Speaker
Description
The aim of the presented research was the development of a non-destructive method for automatic verification of the operating temperature of a semiconductor device. The primary objective was to correctly classify a device into one of two categories: packages with a junction operating at an excessively high temperature and packages with a junction operating within the acceptable temperature range. In addition, the proposed algorithm was required to detect and indicate local areas of the package surface exhibiting elevated temperatures relative to the surrounding regions. The investigated component was the MBR2550CT device, consisting of two semiconductor diodes with a common cathode enclosed in a TO-220 package. The core of the experimental setup was an ImageIR 8300hp infrared camera (Infratec, Dresden, Germany), featuring a spatial resolution of 640 × 512 pixels, a detector pitch of 15 μm, a spectral range of 1.5–7.5 μm, and a noise-equivalent temperature difference (NEDT) of 25 mK. The camera enabled image acquisition at 335 Hz in full-frame mode or up to 5000 Hz using subwindowing. The system was complemented by a current source, ammeter, and voltmeter, while the camera was mounted on a tripod and operated using IRBIS 3.1 Professional software. To validate the thermographic measurements, a Pt1000 temperature sensor (Vishay Intertechnology, Malvern, PA, USA) in a four-wire configuration was attached to the observed package area. The sensor was housed in an SMD 0805 package and glued using WLK 5 adhesive (Fischer Elektronik, Germany) with known thermal conductivity. During the experiments, currents of 1.25 A, 2.5 A, 3.5 A, 4.35 A, and 4.88 A were applied to one diode junction. For each current level, 100 thermograms were recorded. Measurements were performed on six devices, resulting in a total of 3000 thermograms. To establish the relationship between package temperature and junction temperature, an additional MBR device (not used in the main experiments) was physically opened, allowing precise internal measurements. This enabled the creation of a three-dimensional model and thermal simulations in the SolidWorks environment. As a result, the junction temperature Tj was estimated based on the thermographically measured case temperature Tc. The junction temperature of 70 °C was arbitrarily selected as the acceptable operating limit, corresponding to a case temperature of Tc = 58.9 °C. Using the Arrhenius model, it was shown that reducing the junction temperature from 150 °C to 70 °C decreases the aging rate by approximately 169 times (AF ≈ 0.0059 for Ea = 0.8 eV). Based on these thresholds, the thermograms were divided into two classes: correct operation (Tj < 70 °C, Tc < 58.9 °C) and incorrect operation (Tj ≥ 70 °C, Tc ≥ 58.9 °C). The resulting dataset consisted of 1500 thermograms for each class. Temperature matrices were extracted and processed, and the RetinaNet algorithm achieved a classification accuracy exceeding 90%.
