Fig. 6: Physical performance modeling results.

a Time and (b) energy to solution of the proposed WalkSAT/SKC and high-order HNN solvers for random uniform 3SAT problems. The shown data are computed based on developed hardware architecture (Supplementary Figs. S18, S19) and their models (Supplementary Table S1–S4) – see Supplementary Notes 7, 8 for more details, and hardware-aware algorithmic time-to-99% solution certainty post-processed from cumulative distributions (Supplementary Fig. S25). The shaded area in the plot shows the interquartile range (25–75%), and the markers indicate the median across instances of that problem size. Hardware-aware algorithmic simulations are taking into account errors in break/gain values due to memristor tuning errors, with the assumed relative value of ~ 2.4% and ~ 20% (corresponding standard deviation for the conductance tuning of 3 μS and 0.25 μS) for the on- and off-state memristors, respectively.