Research On 1/f Noise Statistical Model And Spectrum Estimation Of Semiconductor Devices

In the fields of nuclear industry,aerospace,national defense and military industry,the research on reliability screening of semiconductor devices has aroused great attention.It is now well established that method of noise analysis is one of the most universally used non-destructive methods in device reliability screening.Through a great deal of experimental results,it is proved that 1/f noise spectrum(for short,1/f spectrum)has been generally expected to be qualified for indicating the reliability of devices.Therefore,1/f spectrum estimation is an important method of semiconductor devices reliability screening.In this paper,1/f noise statistical model and spectrum estimation method for semiconductor devices are established and developed.The main research work and innovations are as follows:1.Firstly,the relationship between 1/f spectrum and reliability of semiconductor devices is welly discussed.The basic theory of alpha-stable distribution and its applications are introduced,which lays a theoretical foundation for the research work in this paper.2.In order to analysis the statistical model of 1/f noise in semiconductor devices,a dual-channel system for measuring noise of semiconductor devices is established.The Zener diode is taken as the experimental subject.3.In the past,it was considered that 1/f noise of semiconductor devices accords with Gaussian distribution.In this paper,the probability model of the observed noise of semiconductor devices and the background noise of measurement system was validated and analyzed,and we propose that 1/f noise of semiconductor devices obeyed alpha stable distribution.Firstly,through the infinite variance test,goodness of fit test of alpha-stable distribution,consistency test of parameter estimation of characteristic function and experimental analysis,it is proved that the system interference noise obeys Gaussian distribution,in addition to Gaussian noise,alpha-stable distribution noise(for short,alpha noise)is found in the observed noise of semiconductor devices.Then,it is concluded that 1/f noise of semiconductor devices obeys alpha-stable distribution,and that 1/f noise of semiconductor devices can be regarded as the output of linear models driven by i.i.d.alpha noise(AR,ARMA model,etc.).Finally,the problem of 1/f spectrum estimation of semiconductor devices is transformed into the problem of linear model parameter estimation of alpha-stable process,which opens up a new development of 1/f spectrum estimation of semiconductor devices.4.Aiming at the parameter estimation of alpha-stable distribution process,a method of estimating dispersion coefficient and characteristic index is proposed.The principle of this method is to estimate the dispersion coefficient and the characteristic index respectively based on the sample characteristic function and the stability property.The simulation results show that the proposed method has good performance on estimation accuracy.5.Under the background of independent zero-mean Gaussian colored noise,normalized cross-correlation function(NCC)is proposed.Based on the discussion of NCC operator's properties,a method based on NCC operator to estimate the AR and ARMA parameters and 1/f spectrum of semiconductor devices are proposed,which effectively improve the accuracy of the parameters estimation and 1/f spectrum estimation.6.In the environment of correlated and independent Gaussian noise,normalized third-order cumulants(NTC)and normalized third-order cross cumulants(NTCC)are proposed.Based on the analysis of operator's properties,a method based on NTCC operator to estimate the AR and ARMA parameters and 1/f spectrum of semiconductor devices are proposed,which further provides a novel method for reliability screening of semiconductor devices.