Thermal data supplied by manufacturers and used as a benchmark in a co
mparison of the thermal performance of different packages must be trea
ted with caution, as the results are specific to the measurement envir
onment of the supplier. For this reason, calculations of expected comp
onent thermal performance based on these data can be erroneous. This p
aper is an analysis of the discrepancies observed in thermal test resu
lts owing to small changes in the exact implementation of the measurem
ents and in the calibration of the device temperature-sensitive parame
ter (diode V-f). It concludes with recommendations for ensuring more a
ccurate and reproducible thermal resistance measurements. (C) 1998 Pub
lished by Elsevier Science Ltd. All rights reserved.