Report on a second round robin measurement of the thermal conductivity of CVD diamond

Abstract The results of a round robin series of measurements of the thermal conductivity and thermal diffusivity of chemical-vapor-deposited (CVD) diamond are reported. The present round robin (RR2) is an outgrowth of an earlier series (RR1) aimed at understanding which measurement techniques are best suited to measuring the high conductivity of CVD diamond. RR2 includes diamond specimens of higher homogeneity, non-diamond specimens of lower conductivity as controls, and measurements by 14 laboratories using five techniques, with all techniques but one used by more than one laboratory. The data are converted to thermal conductivity at 25 °C to facilitate comparisons among specimens and laboratories. The statistical analysis excludes outliers using Chauvenet's criterion, resulting in data from typically two or three laboratories being excluded for each specimen. The analysis arrives at mean values in the range 1300–2000 W m −1  K −1 for the diamond samples, as expected because of the conditions of preparation, with uncertainties in the range 1.5–4%. For the non-diamond materials, SiC and AlN, the results are 268 W m −1  K −1 ±2.2% and 178 W m −1  K −1 ±2.5%, respectively. Comparing techniques, the d.c.-heated bar is found to be the most accurate, typically ±5% or better. Under favorable conditions, Angstrom's thermal wave method can apparently yield relative uncertainties of ±5–10%, and the mirage effect ±5–15%. Too few laboratories used the transient thermal grating and laser flash methods to make general comments on their accuracy, but one expects a comparable accuracy if used on specimens that are fine-grained and thermally isotropic. Deviations, sometimes large, from these optimal accuracies are examined, and it is suggested that certain experimental details are important for achieving accuracy. It is also found that the estimates of experimental uncertainty provided by most laboratories seriously underestimate the actual deviations of their data from the mean conductivity.