Fig. 4: Reusable 100-bit two-memory winner-take-all neural network.
From: Heat-rechargeable computation in DNA logic circuits and neural networks
a, Abstract circuit diagram, DNA implementation and number of strands. 1 ≤ i ≤ 100; 1 ≤ j,k ≤ 2 (k ≠j). For the input, input inhibitor and total number of strands, the two values to the left and right sides of the vertical line correspond to the specific input patterns shown in this figure and all possible inputs, respectively. Grey bars next to species names correlate with their reset temperatures; darker means reset at a higher temperature. b, Simulation of the reset. For simplicity, only the desired product species are shown; the concentrations of other species can be inferred from the conservation laws and are shown in Supplementary Fig. 11. c, Steps for testing the reusability of the system. Each reset involves input inactivation, heating and cooling and introduction of a new test pattern. d, Weighted sum analysis of ten representative Modified National Institute of Standards and Technology digits. A test pattern closer to the diagonal line is harder, and one farther from the diagonal line is easier to classify. e, Simulations and fluorescence kinetics experiments for classifying ten test patterns sequentially added to the same test tube. The DNA sequences are listed in Supplementary Table 5.
