Analog spiking neuron
Context
Many different neuron models have been developed since the modelling work of Hodgkin and Huxley in the 1950's, and some models have been successfully integrated on silicon. Analog neurons are power and cost-effcient: an analog computation (temporal integration, spatial summation, leakage) performed by an electronic neuron takes direct advantage of the laws of physics (capacitive integration, current summation, leakage current), and it is well suited for low resolution real-time computation.
Our approach
Our goal is to design a Leaky Integrate-and-Fire (LIF) neuron that is suitable for computing purposes. To this aim, it has to be suitable for VLSI integration (compact & low power), robust to process variability (UDSM process is used), while maintaining high linearity and accuracy (computational abilities).
Structure of an analog leaky integrate-and-fire neuron.
