METHOD OF TRAINING A NEURON DEVICE

摘要

A method is described of training a pRAM device to produce a desired output behaviour in response to real-valued inputs. In a simple form thereof, each address input of the device is supplied with a pulse train having a length R equal to a first length R1, and the device computes an output as a function of the pulse outputs produced by the device in response the pulses of the input pulse trains. The behaviour of the device is assessed after carrying out this procedure at least once, and a performance measure produced therefrom. A new value of R = R2