A new computing rule for neuromorphic engineering

Neuromorphic engineering has helped to build a brain-inspired intelligent paradigm based on VLSI, and to promise a new application space on smart devices. Throughout most of the state-of-the-art neuromorphic systems, including analog, digital, the mixed one, as well as some other memristor-based systems, the dominated computing rule in neuron is simply based on the linear-superposition operation of the excitatory and inhibitory pre-synaptic inputs. However, recent discoveries in neuroscience field reveal that the post-synaptic potential at the soma cannot be directly achieved by linearly adding up all the pre-synaptic inputs, which indicates the prevailing computing rule in current neuromorphic systems is not very biologically plausible. In this paper, we introduce a new computing rule in neuron block, which nonlinearly depends on the pre-synaptic inputs. Besides the superposition operation, the membrane potential is also related to the product item of the excitatory and inhibitory inputs. Furthermore, we design a heuristic circuit for the bio-plausible computing rule, which is compatible with the crossbar structure based systems. These results provide a new insight into the role of inhibitory signals in the brain, which is very helpful to explore more reliable computing principles in future neuromorphic devices.