Research And Theory Of1/f Noise Detection Methods In Semiconductor Devices

Semiconductor is the most basic unit of all kinds of electronic equipment; itaffects the stability and safety of electronic equipment by quality andstability.Especially when it comes to area of high cost and high risk like aerospaceand nuclear, any defects could cause economic losses, even threatens personal safetyof staffs.With the constant improvement of reliability requirements, how to accuratelydetect the merits of the semiconductor devices is becoming theoretical and practicalsignificance, it is also a study field for many scholars.‘1/f noise’ is one of theimportant indexes to measure the quality of semiconductor devices.The paper takeszener diode as the experimental object and brings several new analysis methods forlow-frequency noise which are more stable. These methods can estimate the1/f noiseparameters accurately and also can set up a test system for both software andhardware. They provide theoretical support and physical implementation for thereliability evaluation of the semiconductor devices. The specific research in thisarticle is as follows:1) First, clarify the relationship between low frequency noise and reliability ofzener diode; introduce the noise characteristics of semiconductor device that qualitydefects generated.2) Building the hardware platform of low-frequency noise test system, anddevelop the corresponding softwar system. The system mainly includefrequency-domain analysis, time-domain analysis and data acquisition.3) In the time-domain analysis, apply robust estimation to estimate the1/f noisefrequency factor base on the wavelet analysis, and use the bionic optimizationalgorithm for robust estimation equation optimization, comparative advantages anddisadvantages of several kinds of method.4) In the frequency-domain analysis, there are three ways: The first method is theclassical Fourier transform cross-spectrum method repetition; The second method is used based on adaptive threshold denoising algorithm of wavelet packet analysis tocalculate the denoising signal power spectral density,1/f noise frequency byparameter fitting index; The last one is to obtain the fourth-order spectrumone-dimensional slices of the measured signal by the cross fourth-order cumulants ARmodel method. Then estimate the index of the1/f noise frequency.