ceramics are increasingly used in packages and printed circuit boards, because they have a number of advantages over plastics: much higher thermal conductivity, possible match of the coefficient of thermal expansion, and hermetic sealing.

unfortunately, the cost of ceramics is still much higher than that of plastics, prohibiting their use in low-cost high-volume products. nevertheless, their use in future products seems unavoidable, given the trends in the semiconductor industries regarding the expected much lower thermal resistances of the packages and the concurrent design of packages and boards to facilitate heat spreading.

the problem with the thermal conductivity of ceramics is the dependence on the composition, grain size, and manufacturing process, which make it rather difficult to obtain a reliable value from literature only. looking at the values quoted in various handbooks, papers and data sheets, two things are observed. 1) large variations exist, and 2) many authors seem to copy values from the same, but untraceable sources.

an intriguing example is aluminum nitride (ain), an increasingly popular ceramic. the value most often quoted is around 180 w/mk. however, in reference 1 some interesting data can be found in a graph in which thermal conductivity values from seven manufacturers are displayed as a function of temperature. the highest value found at room temperature is 200 w/mk; the lowest is 80 w/mk. furthermore, these values drop by more than 30% going from 20° to 100°c.

the same drop, by the way, is observed for other ceramics such as beo and al₂0 ₃. (see also the technical data section of the previous issue.)

the table shows the range of values at room temperature for a number of frequently used ceramics from various sources. note that the wide variation in the values presented for bn is caused by different modifications; for example, the highest value is for cubic boron nitride.

in summary, the published thermal conductivity values of ceramics are ideally suited to match experimental results to numerical simulations, and i am afraid that this is exactly what happens in practice. measurements in-situ, using well-designed experimental benchmarks, are the preferred way to get accurate values.

reference

1. r. dindwiddie, advanced electronic packaging materials, vo1.167, boston, 1989.

clemens j.m. lasance
associate editor
philips research laboratories