ANN Modeling Of Nano MOSFET Noise For Low-power And Millimeter-wave Application

The characterization of RF noise characteristics of nanoscale MOSFET is the basis of RF IC design.In new wireless communication networks and systems,the low-voltage and low-power design requirements make it more and more important to study the characteristics of MOSFET in the weak inversion region.As the feature size of MOSFET devices continues to decline and the feature frequency continues to increase to millimeter wave band,a new study on the noise characteristics of 40 nm MOSFET shows that the noise characteristics in the weak inversion region are different from those in the strong middle inversion region,showing obvious thermal noise characteristics.Therefore,this paper mainly studies the following contents:Firstly,the noise parameter extraction technology and noise model analysis theory which are widely used at present are used to clarify the noise mechanism of the weak inversion region of 40 nm MOSFET.Combined with the small signal noise equivalent circuit of nanoscale MOSFET,a semi empirical model for characterizing the drain current noise of40 nm MOSFET is established.At the same time,it is found that there are some limitations in the related research methods and noise modeling techniques.The main problems are the complexity of the modeling process and the inability of the model to accurately predict the noise value of unmeasured bias point or unmeasured frequency and temperature range.Artificial neural network(ANN)has the advantages of simplicity,rapidity,high precision,good extrapolation and interpolation in the field of RF / microwave modeling.In view of the limitations of the semi empirical model,an ANN model is established to characterize the thermal noise characteristics in the weak inversion region of 40 nm MOSFET.The model can well fit the input-output relationship,and there is no under fitting or over fitting phenomenon.The application frequency of ANN model is 60 GHz,which can characterize and accurately predict the noise value of the undetected bias point or the undetected frequency and temperature range in the weak inversion region of 40 nm MOSFET,and has good interpolation and extrapolation function.Finally,the ANN model is verified.The semi empirical model and ANN model built inthis paper are embedded in the simulation Software of four noise parameters of the device,and the simulation data of four noise parameters of the two models under the same bias,frequency and temperature conditions are obtained.The experimental results verify the superiority of ANN model in precision and interpolation and extrapolation.