Training method and training device for neural network based on memristors

A training method for a neural network based on memristors, wherein the neural network comprises a plurality of neuron layers connected one by one and weight parameters between the plurality of neuron layers, and the training method comprising:

• during an off-chip training process, training the weight parameters of the neural network to obtain quantized weight parameters, and programming a memristor array based on the quantized weight parameters to write the quantized weight parameters into the memristor array;
• and
• during an on-chip training process, updating a critical layer or several critical layers of the weight parameters of the neural network by adjusting conductance values of at least part of memristors of the memristor array,
• wherein the programming a memristor array based on the quantized weight parameters to write the quantized weight parameters into the memristor array comprises:
• for each memristor of the memristor array:
• determining a target interval of a conductance value of the memristor based on a quantized weight parameter corresponding to the memristor, wherein the target interval is expressed as (It−ΔI, It+ΔI), It is a current value corresponding to a, conductance value of the memristor at a preset voltage, and ΔI is a current error tolerance corresponding to the conductance value of the memristor;
• and
• in response to the conductance value of the memristor being within the target interval, determining the quantized weight parameter corresponding to the memristor to have been written into the memristor.