Online proceedings - EDA Publishing Association

24-26 September 2008, Rome, ItalyEvaluation of short pulse thermal transientmeasurementsV. SzékelyBudapest University of Technology & EconomicsDepartment of Electron Devicesszekely@eet.bme.huAbstract-Thermal transient recording and the time constantspectrum analysis are widely used methods in the testing andqualification of IC packages. A limitation of these methods isthat recording of the complete transient response requires longtime. The paper offers algorithms to evaluate short pulse andshort time measurements. These methods are suitable if onlythe extraction of the short time constants are needed. This is thecase if the transient method is used for die attach qualitychecking.I. INTRODUCTIONThermal transient recording and the time constantspectrum analysis are more and more accepted and usedmethods in the testing and qualification of IC packages. Themethod is essentially as follows. First the response for a stepfunction excitation is recorded. A deconvolution step leadsto the time-constant spectrum. The latter can be transformedto the structure function [1],[2]. The time-constant spectrumand/or the structure function are highly suitable to recognisethe details of the heat-flow, to extract partial thermalresistances and capacitances. The displacement of the timeconstant lines may indicate the changes of the interfacethermal resistance [3] etc.Although this procedure has been working very well formany years, some limitations can be recognised during theproduction testing. First of all, recording of the completetransient is needed for the correct evaluation, even in case ifwe are interested only in the short-term time constants. Thisrequires excessive amount of time since the time to reachsteady state (t ss ) is in the range of 5-20 minutes. This isespecially disagreeable if the cooling curve is recorded sincefirst we have to wait for achieving the heated steady state(t ss ), after this we have to wait for the recording of thecooling curve. This leads to 2 t ss as total time.Another constraint is that if we are waiting until the steadystate, the power driving level has to be limited in the steadystate rating of the DUT. In order to detect time constantswith small amplitudes it would be advantageous to applypower overdriving, but if we wait to t ss this is obviouslyinhibited.A possible solution for both of the above mentionedproblems is to apply driving power pulses of short duration.There were some attempts for this in the last decade. In aprevious paper [4] we demonstrated that if we record thecomplete answer for a single short power pulse, the entireheating curve could be restored from this recording. Thismeans that all the information carried by the heating curveare present in the short pulse answer as well. Masana hasreported a method [5] to improve the answer if both thepowering and the recording are short. This improvement isbased on the estimation of the first time constant not coveredby the measurement time.In this paper we are dealing with further possibilities todiminish the time requirement of transient measurements,together with the exact evaluation of short pulsemeasurements.II. EVALUATION OF SHORT PULSE, LONG RECORDINGMEASUREMENTSIn this proposed new measurement strategy we apply arelatively short power pulse to the DUT and then wemeasure the complete temperature transient started at thetrailing edge of power pulse (see Fig. 1). In this way• the driving power level can be increased significantly,• the cooling curve measurement is applied which ismore advantageous in many view-point,• the required time is only one t ss instead of two.Fig. 1. Pulse driving and subsequent transient recordingSince the applied pulse has to be short the heat wavefrontdo not have time enough to reach the farther regions of theDUT. This means that the large time constants of the systemare only weakly excited and have small effect in thetemperature response. So, our expectation is that the largetime constants will appear in the time constant spectrumonly in an attenuated form. Obviously the question arises:how can one calculate the exact amount of this attenuation?Having this knowledge the systematic measurement errorcan be (at least partly) compensated.Let we suppose that the thermal system under test can bedescribed by a finite set of time constants τ i and the©EDA Publishing/THERMINIC 2008 20ISBN: 978-2-35500-008-9