IP개요 |
The investigation on neural networks using standard CMOS processes with memristor crossbar array has become continuous trend where reaction is silicon neuron with electronic synapses, large dynamic range and relatively low power consumption. This work presents a novel hybrid system based on an CMOS circuit with memristor crossbar array for multiple-layer neural networks. The proposed neural network consists of synapses and neurons. For mimicking synapses, Ag memristors have been proposed. The proposed neuron circuit is designed, analyzed and simulated using 0.18µm CMOS process. SPICE simulation results which used supply voltage of 3.3 V, showed that a neuron pulse is generated above a threshold input current of 70 ㎁. The classification of 5 ×5 pixel black/white image is demonstrated by the 5 × 5 cross bar memristor with pre- and post-neuron system. The proposed memristor devices with the Ag electrode demonstrate the adequate performance compared to the Au electrode, and may present noteworthy advantages in the field of neuromorphic computing. |