The Measurement System Design For The Low Frequency Noise Of OCDs And Study On Analysis Approach

Optoelectronic Coupled Devices (OCDs) are widely used in areas which meetrequirement of high reliability. So it is a subject of great significance on theory andapplication to study on a screening method to estimate the quality of devices.There are some conventional screening methods such as life test, ultimatestress test, environmental test and destructibility physical test. But these, especiallyultimate stress test, are expensive, slow, high limitation and destructibility. Thedevices under those tests should be rejected because of damage. Therefore only thedegree of unavailability, average life and failure mode can be statistically analyzedbased on the test results of a batch of devices. But the devices with defection can'tbe exactly screened from the faultless devices. There is another method like checktest which is nondestructive and unambiguous for each device, but it can't screenthe semiconductor devices and material with inherent defects.The noise of OCDs is a critical indication of sensitivity and stabilization ofdevices and a sensitive parameter of quality and reliability of devices. There issome basic noise in every device, but the noise will increase when defections liedon Si-SiO2 surfaces, P-N junction or channel et al. The last one is called excessnoise which is relative to the different faults in devices. Thus through the amplitudeof the excess noise such as excess 1 f noise, excess G-R noise and burst noise, itis possible to analyze the quality of devices and estimate the reliability of devices.As the development of the mechanism of low frequency noise and the research oftest approach, the analysis and test of low frequency noise is a significant measurefor reliability estimation, which is quick, simple, nondestructive and specific.There are some instance of imperfect and shortage in the measurement systemin existence. The measurement system is expensive because of the CF920 worthhundreds thousand RMB which is a FFT analysis apparatus. The measurementsystem is not flexible enough. In respect that not accurately analyze the componentof the low frequency noise, the value of Vn (1 Hz) approximately indicate 1 fnoise and the value of V n (1 0Hz)Vn(1Hz) approximately indicate G-R noise by thereliability class judgement criterion of semiconductor devices. Personal error mustbe leaded in for judgment of burst noise.A new acquisition, test and analysis system for low frequency noise of OCDsis built on LabVIEW software based on the reliability estimation method ofsemiconductor devices with the algorithm of crosscorrelation and FFT.In this paper above all the imperfects of traditional screening method areanalyzed and the feasibility of analysis inherent defects of devices with amplitudeof excess noise is approached.In the next place burst noise will be detected by wavelet based on the analysiswith character of burst noise and wavelet, as an effective supplement for parametersestimation. Wavelet detecting program is programmed with wavelet toolkit inLabVIEW software which can exactly detect burst noise. This part avoids thepersonal error, advances the automation level of measurement system, supplies theparameters estimation and detects burst noise real time. The follow conclusions aregained through abundant test. Quadrature mother wavelet with compact support candetect burst noise, but the effect of mother wavelet with high order vanishingmoments is better than mother wavelet with low order vanishing moments. Thedecomposition effect is satisfying when decomposition level equals to 4. If thedecomposing level is too high some useful singularity will be filtered, but if thedecomposing level is too low relatively more singular point can't be detectedexactly. The test effect relate to both mother wavelet and decomposition level.Decomposition levels with the most satisfying effect and most completeinformation are the third detail and the forth detail. The forth decomposition detailof the Daubechies2 wavelet is the best effect.And then the parameters estimation based on Genetic Algorithms is discussedthrough which the parameters A, B, C i, f oi and N can be precisely estimate. It istestified that the analyzed result of noise spectrum is exact, which is receivedthrough the algorithm above mentioned, for that the estimation is accurate and thecurve fit well. In accordance with the actual problem, it is adopted in algorithm thatreal number coding mode, wheel of fortune choosing mode, simple point crossingmode and heterogeneous variance mode. After abundant testing, it is gained that thesuitable maximum of generation is 80. If the number of generation is too big thetime of program will increase but the adaptability advance inconspicuous, and if thenumber of generation is too small it is easy to generate infeasible solution.At last the blue print of whole measurement system of low frequency noise ofOCDs is discussed based on the existed measurement system of semiconductordevices, and the system is built on LabVIEW software. This system can acquire thesignal from the two preamplifiers by the DAQ device of NI Company and processthe signal with the controls of LabVIEW. The system still can detect the burst noisewith wavelet toolkit and estimate parameters. This system enhances the automationlevel for the function of real time display, document save, database establishing andreport automatic generation.