Online proceedings - EDA Publishing Association

24-26 September 2008, Rome, ItalyAccording to the above definition, it is formally notpossible to define a thermal resistance between two points,e.g. die and case. In other words, R th die-case is only correctprovided:• The die and case surfaces are at uniform temperature• We know the heat flux between die and caseRegarding the first bullet point: except for high-powerLEDs (e.g. > 3 W) the assumption of a uniform dietemperature is correct. It is the case surface that causessevere problems because the heat spreader (or alternativelythe board) cannot be considered to be at uniformtemperature. The consequence is that the measured casetemperature becomes dependent on the heat transfercoefficient h that describes the rate of heat transfer from theheat sink to the environment, usually including bothradiation and convection. The second bullet point is usuallyalso met, but should be checked in case some heat is leakingaway through the optics, either by radiation directly from thesource or by conduction and convection/radiation from thetop surface. However, it should be stressed that in contrast toincandescent lamps the corrections involved are of secondorder.II.HISTORY OF IC PACKAGE STANDARDISATIONIt is always instructive to look back and see if we canlearn from the mistakes made in the past. In the field ofthermal standardisation we see a remarkable parallelbetween what happened in the semiconductor world andwhat is happening right now in the LED-world.Around 1990 it became clear that thermal characterisationof IC packages was a chaos. Manufacturers all over theworld used different standards, and even within a singlemanufacturer’s environment intolerable differences showedup, as Table 1 nicely demonstrates.TABLE 1: Round robin data for a variety of DIL packages (Philips, 1991)Various DIL packages were sent to three different sites of alarge semiconductor manufacturer, with the question toreturn thermal data according to the standard that was in use.The returned numbers clearly indicate the problem at hand,the differences being up to 100%.While the manufacturers were complaining about the lackof well-defined standards, users started to complain, for acompletely different reason, that they could not use thepublished data for their application.A. Why end-user rebellion started around 1990Apart from the reasons why standards were lacking endusersdid rarely raise their voice before 1990. The mainreason this situation changed was because a new tool arrivedthat made it possible for designers to predict the thermalbehaviour of their products in a specific application:Computational Fluid Dynamics (CFD). The accuracy thatcan be expected from these tools is not discussed here (seee.g.), but in principle the only data that are lacking for whichthe end-user has no responsibility are the component thermaldata provided by the manufacturer. From this point of view itis obvious that the accuracy that can be reached is ultimatelylimited by the information in the data sheets. End-usersstarted to realise that it did not make sense to spend a lot ofmoney in software and training to facilitate virtualprototyping only to be seriously limited by the lack ofreliable information in the data sheets.However, it is one thing to say: we cannot use the datasheets; it is quite another thing to formulate what is reallyneeded. Because the use of user-friendly CFD tools in theelectronics industries started in Europe, end-users in Europestarted also the initiative to combine efforts to try to answerthe question what they really needed. These needs wereclearly identified as early as 1991 resulting in a Europeanproject proposal championed by Philips and finally lead tothe launch of the EU-funded DELPHI project in 1993. Whenthis project was finished, the members (all end-users) had aclear idea about what they really wanted, and they invitedthe three largest semiconductor manufacturers in Europe tojoin the team to investigate if the manufacturers could indeedrealise what the end-users wanted (SEED, 1996). SEED wasfollowed by PROFIT (2000) in which new membersparticipated, focusing mainly on the extension to thetransient domain and on trying to get the methodsstandardised. From 1990 onwards, European membersplayed an active role in the JEDEC JC15.1 subcommitteethat mainly concentrates on the standardisation of thermalcharacterisation of IC packages.A major breakthrough in thermal characterisationemerging from DELPHI was the notion that we needed aboundary between the responsibilities of the manufacturersand the end-users:©EDA Publishing/THERMINIC 2008 210ISBN: 978-2-35500-008-9