where silicon was exposed — the gate top and the diffusion regions — and leaving the spacer and isolation surfaces bare. A second, higher-temperature anneal converts the silicide to its final low-resistance phase.
## 3. Why This Is the Same Self-Alignment Philosophy as the LDD Spacer
Just as Step 4's sidewall spacer used the gate's own geometry to define implant boundaries without a separate lithography mask, the salicide process uses the existing pattern of exposed silicon versus dielectric — already defined by the gate, spacer, and isolation structures from earlier process steps — to define where silicide forms, without a separate silicide mask. This is a recurring 1990s deep submicron theme: as mask count and alignment tolerance both become scarcer resources, process engineers increasingly favor structures that use an already-patterned feature to self-define a later step, rather than adding a new masked step for each new material or dopant that needs placing.
## 4. The Net Result at the Transistor Level
A finished salicided device has a low-resistance silicide cap on both the polysilicon gate (reducing gate-line RC delay, a concern that compounds with the interconnect RC issues Steps 10 and 11 of this series take up) and on the source/drain diffusions (reducing series resistance and improving the contact that subsequent tungsten-plug interconnect, covered in Step 9, lands on) — achieved with the same mask count the device needed before silicidation was added at all.