Analysis Of Electrical Devices Noise By Using Multi-scale Entropy Method

The interior noise of the electronic devices (especially the low-frequency noise such as 1/f noise) contains a large amount of information of the quantity and reliability of the devices. Conventional methods for analyzing the noise based on the power spectrum in the frequency domain focus only on the whole singularity of noise signal but neglect the interior complexity of time series. Usually entropy is used to indicate the irregularity and complexity of time series. However, traditional methods based on entropy always get different results when processing a real signal, because they are based on single-scale. Real noise signals such as 1/f noise have correlation over different time scale, whose interior complexity has close relation with correlation on multi-scale. For solving those problems existed in the traditional methods based on entropy, multi-scale entropy approach is applied to analysis of the noise of electric devices with the consideration of the correlation over different scale of time series.The principle of multi-scale entropy method is presented with its arithmetic, which is realized by designing software. When applied to white noise and 1/f noise, it is proved that the complexity of 1/f noise is more complicated than white noise. According to the different-shaped entropy curves, multi-scale entropy method can not only distinguish 1/f noise produced by four different modules (Brown, Soc, Dis and RTS) but also noise time series generated by the same modules under different conditions. In practical application, multi-scale entropy has proved that the little size MOSFET1/f noise is possibility superposed formed by RTS noise from its module, the result and theory prophecy are totally identical. In addition, upon analyzing the noise signal of metal interconnection electromigration, multi-scale entropy method can characterize the change of noise modules during electromigration and the change of complexity induced by altering the interior mechanism, so it offers a more full-used method than using the frequency exponent to indicate electromigration.This research proposes a new method for processing noise of electronic devices available, thus proving its validity and superiority from the theory and the analysis of real signal, and further it offers an effective signifies method for the research and appraisal of electronic low-frequency noise and module.