A 32-byte key driving placement, CTS, and routing marks is a fun way to make the watermark follow the design around (instead of living in one brittle spot).
The 10^-32 coincidence number sounds very tidy, though, until you ask what “weakening the watermark” costs in actual PPA...
> The 10^-32 coincidence number sounds very tidy
Yeah, thats the part I dont buy, it smells like per-instance math, not an end-to-end attack metric. What does the flow pay to keep that number?
probably mostly a handful of extra placement/cts/routing constraints plus some reroute churn around the marked spots, so the cost is whatever ppa slack you burn to keep those coincidences pinned down. if the paper says comparable or smaller overhead, im still curious how many hotspots they had to touch before the key starts hurting timing or congestion in a real design.