Title :
Neural learning in analogue hardware: effects of component variation from fabrication and from noise
fDate :
4/15/1993 12:00:00 AM
Abstract :
Simulations of backpropagation networks in analogue circuitry with on-chip learning are presented. Forward and backward computations are performed using Gilbert multipliers. Component variations fixed in fabrication are shown to be adapted to and tolerated much better than similar variations due to noise.
Keywords :
analogue computer circuits; backpropagation; multiplying circuits; neural nets; Gilbert multipliers; backpropagation networks; component variation; fabrication; noise; on-chip learning;
Journal_Title :
Electronics Letters
DOI :
10.1049/el:19930464
