Refresh Method for the Weights of the Analogue Synapses

The analogue implementation of spiking neural networks have the critical advantage of parallel operation and information transmission. The implementation in hardware of the synaptic weights and learning mechanisms represents a challenging task. A cheap and efficient method to store the weights of analogue SNN which benefits from real time adjustment represents the capacitors. However, the main disadvantage of this method represents the leakage currents which make the synaptic weights volatile. This paper proposes and evaluates experimentally a method for refreshing the weights of the analogue synapses by rhythmic activation of the neurons. The results show that the long term memory of analogue neurons improves significantly if the neurons are activated at certain periods of time. Moreover, in order to obtain non-volatile memory the refresh rate can be determined as a function of the initial value and the variation of the synaptic weights.