Testing Thermal Response of Diodes, Chapter 1
If there is a single most important thing that I can tell you about Thermal Response testing, it is this: know your product and set your own test limits accordingly!
Industry wide specifications are certainly useful and set a standard that no credible product should fail to meet. However it is entirely possible that a bad part by one manufacturer (process/design) may test better than a good part by another.
How could that be?
The thermal properties of a given diode design are limited by the geometry and metallurgy of the diode package and the semiconductor chip within it. Given a certain design, if all of the parts are assembled perfectly and perfectly bonded together, the part will have predictable thermal properties.
A part designed with a slightly different geometry or different metal or alloy in any of its parts would have different thermal properties if processed perfectly.
Both parts may be suitable for the application and be reliable if assembled perfectly (or nearly so). Suppose the measured response of a particular part from the "best" design varies significantly from its theoretically perfect reading. It is highly likely that there is a poor bond within that part.
Parts that have detectable imperfect bonds early in their life have a much higher probability of premature failure!
A part from a design with a slightly poorer initial thermal performance, but perfectly bonded (or nearly so) will be much more likely to survive the stress of repeated heating and cooling that diodes are subject to in typical applications.
What to do
Thermal Response testing with carefully selected test conditions and limits can detect defects quickly and non-destructively. Before getting into details of setting up a test program, I would like to share the following typical case history that I have experienced several times in my years of testing Thermal Response.
A manufacturer is experiencing excessive returns of rectifiers with catastrophic failures. The parts are cut open to reveal poor bonding of the chip or cell to the package. If no Thermal Response tester is available, the engineers try various tests such as high current SURGE pulses or tight VF limits. It doesn't work!!
We supply one of our Thermal Response testers and help our customer to evaluate the product. Often we pick out parts that have fairly small deviations from the product norm. These parts are opened up and, sure enough, bad bonds.
But this is only part one of the story. Now that the means of evaluating the product is available, the manufacturing process can be adjusted and the results monitored. The optimum process settings can be found and in the end the process norm is shifted significantly toward the best that the design will allow. Problem solved!
Finding the bad parts. A quick overview
(long-winded justification to follow)
Make sure all the samples are good parts electrically.
Electrical rejects may produce unreliable thermal readings which will mess up your statistics. Eliminate the electrical rejects FIRST.
Use DVF, not THETA
THETA reads in Degrees Celsius/Watt and DVF reads in Delta VF. DVF works better.
Find the optimum SAFE value for measurement delay (TMD)
Use the plot function "Cooling Plot" for every test program and for every part number and (if applicable) from every manufacturer. This allows you to select a low TMD but not so low as to be invalid.
Use a higher rather than lower IM
10mA is a good general purpose value although some large slow rectifiers may need more and some published specifications call for less. For internal statistical analysis, you can choose.
Collect data to plot histograms at several pulse widths (TH)
I would suggest 10mS, 50mS, and 250mS. Bad parts may show up best at one or more of these values.
Use a test current at each pulse width that produces a DVF of roughly 50mV to 100mV
For silicon junction rectifiers that would be a delta temperature of roughly 25 to 50 degrees Celsius. Schottky rectifiers would get a bit hotter but still ok.
They may ALL be bad (sort of)
If you are suspicious of this population already, consider the possibility that although some parts are worse the best ones could be better.
ELECTRICAL TESTS FIRST
This is intended to apply to statistical analysis of Thermal Response, not overall acceptance testing. If you include Thermal Response in a test program including electrical tests, you could test Thermals first if you want. This would ensure that in the unlikely event that the part was damaged by the Thermal test, it would be rejected by the following electrical tests (especially the IR test).
Please keep in mind that the Thermal test leaves the junction temperature somewhat higher than the ambient. If the pulse width is on the high side, such as 250mS, it make take a few seconds to cool. The 10mS end of the range would of course cool much faster.
DVF vs THETA
If your final test is controlled by some published specification requiring reading Degrees/Watt then you should use THETA. You can still do your internal analysis as DVF and establish correlation between the DVF limits you establish and the THETA limits in the final program. You will find the correlation pretty good but not perfect and here is why.
