Fsg Process
The FSG process deposits fluorine-doped silicate glass — PECVD oxide with a few atomic percent of fluorine substituted into the silicon-oxide network — and it is the one dopant swap in this oxide family aimed at lowering dielectric constant rather than enabling reflow or gettering, making it BEOL's answer to the same trade-off the BPSG, PSG, and USG processes already solved for the pre-metal dielectric sitting closer to silicon.
The physics is straightforward substitution, not a new deposition mechanism. TEOS or silane PECVD oxide normally forms a dense Si-O-Si network with a dielectric constant around 4.0-4.2; adding SiF4 to the plasma chemistry lets fluorine atoms terminate some of those Si-O bonds in place of oxygen, and because fluorine is more electronegative than oxygen, the resulting Si-F bonds polarize less easily under an applied field, pulling the bulk dielectric constant down toward 3.5-3.7 at a practical 2-4 atomic-percent fluorine loading. That drop matters directly at the interconnect level: capacitance between adjacent metal lines scales with the dielectric constant of whatever sits between them, so an FSG interlevel dielectric cuts RC delay and crosstalk on the same Al or early-Cu wiring pitch without requiring any change to the metal itself.
| Oxide family | Dopant | Typical k | Primary benefit | Main risk |
|---|---|---|---|---|
| BPSG | Boron + phosphorus | ~4.0-4.2 | Low-temperature reflow, Na+ gettering | Boron out-diffusion into silicon |
| PSG | Phosphorus only | ~4.0-4.1 | Na+ gettering, no boron risk | Still somewhat hygroscopic |
| USG | None | ~4.0-4.2 | Chemically inert, safe against metal | No reflow, no gettering |
| FSG | Fluorine | ~3.5-3.7 | Lower k, less RC delay/crosstalk | HF formation above ~4-6 at% F |
| OSG / CDO | Carbon (methyl groups) | ~2.7-3.0 | Lower k than FSG, no corrosion risk | Lower mechanical strength, CMP-sensitive |
That production window is also exactly why FSG never chased the lowest achievable k the way its successor did. Pushing fluorine loading past roughly 4-6 atomic percent keeps lowering the dielectric constant on paper, but the film stops being thermally and chemically stable at that concentration — unincorporated or weakly bonded fluorine reacts with ambient moisture during subsequent anneal steps to form hydrofluoric acid, which then slowly attacks any exposed aluminum interconnect it contacts, a corrosion failure mode with nothing in common with BPSG's boron diffusion or PSG's gettering trade but cutting from the same cloth: every dopant buys a property improvement by accepting a new release-and-react risk at the margins of its process window. That ceiling is what eventually priced FSG out of leading-edge BEOL — once copper interconnect and sub-130nm pitches needed k well under 3.0 to keep RC delay in check, fabs moved to carbon-doped oxide instead of pushing fluorine content any further.
Read the FSG process as fluorine's answer to USG's inertness: USG accepts zero dopant benefit to guarantee zero dopant risk, while FSG accepts a real but bounded corrosion risk in exchange for a dielectric constant improvement that BPSG, PSG, and USG — built from the same base PECVD chemistry — have no path to delivering at all.