Standard cells for auto scaling pulsed analog neural circuits

The authors present an approach to implementing neural networks that uses pulse streams modulated by nonlinearly weighted synapses and neurons that generate individual pulses whenever their threshold is exceeded. The distribution of the membrane capacitance and the related elements to the synapses automatically scales the capacitance to the incoming excitatory charges, allowing the synapses to be developed as cells and employed over a very wide range of connectivity. Standard cells are presented along with simulation results for a number of different neural networks.< >