Copper CMP Dual Damascene
Once copper has been electroplated into a dual-damascene trench and via until it overflows well above the dielectric surface, that excess metal — the overburden — has to come off completely and the field dielectric has to end up flat before the next interconnect level can be patterned on top of it, and that job belongs entirely to copper chemical-mechanical polishing (Cu CMP), a two-stage polish that strips bulk copper first and then clears the underlying tantalum-based barrier without gouging the soft copper lines it just finished planarizing. Cu CMP sits immediately downstream of the electroplating fill step and directly upstream of the CoWP selective capping already covered in cowp-cap-process, and it inherits the same barrier/seed stack that pvd-barrier-seed-redeposition built at the bottom of every trench — because that barrier is exactly what the second polish stage has to clear without damaging the copper lines or digging into the low-k dielectric around them.
The polish is deliberately split into two chemically distinct stages because copper and the tantalum-based barrier beneath it respond to completely different slurry chemistries, and running one slurry for the whole stack would either leave barrier residue shorting adjacent lines or remove so much copper that line resistance rises out of spec. Stage one uses a slurry formulated for a high copper removal rate — typically an oxidizer-plus-complexing-agent chemistry similar in spirit to the bath-chemistry tuning already familiar from copper-electroplating, but aimed at dissolution rather than deposition — and it runs until the bulk overburden is cleared down close to the barrier layer, deliberately stopping short of fully exposing it so the more aggressive bulk-removal chemistry never touches the dielectric directly. Stage two switches to a slurry selected for a high selectivity ratio between the Ta/TaN barrier and both the copper lines and the surrounding low-k or SiO2 dielectric field, so it can clear the remaining barrier film essentially everywhere at once without preferentially eating into the softer copper or punching through the dielectric in isolated spots.
Endpoint detection is what keeps the two-stage polish from simply running on a fixed timer and gambling on removal-rate uniformity, because the moment the barrier clears is exactly the moment the process has to stop before it starts removing copper and dielectric at the much higher rate characteristic of over-polish. In-situ optical reflectance monitoring and polish-head motor-current signatures both shift measurably when the exposed surface transitions from metal barrier to dielectric, since the two materials interact with the polish pad and slurry very differently, and that shift is what the tool uses to trigger the transition from stage-2 polishing into a short, lower-pressure buff step rather than letting bulk removal rates continue unchecked across a surface that no longer needs them.
| Removal-rate / defect metric | Dense narrow-line regions | Wide / isolated-line regions | Primary lever to control it |
|---|---|---|---|
| Copper dishing | Lower risk; narrow lines offer less total copper area for the pad to dish | Higher risk; wide copper areas dish more under sustained pad pressure | Pad hardness, down-force, polish time tuned to line-width distribution |
| Dielectric erosion | Higher risk; densely packed lines leave less dielectric support between them | Lower risk; isolated lines surrounded by ample dielectric field | Slurry selectivity (Cu/barrier vs. dielectric), pattern-density-aware recipes |
| Barrier residue / micro-shorts | Higher risk if stage 2 under-polishes; residual Ta/TaN can bridge adjacent lines | Lower risk; wider spacing tolerates minor under-clearing better | Stage-2 time/selectivity, endpoint-detection sensitivity |
| Scratch / particle defects | Comparable across pattern density; driven more by slurry particle control | Comparable across pattern density | Slurry filtration, pad conditioning, post-CMP clean |
Dishing and erosion are not cosmetic blemishes; they are a direct electrical and mechanical liability, because a dished copper line has less cross-sectional area than its design width implies, raising resistance exactly where the interconnect is supposed to be lowest-resistance, and an eroded dielectric field around densely packed lines leaves less structural support for every polish and deposition step that follows. This is why pattern-density awareness — tuning pad pressure, slurry chemistry, and polish time differently depending on whether a given die region is densely or sparsely wired — has become as important to Cu CMP recipe design as the raw removal-rate numbers themselves, since a recipe calibrated only against isolated test structures will reliably underperform across the mixed dense/sparse wiring that real interconnect layouts actually contain. Once the polish clears cleanly with dishing and erosion held within spec, the wafer surface is flat enough and the copper sufficiently exposed for cowp-cap-process to selectively cap each line, continuing the same interconnect-reliability chain that pvd-barrier-seed-redeposition started at the bottom of the trench and that via-etch-stop-dual-damascene-integration carries forward into the next metal level.