a first neuron circuit, including a first computing layer set of a trained neural network;

a second neuron circuit, including a second computing layer set of the trained neural network;

a pre-processing circuit, performing pre-processing on analog input signals to generate a plurality of inputs of the first neuron circuit, the pre-processing comprising one of signal amplification, filtering, noise suppression, compensation, digital-to-analog conversion, analog feature extraction, or a combination thereof;

a processor, coupled to the first neuron circuit and the second neuron circuit, configured to:

execute a first neural network computation for a fixed feature pattern through the first computing layer set; and

execute a second neural network computation for an unfixed feature pattern through the second computing layer set, wherein a computing speed of the first neuron circuit is higher than that of the second neuron circuit or a power consumption of the first neuron circuit is lower than that of the second neuron circuit;

wherein a training process of the trained neural network comprises:

building a neural network;

sending training data to the neural network for training, so as to obtain the trained neural network, and dividing the first computing layer set and the second computing layer set in the trained neural network;

in response to receiving new training data, sending the new training data to the trained neural network obtained by previously training for retraining to generate a new neural network,

wherein in a process of retraining the trained neural network, an update of a structure and weights of the first computing layer set is not performed, but only an update of at least one of the structure and weights of the second computing layer set is performed.