### Extended Data Figure 8-2

Computer Simulations of flip-flop circuits: setting flip-flop synaptic weights and excitatory bias currents. *A*, Connection weighting was tuned prior to setting the level of excitatory bias current. Initial connection weights were divided by a factor, *d*, ranging from 1.8 to 2.8 in 0.1-unit increments. N_{j}→R_{i} weights were changed independent of R_{j}→N_{i} weights. A total of 121 combinations of N_{j}→R_{i} of and R_{j}→N_{i} weighting were used (Extended Data Fig. 9-1*A*). *B–D*, Examples (3/121) of the simulated combinations of flip-flop weighting. For each weighting combination, we calculated the difference in population firing rate over time and converted these data to a histogram showing the frequencies of binned R-N firing rate differences. Bimodal histograms indicate flip-flops that spontaneously switch between N-state and R-state. The height of the left and right peaks of the histograms indicate the prevalence of the N-state and R-state, respectively. The height of the intervening trough indicates the prevalence of N/R intermediate states. *E*, *F*, 11 × 11 simulation spaces where individual cells in the grid correspond to a given parameter combination and the color coding is a representation of the firing rate histogram generated from that particular set of simulations: the height of the N-peak, N/R trough, and the R-peak are indicated by the color of the left, middle, and right bars, respectively. The parameter combinations selected for experimental simulations are outlined in yellow. *E*, The stimulation space for synaptic weight (columns correspond to R→N weighting; rows correspond to N→R weighting). *F*, The simulation space for excitatory bias current (columns correspond to R neuron current; columns correspond to N neuron current). Download Figure 8-2, TIF file.