Design of a High-Precision ANN Circuit with Sigmoid Activation Function
An artificial neural network(ANN)circuit with Sigmoid activation function(AF)is proposed to realize the classification of tar-get data.Based on the TSMC 180 nm CMOS process,the circuit design,layout design,simulation and chip test are completed.The chip area is 0.337 5 mm2,and the circuit power consumption is 216 μW at 1.8 V supply voltage.The test results of three chips show that the classification accuracy of the circuit for target data is 97.5％,the error is between-3.2％and 3.27％,and the average error is-1.51％.The ANN circuit consists of 3 network layers and is composed of 15 Sigmoid AF neuron circuits.The two-inputs Sigmoid AF neuron cir-cuit is composed of two weighted circuits and a Sigmoid AF circuit.The post-simulation results show that the power consumption of neu-ron circuit is 8.02 μW at 1.8 V supply voltage,the maximum noise is 80 pA/√Hz,the layout area is 0.001 mm2,and the maximum rel-ative error is 1.76％.
万方数据
维普
