Frequency-based multilayer neural network with on-chip learning

In this paper, frequency-based digital architecture of Multilayer Neural Network (MNN) with on-chip learning is proposed. In the frequency-based neural networks, pulse-streams are used for representation of the neuronal signals so that the proposed MNN can be effectively implemented on hardware. The back-propagation algorithm is modified and employed for the on-chip learning. The learning behavior of the MNN is tested by simulations and the feasibility of the architecture is verified.