Price a Maskset Respin When Even One Bit Is Wrong

# Price a Mask-Set Respin When Even One Bit Is Wrong

## 1. Treat a Single Wrong Bit as a Whole-Set Problem

A mask ROM's bit pattern was committed to a physical plate, not to a rewritable medium, so a single bit discovered wrong after mask fabrication could not be corrected by changing that one bit on the existing mask. The crossing in question either had the fabricated connection or it did not, and the mask that defined which crossings got one was already cut. Correcting even one bit, in general, meant going back to mask design with the corrected pattern and running the entire multi-week design-and-fabrication sequence from Step 5 again for the affected layer or layers — a full respin, not a patch.

## 2. Add the Real Costs a Respin Actually Carried

A respin cost real money for the mask set itself (a fresh plate run at the mask shop, per affected layer), real calendar time (the same four-to-eight-week stretch from Step 5, run again, on the critical path of whatever product was waiting on the corrected ROM), and often real wasted silicon — any wafers already started with the wrong pattern before the bug was caught had to be scrapped or diverted, since there was no way to rewrite their already-fabricated bit pattern after the fact. For a product whose schedule assumed the first mask set would be correct, a late-discovered single-bit error could push a ship date back by the length of an entire respin cycle, which is exactly why bit-pattern verification before tape-out, covered conceptually in Step 7, carried so much weight in this industry's actual design discipline.

Step 7 names the tape-out commitment point precisely: what can still be fixed afterward, and what genuinely cannot.