| Noise is a general phenomenon that exists in most of the natural and artificial systems, such as electrical circuits and electronic devices. The study of noise phenomena is of great significance both in scientific research and in practical application. During the up to one century investigation process, noise was found in almost all types of electronic materials, devices and systems with a variety of expressions and contents. In order to deepen the understanding of electronic noise, some recently developed signal processing algorithms were applied to the analysis of the low frequency noise in semiconductor devices, and at the same time, to explore the capacity of low frequency noise in the characterization of the quality and reliability of semiconductor devices.The conventional analysis method of noise in semiconductor devices is spectrum analysis, and from which a set of characteristic shape parameters can be obtained. However, these parameters can not fully describe the features of the noise signal. For example, the noise is a mixed signal of different noise sources in the device, so it is hard to extract the parameters of all noise components because of their overlaid spectra. Further, with the increase of the complexity of the noise sources involved, the errors of the extracted parameters increase, which degrade the precision of the analysis. Moreover, with the emergence of the advanced devices, non-Gaussian, nonlinear, non-stationary noise was found in more and more samples with newly developed signal analyzing techniques. These new peculiarities can not be characterized by conventional spectrum analysis. For these reasons, the multi-parameter method (MPM) including spectrum parameters should be employed to fully describe the features of noise signal in electronic devices.The MPM analysis of the noise signal in electronic devices can be conducted in the following ways. First, some feature parameters of the noise signal are extracted by time domain, frequency domain and time-frequency domain analysis. Second, the peculiarities of the noise signal are constructed by the extracted feature parameters. Finally, the peculiarities are used to uncover the microscopic mechanisms of the corresponding noise sources at the light of device physics.Three category innovative results concerning the MPM analysis of noise signal in electronic devices were acquired, such as a new noise generating model, three novel applications of signal analyzing methods, and some conclusions. In the new model, a density distribution function of the border traps by activation energy was brought forward, and on which a 1/fγnoise model for MOSFET devices and a 1/fγnoise Monte Carlo simulation routine based on the thermally activated trapping and de-trapping of border traps were presented. With this new model, 1/fγnoise time series with controllable exponent index y can be obtained.The three novel applications are the applying of signal analyzing methods, such as wavelet analysis, higher-order statistics analysis and complexity analysis, in the extracting feature parameters of noise signals. Their details are as follows.1. Based on wavelet transform, four analyzing methods were brought forward, such as wavelet modulus maxima statistics, average Lipschitz exponent, wavelet similarity coefficient and local Lipschitz exponent analysis, in which the statistical distribution of noise wavelet modulus maxima were depicted by wavelet modulus maxima statistics analysis, the local singularity of noise signal was described by the local Lipschitz exponent analysis, the entire singularity of noise signal was described by the average Lipschitz exponent analysis, and the shape similarity between two noise signals was characterized by wavelet similarity coefficient analysis.2. The quantitative analysis of the Gaussianity and linearity of electronic noise were conducted by using the quadratic sum of the bicoherence S as a feature parameter. Based on the summing-up of the computer simulated noise signals, a quantitative S criterion for the classification of noise signals was established, in which linear Gaussian noise has an S value in the interval of [0,2), non-linear Gaussian noise pertains an S value in the interval of [2,60), linear non-Gaussian noise holds an S value in the interval of [60,500), while non-linear non-Gaussian noise possesses an S value in the interval of [500,+∞). It was also proved that the widely existing noise in electronic devices can be distinguished by this method.3. The LZ complexity and fluctuation complexity analysis for noise signal were realized and improved, from which the complexity of noise signal can be effectually portrayed. During the computation of fluctuation complexity, the conventionally adopted mean value binary coarse-grained process was replaced by the newly introduced difference coarse-grained process, and the effectiveness of this new grain mode was verified by actual analysis.The conclusions were achieved by applying the above mentioned analyzing methods to the real case of MOSFET devices.1. According to wavelet modulus maxima statistics, some conclusions regarding the noise of MOSFET devices can be drawn.1) The wavelet modulus maxima statistics of the experimental noise data in both nMOSFETs and pMOSFETs pertained values similar to that of the RTS superimposition simulated low frequency noise.2) Radiation did not alter the noise generation mechanisms in both nMOSFETs and pMOSFETs. And 3) Radiation degraded nMOSFETs more severely than pMOSFETs, but did not create new types of defects, on the other hand, aggrandized the population of the prototype ones.2. From the results of the quadratic sum of the bicoherence, it is known that,1) non-Gaussian noise occurs in nMOSFETs,2) the noise's non-Gaussian degree in small size devices is stronger than in large size devices, the non-Gaussian degree of nMOSFET's noise in strong inversion and linear regime increase with drain-source voltage. All these phenomena can be attributed to the central limit theorem and the noise physics of nMOSFETs.3. Based on LZ complexity analysis we know that, the LZ values of MOSFET devices biased in linear region decrease with drain voltage, and increase with gate bias, while the variation amplitudes of LZ values in nMOSFETs are more prominence than in pMOSFETs.4. The two parameters for fluctuation complexity analysis are fluctuation complexity (FC) and mean information gain (mIG). In MOSFET devices, with the increase of drain voltage, FC value decreases, mIG value increases, and C-R value move to the down-right side of Bernoulli curve. Whereas, with the increase of gate voltage, FC value increases, mIG value decreases, and C-R value move to the up-left side of Bernoulli curve.All the results acquired in this thesis provide solid experimental and theoretical bases for the future noise research in semiconductor devices.
|
PDF Full Text Request