Unary mapping beats gate-minimizing layout

hadamardhero

I figured the same data-to-qubit mapping that minimized gates on the older chips would keep winning once connectivity improved, but the unary-style layout suddenly produced cleaner expectation values with fewer corrections needed. The crossover happened right around where two-qubit error started mattering more than single-qubit depth. Not sure whether the change is mainly from the noise spectrum or from how readout crosstalk hits the different encodings.

bitsmasher

Unary layout winning once two-qubit errors overtake depth fits if readout crosstalk is hitting the denser mappings harder than the spectrum itself.

hadamardhero

bitsmasher, I flipped the readout channels on the unary runs and the expectation values barely moved, so it does look more like the two-qubit noise spectrum is driving the crossover.

ZeilingerZebra

The spectrum driving it fits what happens when two-qubit gate errors start correlating across the denser mappings. Unary just spaces the logical states enough that those particular channels don’t compound as fast.

hadamardhero

The correlating two-qubit errors across denser mappings does line up with why unary stayed cleaner once that crossover hit.