A Study On The Design Of Kev Modules Of Asynchronous Mixed-Signal Neuromorphic Chip

With the development of artificial intelligence,more and more spiking neural network algorithms have been widely used in pattern recognition,voice analysis and sensory processing.However,these software algorithms always consume a large number of hardware resources.In order to apply neural network paradigm to efficient neuromorphic system,it is necessary to develop dedicated artificial intelligence chip.A reconfigurable mixed-signal spiking neuromorphic architecture is proposed.The chip comprises a total of 256x4 neurons,256x4 synapses and router circuits.Due to the biology plausibility and the implementation feasibility between different neural models,the integrate-and-fire(I&F)neural model is adopted.The I&F neuron implements spiking frequency adaptation,adjustable spiking threshold voltage and refractory period.While the weight update of spike-timing-plasticity(STDP)only depends on the time difference between the presynaptic and postsynaptic spike,this algorithm is not suitable for both encoding information represented in a spike correlation code and a mean rate code without spike correlations.For this reason,a more elaborate plasticity mechanism that not only depends on the timing of presynaptic spikes but also on other state variables presented at the postsynaptic terminal,such as the postsynaptic depolarization or neural calcium concentration,is adopted.What's more,the stop-learning mechanism is added to synapse so that the synapse can store more complex patterns.The implementation of AER protocol is used to realize the connection between different chips.In order to minimize memory resources and maximize network programmability and flexibility,a two-staged routing scheme is adopted.The key modules of neuromorphic chip are realized in SMIC 180nm technology.Simulations about different modules are made and results show the correct operation of proposed key modules.