Researches On The Algorithm And Software Of The Complexity Analysis Of Electrical Noise

Noise inside the electronic devices (especially the low-frequency noise such as 1/ f noise) is the key factor limiting the sensitivity and precision of the devices. And at the same time, it is the important sensitive parameter to characterize the quality and reliability of the devices. Therefore, noise has been studied and used widely as the diagnostic and evaluating tool.Because electrical noise is a stochastic signal, the traditional analysis approach is power-spectrum. The parameters extracted from power-spectrum such as spectral denisity, fitted magnitude and frequency exponent are used to denote the changes of devices mechanisms. However, power-spectrum is a linear approach which is not so suited for non-linear, non-stationary and singulary signal like noise. It is required to present a new analysis method for electrical noise. Apparently, electrical noise is irregular, uncertain and random. In other words, the signal is complex. Therefore, complexity analysis approach is applied to electrical noise in this thesis.In this thesis, we focused on some typical methods such as correlation entropy, correlation dimension, Lempel-Ziv complexity and fluctuation complexity. We realized their algorithms and verification. Then an exploration into the multi-parameter characterization was conducted based on the time domain of electrical noise. Gathering all the complexity analysis approaches appear in this thesis and Lab, we designed and realized the Complexity Analysis Software of Electrical Noise.After applying these methods to emulational noise and experimental noise, we find that correlation entropy and correlation dimension based on correlation integral can distinguish the random telegraph superposition noise from other 1/ f models. With the frequency exponentγincreases, correlation entropy, correlation dimension and Lempel-Ziv complexity decreases but fluctuation complexity increases and then decreases. Complexity analysis approaches are valid in characterizing elecromigration process and predicting interconnection failure. They can also show the complexity changes of MOSFET at different bias, gate voltage and radiation dose.This thesis proposes a new method for processing electrical noise, thus proving its validity and superiority from the emulational noise and experimental noise, and further it offers an effective method for the research and appraisal of electrical low-frequency noise and its model.