Search

In recent decades, the modeling and designing of neuromorphic systems have received considerable attention, leading to the development of electrophysiological devices that can mimic the dynamic behavior of cortical networks. Some features, such as parallel processing, real-time performance, reconfigurability, low production cost, and good accuracy, make field-programmable gate arrays (FPGAs) an ideal hardware platform for implementing spiking neural networks (SNNs), which has become very popular in the field of neural computing. In this work, we investigated a high-performance FPGA design of the auditory attention (AA) to simulate the on-chip “cocktail party” effect similar to what happens in the brain. In the first step, we introduce a novel spiking winner-take-all (SWTA) network consisting of a hardware-friendly Izhikevich neuron model and explore its biological characteristics and stability. We then use this network as a receptive field in the primary auditory cortex, which is tuned by attention signals. The desired sound frequency will be selected as the target and transmitted to the higher layers of the auditory cortex for further processing, while other frequencies are severely suppressed. Experimental results show that the proposed network accurately follows the dynamic behavior defined in the AA theory and emphasizes the resource consumption after post place and routes on a fully efficient hardware implementation, where less than 4% of the resources were used to implement a network with 1856 neurons.