Fig. 4: Experimental results.
a, b Measured output (bit-line) currents for (a) forward and (b) backward passes. In panel (a), currents are grouped according to their ideal clause literal sum values. In panel (b), outputs are shown against calculated ideal break values. In both panels, data are collected during a single iteration of an algorithm. c, d Functional performance comparison between ideal software model and experiment, showing (c) evolution of the number of unsatisfied clauses, and (d) run-length distribution curves obtained across 200 iterations (restarts) of the algorithm, each time with new randomly initialized variable assignments. In all experiments, MAX_FLIPS = 10,000.
