Research On Modeling And Parameter Estimation In 1/f Noise Based On Alpha Stable Distribution

With the rapid development of science and technology, electronic equipment is widely applied to many fields. In the fields of nuclear electric power generation, aviation and military industry and undersea exploration, the reliability of electronic equipment requirements is extremely high. As an important component unit of electronic devices, the reliability of semiconductor devices is closely related to the system stability, however, potentially defective devices were produced inevitably in volume production, they have a shorter service life and exist significant hidden danger to the stability of the system. Once they have a failure and they may cause irreparable damage. Therefore, reliability analysis method of semiconductor device has been the focus of research direction. In resent years, the research of reliability analysis method of semiconductor device focuses on the accelerated aging experimental methods and noise analysis methods. Compared with the accelerated aging experimental method noise analysis method has the shorter test time, low equipment requirement, high test efficiency and free of test device destruction. It has become the mainstream method of analysis and research. The characteristics of 1/f noise as a main part of low-frequency noise is closely related to the potentially defective of devices. According to the study, the inherent nature of 1/f noise has important implications for the reliability of semiconductor devices. The paper proposes that modeling approaches for 1/f noise based on alpha stable distribution and the method of parameter estimation. The specific research in this article is follows:First of all, this paper takes zener diode as the experimental object which is used to set up test system. The measurement method for noise, the anti-interference measures and the hardware structure and function of system are introduced in detail. According to the system structural noise and the analog circuit noise, etc, this paper chooses adaptive threshold denoising algorithm of wavelet packet analysis to process the test data, the experimental results show that the denoising effect is better. The composition of Low-frequency noise of semiconductor device and the basic concepts about alpha stable distribution are introduced. Secondly, according to the infinite variance test method 1/f noise is validated that belongs to the non gaussian distribution, and based on that, the modeling approachs for 1/f noise based on alpha stable distribution is proposed and the modeling theoretical researches is conducted. Three methods to estimate model parameters were studied, and the consistency of the experimental results show that 1/f noise meet the necessary conditions of alpha stable distribution. The gaussian distribution model and the alpha stable distribution model were established based on 1/f noise. According to compare the deviation degree of theprobability density curves 1/f noise is considered to meet the alpha stable distribution preliminary. To verify the goodness of fit of alpha stable distribution models, a method of hypothetic test based on the eigenfunction and properties of stable distribution has been proposed in this paper. Thirdly, based on conventional spectral estimation theories, the methods of parameters estimation of 1/f noise based on covariation spectrum, fractional lower order covariance spectrum and the generalized fractional low-order covariance spectrum were proposed, and the parameters were got from the experiment. Finally, the relation between Fractional autoregressive integrated moving average(FARIMA) model and alpha spectrum is deduced. An methods of parameters estimation of 1/f noise based on FARIMA model is proposed, several FARIMA model parameter methods were studied, then alpha spectrum is obtained by experiments and estimate the index of the 1/f noise frequency.