A novel hardware-efficient liquid state machine of non-simultaneous CA-based neurons for spatio-temporal pattern recognition

In this paper, a novel liquid state machine (LSM) comprising neurons whose nonlinear dynamics are described by a non-simultaneous cellular automaton (CA) is proposed. The proposed LSM is applied to a supervised classification task. It is shown that the proposed LSM can recognize spatio-temporal spike patterns with high accuracy. Furthermore, the non-simultaneous CA-based neuron is implemented on a field programmable gate array (FPGA), and an experiment validates its spiking function. It is then shown that the non-simultaneous CA-based neuron occupies fewer FPGA resources compared with typical conventional neuron models, such as the Izhikevich, leaky integrate-and-fire, quadratic integrate-and-fire, and Morris–Lecar neurons.