Spin-on Glass Process
The spin-on glass process applies SOG as a liquid — a siloxane or silicate precursor dissolved in solvent, spun onto the wafer exactly like photoresist — rather than growing or depositing it from a gas phase the way every other oxide covered in this glossary is made. That liquid starting point is the whole distinguishing feature: SOG planarizes by ordinary liquid self-leveling during the spin itself, before any bake or cure ever happens, instead of relying on a later thermal reflow (the BPSG process) or on flow properties built into a vapor-phase deposition chemistry (the SACVD process).
Spinning a liquid onto a wafer planarizes for the same reason spinning photoresist does: centrifugal force and surface tension pull a liquid film toward a flat, self-leveling surface well before any chemical reaction or thermal step comes into play. That gives SOG a planarization mechanism no vapor-deposited film can match in kind — it isn't flowing because the deposition chemistry happens to leave the film mobile (SACVD) and it isn't flowing because a later anneal softens an already-solid network (BPSG); it's simply a liquid obeying ordinary fluid mechanics while it's still a liquid, the same physics that makes spin-coated photoresist thickness so uniform and reproducible across a wafer.
That cracking risk, not just cost or throughput, is the real reason SOG is almost never used as a standalone thick gap-fill film the way SACVD or TEOS oxide can be. As the spun-on solvent evaporates and the film cures, it shrinks — and a thick, continuous region of cured SOG (typically over open field areas rather than narrow gaps) can't relieve that shrinkage stress without cracking, while a thin film sitting only inside a narrow trench or via gap shrinks under far less absolute stress and survives intact. Production flows exploit that difference directly: spin on SOG thick enough to fill the narrowest gaps completely, then blanket-etch the film back until only the material actually sitting inside those gaps remains, deliberately removing the crack-prone thick regions rather than trying to prevent them from forming in the first place.
The oxide/SOG/oxide sandwich structure exists because SOG alone isn't a good enough dielectric or moisture barrier to use on its own, even where it survives the cure without cracking. A typical multilevel-interconnect planarization stack from the SOG era deposits a CVD oxide first (for a clean, well-controlled lower interface and moisture barrier), spins on SOG to fill the gap and locally planarize, etches that SOG back, and then caps the whole structure with another CVD oxide layer — SOG does the planarization job in the middle of the sandwich, while CVD oxide handles the interface quality and barrier properties on both sides of it.
Siloxane-based and silicate-based SOG trade crack resistance against dielectric density in opposite directions, the same kind of material trade-off every doped-glass variant in this glossary makes somewhere. Silicate SOG is closer to pure, dense silicon dioxide and gives better dielectric properties, but is more prone to cracking at a given thickness; siloxane SOG incorporates organic methyl or similar groups into the network, which makes the cured film more flexible and crack-resistant at the cost of a higher dielectric constant and somewhat lower thermal stability — a choice between a denser, more brittle film and a tougher, lossier one.
Read the spin-on glass process as the one dielectric gap-fill method in this glossary that planarizes before it's even a solid, using the same centrifugal spreading mechanism that makes photoresist coating uniform, and then has to be deliberately thinned back everywhere except the gaps it was spun on to fill, because the same liquid-to-solid shrinkage that makes it planarize so well is exactly what makes a thick cured film crack on its own.