Fig. 5. Letter recognition with hardware-based circuit simulation by reflecting the measured characteristics of single-transistor neuron and synapse.

(A) Input image of the 3 × 3 pixel letter pattern. (B) SLP for a classifier and classification results. Each input layer represents each pixel, and each output layer represents each letter. Classification determined by which neuron expressed spiking first was performed. All other neurons except the first spiked output neuron were laterally inhibited. (C) MLP network for an autoencoder and its encoding/decoding results. Each input layer represents each pixel of noisy input, and each output layer represents each pixel of reconstructed output by the autoencoder. The output neuron that was fired could be newly decoded as a black pixel, and the output neuron that was not fired could be newly decoded as a white pixel to reconstruct a clearer image from a blurred noisy pattern.