At any fixed operating current, a rectifier will get hotter if the VF is higher. This will make the DVF reading higher also. The THETA test compensates for this difference by calculating the rise in temperature per Watt.
However a part operating at a higher temperature is more likely to fail! Also within any given lot of parts made by the same process, the part with the high VF is more likely to have a defect such as poor bonding or a cracked chip. Why compensate for that?
Actually if the part is a zener diode which would normally operate in the "Reverse" direction, there may be some justification for this compensation. You decide.
Find the best TMD
Everyone would like to use the lowest possible value for TMD. The reason for this is that the junction begins to cool immediately as the heating pulse current is turned off. The DVF measured will be slightly (to considerably) lower than the true value depending on the value of TMD used.
However, the lowest value that can be used is not constant. It is effected by several properties of test equipment which vary with the heating current (IH), the measurement of current (IM) and even the wiring to the test fixture and the fixture itself.
In addition, the lowest safe value is also effected by the carrier recombination time of the diode being tested. Fast Recovery diodes can usually use a lower value than the standard recovery types.
If the value used for TMD is marginally too low the results may appear reasonable but in fact be false and misleading.
Be safe. Always use the Cooling Plot function which will plot the Thermal test against TMD with the other test conditions programmed to the values you will be using. The readings plotted for moderate to long TMD values will be correct. As the TMD gets smaller the readings will go up in a smooth curve. At some point the curve may depart from this smooth predictable path. Use a TMD which is safely higher than this point.
Generally you can use a lower TMD if you use a higher IM (within reason). 10mS is a good general purpose IM for most small to medium power diodes. Large, slow rectifiers may benefit from a much higher IM (perhaps 100mA).
Try several pulse widths
I assume that you are trying to detect the bad bonds that can lead to premature diode failures. Diodes will have two (or more) internal bonds at various physical distances from the junction. The amount of time needed for heat generated at the junction to pass through a given bond is related to that distance.
There are many ways to mount a silicon chip within a diode package and so the distance to each bond may vary from part number to part number and from manufacturer to manufacturer.
You could use a very long pulse to be sure that the heat passed through all of the bonds, but that is not necessarily the best. If the problem bond is very close to the junction a long pulse will heat the diode structure well beyond the bond and give a higher reading. This higher reading not only dilutes the small variations due to the problem bond but may also be effected by other normal variations not related to the problem.
Collecting data for several pulse widths allows you to find the optimum width for the particular diode design being evaluated.
Selecting the right heating current (IH)
Since the variation in DVF that you are looking for may be fairly small, you would like to have a reading that is large compared to the normal repeatability errors in the equipment. A reading of 100mV is well out of the noise and is a good choice. Silicon junction rectifiers have a K factor of roughly 0.5 degrees per mV (we express it as 1/K which would be 2.0mV per degree).
The current (IH) needed to get a DVF or 100mV will be higher for low values of TH and vice versa.
A delta temperature of 50 degrees Celsius is a good safe value. That would give a final junction temperature of 75 degrees if the ambient is 25 degrees. You could go a bit higher if necessary, but avoid very high temperatures as they could distort the data or damage the diode.
Also keep in mind that Schottky diodes generally have lower K factors.
Analyzing your histograms
Let us assume that you are already suspicious of the quality of the sample being tested. I am assuming also that you have removed any electrical rejects.
If the sample contains good parts with only "normal" manufacturing variations you should see a "normal" bell curve with a Sigma of (dare I say?) well, not too many %. If you see "flyers" on the high side, these are probably parts with significantly bad bonds.
Flyers on the low side however show a bigger problem. For any given diode design, it is impossible for the process to be more than 100% perfect (of course). If there are flyers on the low side, these are probably parts that are at least approaching perfection and the main distribution is centered on something that well let's say could use some improvement.
We at FEC have been through this process a number of times over the years and we stand ready to offer assistance if you wish to call on us.