This paper describes the design principles for an implementation of an artificial neural network (ANN) in the form of a silicon integrated circuit based on charge‐coupled device (CCD) and metal‐nitride‐oxide‐semiconductor (MNOS) technologies. The significant features of this design are: (1) the synaptic coupling strengths stored in the MNOS devices can take on continuous, analog values and (2) the synaptic weights can be reprogrammed at any time under electrical control and in response to conditions in the network. These features should make possible ANNs that are capable of dynamic, in situ learning.