Prediction of IGBT junction temperature using improved cuckoo search-based extreme learning machine. (September 2021)

• Record Type: Journal Article
• Authors: Liu, Boying; Chen, Guolong; Lin, Hsiung-Cheng; Zhang, Weipeng; Liu, Jiaqi
• Is Part Of: Microelectronics and reliability. Volume 124(2021)
• Publication Date: 2021-09
• Subjects: Failure characteristic parameters -- Junction temperature prediction -- Cuckoo Search -- Extreme learning machine; Electronic apparatus and appliances -- Reliability -- Periodicals; Miniature electronic equipment -- Periodicals; Appareils électroniques -- Fiabilité -- Périodiques; Équipement électronique miniaturisé -- Périodiques; Electronic apparatus and appliances -- Reliability; Miniature electronic equipment; Periodicals; 621.3815
• DOI: 10.1016/j.microrel.2021.114267
• Languages: English
• ISSNs: 0026-2714

Abstract

Abstract: The insulated-gate bipolar transistor (IGBT) is one of the most widely used power transistors in switching and industrial control systems. Its actual junction temperature plays a critical factor in determining the dynamic performance, reliability and life-time of the device. Although some noninvasive measurement methods such as optical and physical contact methods may be used to estimate the junction temperature, the measurement accuracy is very sensitive to the measured position. Therefore, the prediction using cuckoo search-based extreme learning machine for junction temperature is developed to reach a high-accuracy solution without measured-location sensitivity. Firstly, the accelerated aging and single pulse tests in IGBT are implemented to collect the IGBT failure related parameters, e.g. collector-emitter saturation voltage ( V CE(sat) ), junction temperature, collector current ( I c ) and the aging cycles number. With the curved surface fitting for the collected data, the relationship between the junction temperature and the other parameters can be formed. Based on the extreme learning machine optimized by the improved Cuckoo Search method, called ICS-ELM, V CE(sat), I c and the aging cycles number are taken as input, and the output is the predicted junction temperature. The performance results reveal that the determination coefficient (R 2 ) by ICS-ELM model achieves the optimal value, i.e. 0.9975, which is superior to the curved surface fitting method, Cuckoo search optimizing extreme learning machine, support vector machine and extreme learning machine. Highlights: The junction temperature prediction using ICS-ELM model is insensitive to the measured-location. The relationship between the junction temperature and VCE(sat), Ic and the aging cycles number can be formulated. The determination coefficient (R2) by ICS-ELM model can achieve the optimal value.