Research On Reliability Prediction Mechanism For High-Power LEDs

Solid State Lighting (SSL) is recently a very promising and government supported industry. High-power LEDs are the key devices for SSL. The improvement of the luminous efficacy and reliability of high-power LEDs is critical for SSL. Studies on the LED reliability are the basis of the improvement of the LED reliability. This thesis researches on the reliability prediction mechanism for high-power LEDs. A new method to realize the quantitive prediction for the LED reliability without long-term life test is proposed. It can be used for the early failure screening and product quality management of the LEDs, and is academically valuable and practically useful to the LED scientific research and industry.A reliability prediction mechanism for high-power LEDs based on the characteristic curves is developed. The mechanisms and measurement methods of the characteristic curves—electrical derivative curve, light output curve, spectral power distribution curve and thermal curve, which are correlated to the reliability of the LEDs are introduced. The characteristic parameters—ideal factor, series resistance, light output saturation, current temperature dominant wavelength shift, current temperature color difference, junction temperature and thermal resistance are derived from the characteristic curves, respectively. The relationship between the characteristic parameters and the reliability of the LEDs are verified by experiment. An artificial neural network is employed to model the quantitive relationships.The measurement system for the characteristic curves of high-power LEDs is setup. The measurement systems for the electrical derivative curve, light output curve and spectral power distribution curve are integrated in a whole system. The thermal resistance measurement system is separated. Different measurement methods are proposed according to the requirements of the measurement for the characteristic curves. The problems of the nonuniform spatial response of the integrating sphere, the spectral mismatch of the detector, the stray light and bandwidth of the spectrometer are analysed. Improved designs and correction algorithms are proposed. The uncertainties in the measurement for the characteristic parameters are analysed. The analysis results show that high precision instruments and strict uncertainty control in the measurement are required, for a precise measurement for the characteristic parameters. The analysis results are used as the guide to the instrument design and uncertainty control in the measurement procedure.Reliability experiments are carried out for both high-power GaN based blue and phosphor converted white LEDs. The reliability prediction mechanism is setup based on the experiment results. The average prediction errors of the 700mA accelerated lives and color degradations are: blue LEDs' lives 23.6%, white LEDs' lives 15.9%, blue LEDs' dominant wavelength degradation 11.1%, white LEDs' color degradation 17.2%. The analysis for the risks of the prediction shows that the prediction exclusive of the time factor is of great risks, due to the changes of the activation energy caused by the environment and application conditions. The robustness of the reliability prediction mechanism is challenged by the boundary problem of the failure mechanisms. The application of the reliability prediction mechanism is also limited. A simplified screening method for high-power LEDs based on the thermal resistance and ideal factor is proposed, which can be used for the early failure screening in the LED production. The failure mechanisms of the aged LEDs are analysed based on the characteristic curves. Electrostatic test is also carried out for the LEDs.Long-term life test is carried out for high-power phosphor converted white LEDs. A new life test system based on spectroradiometric measurement is designed. A more precise Eyring model is employed for the accelerated life test. The life test results are: the average accelerated life under 500mA is 3875hours(U=0.6%, k=2), and the estimated average normal life under 350mA is 12628 hours (U=6.3%, k=2). The uncertainties in the life extra interpolation and acceleration in the life test are analysed. The color failure modes and failure mechanisms are analysed.