Individual prediction of the reliability of high power transistors for electronic devices of medical purposes

• Main Authors: S. M. Borovikov, V. O. Kaziuchyts
• Format: Article
• Language: Russian
• Published: Educational institution «Belarusian State University of Informatics and Radioelectronics» 2021-02-01
• Series: Doklady Belorusskogo gosudarstvennogo universiteta informatiki i radioèlektroniki
• Subjects: semiconductor devices; reliability; individual forecasting; informative parameters; majority logic method
• Online Access: https://doklady.bsuir.by/jour/article/view/2995

When assembling electronic complexes for medical purposes, it is important to install highly reliable semiconductor devices in electronic equipment. Experimental studies and the example of high-power bipolar transistors in this work show how you can select copies of an increased level of reliability for their subsequent installation in critical electronic devices. To select highly reliable samples, individual forecasting was used according to informative parameters measured for a particular sample at the initial moment in time. Experimental studies (training experiment) included measuring at the initial moment of time for each sample of transistors of electrical parameters, which may contain information on reliability, and then conducting accelerated tests of transistors for reliability for a time corresponding to normal operating conditions specified in the technical documentation. The training experiment is performed once and used to obtain a predictive rule, which is applied to other similar samples that did not participate in the training experiment. To obtain a predictive rule, the method of majority logic was used. Prediction is performed in the form of assigning a specific sample to the class of highly reliable samples for a given future operating time. To perform prediction, the values of the informative parameters are measured at the initial moment of time for a particular sample of interest, they are converted into binary numbers (zero or one) using the threshold values found from the results of the training experiment, and the decision on the correspondence of the sample to the class of highly reliable transistors is made by a set of binary numbers. To classify a sample as a highly reliable one, it is sufficient that the number of ones exceeds the number of zeros in the resulting set of binary numbers.