A hardware neural network for handwritten digits recognition using binary RRAM as synaptic weight element

A functional neural network for handwritten digits recognition using single binary RRAM as the synaptic weight element, was presented, with successful recognition rate is up to 81%. The results show that multilevel or continual resistances is not necessary for resistive device used as synaptic weight element in hardware neural network application. Variation of RRAM devices, sometimes, would not affect, but benefit the performance of the neural network. The simulated binary RRAM based network is so robust, that even if half of the basic elements were die, it will still be functional.