Rework Strip
Every single round covered in this entire project, from the storage-node hole etch through the bump and RDL opening, has moved a wafer forward — removing material to build toward a finished device, with no way back once it happened; rework strip is the one exception, the only step in this whole sequence whose entire purpose is to undo work rather than advance it, erasing a defective photoresist pattern so the exact same lithography step can be attempted again from scratch. If inline inspection catches a misaligned, mis-sized, or otherwise defective resist pattern before the etch that would transfer it into the underlying film ever runs, the wafer isn't scrapped — the photoresist gets stripped completely, the surface returns to its pre-litho state, and the lithography step repeats, recovering a wafer that would otherwise represent a total loss of everything built into it up to that point.
The etch step itself is the hard boundary between recoverable and unrecoverable in this entire flow, which gives inline litho inspection a real, concrete deadline rather than a general quality target. Once an etch transfers a resist pattern into the film beneath it — whether that's a storage-node hole cut into dielectric, a trench carved for M1, or a via punched toward a TSV tip — the underlying material has been permanently removed or altered, and there is no equivalent rework step anywhere in this project that undoes an etch the way rework strip undoes a litho mistake; this is exactly why CD-SEM and overlay metrology run between lithography and etch at nearly every patterned layer this project has covered, because that narrow window is the only point in the entire flow where a caught defect still has a cheap, complete recovery option rather than becoming permanent scrap.
Rework isn't free, even though it avoids outright scrap, which is why it functions as a safety net rather than a routine step in the flow. Repeating a lithography step consumes tool time, photoresist, and cycle time that a defect-free first pass wouldn't have needed, and some resist chemistries have practical limits on how many strip-and-recoat cycles a given wafer can tolerate before the underlying topography or surface condition starts to degrade — so fabs track rework rate as a yield and cost metric in its own right, treating a high rework rate as a signal that the lithography process itself needs tuning, not as a sign that the safety net is working as intended.
Read through a last-chance-to-undo lens, the rework strip is this entire project's single exception to an otherwise one-way sequence — every other round has described a transformation the wafer can't come back from, and this one exists specifically to prove that there's exactly one moment in the flow, right before an etch runs, where that's still not quite true.