Usg Process
The USG process deposits undoped silicate glass — plain CVD SiO2 with neither boron nor phosphorus added — and it completes the progression the BPSG and PSG processes started. BPSG trades in both dopants for reflow and gettering; PSG keeps phosphorus's gettering while dropping boron's diffusion risk; USG drops both dopants entirely, accepting zero reflow capability and zero gettering benefit in exchange for a film with nothing in it that can migrate, outgas, or absorb moisture at all.
PSG already solved the boron-penetration risk by removing boron; USG exists because even phosphorus's own properties are sometimes a problem, not just boron's diffusion. Phosphorus-doped glass, even without boron, is still somewhat hygroscopic and can still outgas trapped moisture under thermal stress — not as severely as a high-concentration BPSG film, but enough to matter directly against a sensitive interface. A liner or cap sitting in direct, permanent contact with an exposed metal line, for instance, risks having that outgassed moisture or residual phosphorus-related byproduct attack or contaminate the metal surface underneath it — a failure mode that has nothing to do with dopant diffusion into silicon and everything to do with what the oxide itself can release under heat.
USG's inertness is what makes it the default choice wherever neither BPSG's reflow nor PSG's gettering is actually needed, which turns out to be most of the oxide volume in a modern BEOL stack. The interlayer dielectrics covered in the TEOS and SACVD processes are, in their plain form, essentially USG — undoped oxide deposited for its conformality or flow-fill properties rather than for any dopant-driven benefit. Once a flow reaches the metal levels, where CMP (not thermal reflow) is doing the planarizing and sodium contamination from packaging is less of a concern than it is for the pre-metal dielectric sitting directly over active silicon, there's no reason to accept a dopant's hygroscopic or diffusion risk for a benefit the process doesn't need — so most BEOL oxide defaults to undoped.
Where USG specifically earns its place over PSG, rather than just being the generic choice, is anywhere a film sits in lasting direct contact with an exposed conductor rather than with silicon. A liner or etch-stop layer directly against an aluminum or tungsten surface has to survive subsequent thermal steps without releasing anything that could react with that metal — exactly the scenario where even PSG's modest hygroscopicity becomes a real risk rather than a tolerable one, and where USG's complete absence of dopant removes the hazard outright rather than merely reducing it.
Read the USG process as the baseline against which BPSG and PSG are both deviations, not the other way around: undoped oxide is what you deposit when neither reflow nor gettering buys you anything worth the dopant's cost, and that description covers the majority of the oxide in a modern chip — BPSG and PSG are the special cases, reserved for the specific pre-metal-dielectric and gate-adjacent contexts where a dopant's benefit is worth its risk.