Memory and training method for neural network based on memory
The present invention discloses a memory and a training method for neural network based on memory. The training method includes: obtaining one or more transfer functions of a memory corresponding to one or more influence factors; determining a training plan according to an ideal case and the one or more influence factors; training the neural network according to the training plan and the one or more transfer functions to obtain a plurality of weights of the trained neural network; and programming the memory according to the weights.
1. A training method for a neural network based on a memory, comprising:
obtaining one or more transfer functions of a memory corresponding to one or more influence factors;
determining a training plan according to an ideal case and the one or more influence factors;
training the neural network according to the training plan and the one or more transfer functions to obtain a plurality of weights of the trained neural network, wherein while performing training for a determined influence factor of the one or more influence factors, a plurality of outputs of a plurality of synapses of the neural network are adjusted according to the one or more transfer functions corresponding to the determined influence factor, a plurality of inputs of the plurality of synapses and a plurality of untrained weights of the plurality of synapses; and
programming the memory, by adjusting a plurality of resistors of a plurality of memory cells configured to represent the plurality of synapses of the neural network based on the memory according to the plurality of weights.
2. The training method according to claim 1 , wherein the one or more influence factors comprises temperature, a number of reads and writes, input voltage, change at each read operation, memory retention and programming error.
3. The training method according to claim 1 , wherein determining the training plan according to the ideal case and the one or more influence factors comprises: determining a first training time for the ideal case, and updating the training plan; and selecting at least one of the one or more influence factors, determining a second training time respectively for the selected at least one influence factor, and updating the training plan.
4. The training method according to claim 3 , wherein in the training plan, trainings corresponding to the ideal case are interspersed between the trainings corresponding to the one or more influence factors.
5. A memory, comprising:
a plurality of memory cells, configured to represent a plurality of synapses of a neural network, each of the plurality of memory cells comprising a resistor, wherein a plurality of resistance of the resistors are determined by the following:
obtaining one or more transfer functions of the memory cells corresponding to one or more influence factors;
determining a training plan according to an ideal case and the one or more influence factors;
training the neural network according to the training plan and the one or more transfer functions to obtain a plurality of weights of the trained neural network, wherein while performing training for a determined influence factor of the one or more influence factors, a plurality of outputs of a plurality of synapses of the neural network are adjusted according to the one or more transfer functions corresponding to the determined influence factor, a plurality of inputs of the plurality of synapses and a plurality of untrained weights of the plurality of synapses; and
programming the memory, by adjusting a plurality of resistors of the plurality of memory cells configured to represent the plurality of synapses of the neural network based on the memory according to the plurality of weights.
6. The memory according to claim 5 , wherein the one or more influence factors comprises temperature, a number of reads and writes, input voltage, change at each read operation, memory retention and programming error.
7. The memory according to claim 5 , wherein determining the training plan according to the ideal case and the one or more influence factors comprises: determining a first training time for the ideal case, and updating the training plan; and selecting at least one of the one or more influence factors, determining a second training time respectively for the selected at least one influence factor, and updating the training plan.
8. The memory according to claim 7 , wherein in the training plan, trainings corresponding to the ideal case are interspersed between the trainings corresponding to the one or more influence factors.