An Analog and Time-Discrete Neuron with Charge-Injection for Use in Ultra-Low Power Spiking Neural Networks

This paper shows a new concept for a neuron circuit for use in artificial spiking neural networks. The circuit performs the simplified analog calculations (LIF-neuron) of a biological neuron but uses advanced analog circuit design techniques to achieve functionality in a very small area and with large-scale integration in CMOS technology in mind. This neuron consumes only a tiny amount of power and is well suited for use within a spiking neural network in smart, ultra-low energy sensing applications. Due to the mapping of a spike event directly to a central clocking signal, large-scale integration and the mapping of weights using standard artificial intelligence algorithms have been drastically simplified. Simulations show that the analog computation concept can save a lot of energy.