On the Reproducibility of Thermal Measurements and of Related Thermal Metrics in Static and Transient Tests of Power Devices

Abstract

Traditionally the thermal behavior of power devices is characterized by temperature measurements at the junction and at accessible external points. In large modules composed of thin chips and materials of high thermal conductivity the shape and distribution of the heat trajectories are influenced by the external boundary represented by the cooling mount. This causes mediocre repeatability of the characteristic RthJC junction to case thermal resistance even in measurements at the same laboratory and causes very poor reproducibility among sites using dissimilar instrumentation. The Transient Dual Interface Methodology (TDIM) is based on the comparison of measured structure functions. With this method high repeatability can be achieved although introducing severe changes into the measurement environment is the essence of this test scheme. There is a systematic difference between thermal data measured with TDIM method and that measured with temperature probes, but we found that this difference was smaller than the scatter of the latter method. For checking production stability, we propose the use of a structure function-based Rth@Cth thermal metric, which is the thermal resistance value reached at the thermal capacitance belonging to the mass of the package base. This metric condenses the consistency of internal structural elements into a single number.

A Novel Evaluation Method of Thermal Transient Measurement by Regularization

“…For the system identification purposes, the time constant spectrum R (z) has to be computed from (5) given the function determining its right hand side, the measured transient response a(z). Mathematically, equation ( 5) is the most iconic example of an ill-posed problem, the Fredholm integral equation of the first kind.…”

“…It should be clear that formally expressing the solution to (5) in the form (8) does not alleviate the ill-posed nature of the problem. Indeed, in the context of ( 8) the illposedness manifests itself through the fact, that owing to the smoothness of the convolution kernel (6) its Fourier transform appearing in the denominator of (8) will be diminishingly small for large frequencies Φ.…”

“…that is, in the standard ℓ 1 -regularized constrained least squares formulation of the equation (5). We note that one needs to seek R(•) in the space of measures to guarantee that the infimum in ( 9) is attained, which is of course consistent with (4).…”

“…One of the important research perspectives is to investigate the thermal transient responses. The typical applications are such as to obtain the thermal model parameters [5], [6], identify the inner structure [7], [8], analyze failure modes non-destructively [9], and estimate the junction temperature [10]- [12].…”

“…method, the framework of the NID method has been successfully standardized in [17] and commercialized in [18]. The majority of publications about the NID method are focused on a variety of its applications [5], [6], [9], [10], [14]. The limitations or challenges of the method itself are rarely discussed.…”