IGBT Linear operation

[Randy222]

This is more of a "specs" comment than a "solution" comment.

Specs of the items are almost always come from modeling or of the package. It's rare to see a manufacture do actual testing when it comes to thermal controls.

"watts" almost always means it's a temp related issue. But we don't see maker mounting the package to a 1ton block of aluminum (or ideal finned sink) at room temp to sink heat out during the "max" test.

IIRC, few if any show graphs for "max". It's temp that kills the device, but you could exceed a max number if the time there is very small, because it takes times for temp changes, etc.

We also don't really see spec sheets calling out heatsink specification numbers, they kinda leave the thermal design up to you.

[T3sl4co1l]

This is more of a "specs" comment than a "solution" comment.

Specs of the items are almost always come from modeling or the package. It's rare to see a manufacture do actual testing when it comes to thermal controls.

Do you have a citation for this statement of fact?  Example, have you worked at one and have read process documents to that effect?  Or you've seen appnotes which document it?

"watts" almost always means it's a temp related issue. But we don't see maker mounting the package to a 1ton block of aluminum (or ideal finned sink) at room temp to sink heat out during the "max" test.

Well... no... why would they?  There are much better ways anyway.

The typical standard is nucleated boiling liquid, a bath of Freon stirred aggressively and held at a pressure such that it's just boiling at ambient temperature.  This holds Tc, very literally every exposed point of the case, at 25°C for the test.

This itself is not well documented, to the consternation of many; but IR did at least release an appnote during the peak of the specsmanship era when this started really getting pushed (and, AFAIK, is still the accepted standard outside of SMT devices; though many tabbed SMTs still also have a useless Tc=25°C spec anyway).

Namely, AN-1140: https://www.infineon.com/dgdl/Infineon-ApplicationNote_MOSFET_Continuous_Current_Rating-AN-v01_00-EN.pdf?fileId=5546d462689a790c0169166671e9454c

IIRC, few if any show graphs for "max". It's temp that kills the device, but you could exceed a max number if the time there is very small, because it takes times for temp changes, etc.

We also don't really see spec sheets calling out heatsink specification numbers, they kinda leave the thermal design up to you.

I...guess?  Not sure what max you mean.  SOA plots don't change much with temperature.  (Tempco and thermal resistance change a little with temperature, which can change size of the instability region.)  There's transient thermal resistance for pulses.  Naming a particular heatsink wouldn't be very useful, though one might argue SMT footprint data (minimal, thermal pours of given area, inner plane, etc.; see JESD51 series) is somewhat analogous -- but also more flexible, and (nearly) free of cost in comparison, so is worth documenting.

Tim

[PCB.Wiz]

A problem with linear-anything, is power dissipation.

The new SMD packages are great for low loss switching and have very low on losses, but are not so great at heat removal.

I did see a new isolated power package from IXYS that could be used for more modern Linear MOSFET designs. It just needs the market demand. 
This isolated package is designed for heatsink clip/pressure mount, not soldering.

[Randy222]

Do you have a citation for this statement of fact?  Example, have you worked at one and have read process documents to that effect?  Or you've seen appnotes which document it?

I don't know of any manufacturer that places every new item into a lab test jig.
Packages, various silicon junctions, etc are well-known things these days. Most of them are already modeled. We know in good terms how heat will move through the packaging along with max junction temp of the design.
We don't really see an abundance of new packages being used, that would require new lab testing for each new item.

The post behind this one shows some "new" ceramic bottom to the package. A model already exists that accounts for that type of configuration, we know how heat moves through the ceramic.

When the heat density goes up (watts per area) it's harder to sink to keep junction temps stable. Why not make the rated 20A 100W IGBT (that comes in TO220) in a package that is 1/2 the surface area of the TO220? It can be made, but controlling the temp inside will be an issue. The makers would love to make things smaller, but heat density becomes an issue even when the heatsink is very large.

[T3sl4co1l]

Those are all very nice assumptions -- but I can extrapolate from physical principles as well; notice I asked for citations, so, I guess I should assume you don't have any?  (In other words: "a 'no' would suffice".)

I'd love to have such references myself; they're dreadfully sparse on the internet as it is.  Oh well, there's always hope.

Why not make the rated 20A 100W IGBT (that comes in TO220) in a package that is 1/2 the surface area of the TO220? It can be made, but controlling the temp inside will be an issue.

Well, packing the die into such a small part is the first challenge.

It's not impossible; maybe not very practical outside of pulsed application, but a few products exist.  DPAK is much smaller than TO-220/D2PAK:
https://www.st.com/content/ccc/resource/technical/document/datasheet/bb/1b/ca/f3/e3/b6/4e/02/CD00058414.pdf/files/CD00058414.pdf/jcr:content/translations/en.CD00058414.pdf

The other part of that is, the heat density doesn't at all go up; it's limited by die thickness and attachment.  DPAKs can be rated fairly generously, in the most generous cases at least, but at 62W maximum, this certainly isn't one of them.  Doesn't help that it's a co-pack, so (a bit less than) half the bond area is committed to a diode.

Tim

[DavidAlfa]

But - Do we agree that continuous power dissipation is just that, no matter what?
Yes, I perfectly understand surge/peak power, meant to be a small, low duty occurrence, let's say 1000% of the of max continuous power, lasting 20ms every 1s.
But "100W continuous power", to me it means "This device can dissipate 100W 24/7 below a defined Tj limit".
No matter if it's a small array of IGBTs or whatever, continous power should account for hotspots and manufacturing tolerances, some parts migh be extremely well bonded and tolerate 200W, while the worst only 130W.
Apply a safe gap and certify for 100W.

[Randy222]

But - Do we agree that continuous power dissipation is just that, no matter what?

Dissipating joules per sec means just that, but it in no way describes the temps.

Mount a TO220 high power metal film resistor to a huge heatsink and dissipate 100w, mount the same TO220 to a much smaller heatsink and dissipate 100w. The smaller one will be much hotter, including the TO220 package.

Watts just gives you an energy rate, how you sink out that energy is a bigger discussion/problem. As long as the sink is big enough to keep junction temp below max, for any application environment, then the package should survive.
The package itself is a thermal resistor, so for any given heatsink there will come a watts number where package temp gets too high. As example, TO220 in open air vs same in open air with a fan vs same mounted to large hunk of aluminum that is water chilled.
If the heatsink is very good (big water cooled, big fan cooled, etc) you can likely have the package dissipate more than rated watts, as long as the package temp does not go high enough to damage it. In real world there are limitations when it comes to physical heatsinks.

[Kleinstein]

The rated power (P_tot) for power transistors is for the limit of a very good heat sink, that can hold the case interface to 25 C.
With a more realistic, no so perfect heat sink one would need to reduce the power from the P_tot parameter.
It would need cooling the case to less than 25 C to safely allow for more than the nominal power.

With plastic cases P_tot is usually calculated for 150° C junction temperature, while for a metal case (e.g. TO3) they usually allow 200° C. A higher temperature will cause a reduced lifetime. With a plastic case presumably from degrading plastics. It can still work for some time.

Especially modern MOSFETs in a TO220 case give power ratings that are not practical to be actually used.

[Phil1977]

That´s exactly the issue with the Tc=25°C values.

They sometimes are tested by connecting the package to an active cooler that pumps a cooling agent through the mount. Then a thermal sensor is attached to the package and the cooler is controlled by a PID-loop so that the package temperature stays exactly at 25°C, though the mount may need to be actively cooled to 10-20K below ambient.

In this system you can test how much wattage P_tot can be dissipated in the semiconductor so that it´s junction reaches e.g. Tj=150°C.

This value is valuable for calculating thermal resistances and setting up thermal simulation models. But it´s quite far from a safe dissipation value in real world usage. For this purpose it´s much better to have e.g. a Tc=75°C value that allows the real heat sink that is connected to the component to heat up during usage.

PS: You can anyhow get quite good estimations of the possible real world dissipation. In a first approximation you can just interpolate between Tc=25°C and Tj=150°C - that means if the component is specified for P_tot=250W at Tc=25°C then it probably survives a dissipation of half of P_tot = 125W for a case temperature half between 25°C and 150°C - which is 87,5°C.