What the record says a brick will take

Thirty-five published machines say what their switches actually ran at. Thirty two of them held and three did not, and they are tabulated in one place by one person who collected them: Mads Barnkob's IGBT chapter of the DRSSTC design guide, where they sit as a table of proven speeds and a shorter table of failures.

That pair of tables is worth more than either alone, and it says something its own author does not spell out: run the two lists against each other and the idea that a brick has a frequency and a current it can take does not survive contact with them.

What it is and why

Every datasheet figure is a hard switching figure. A DRSSTC does not hard switch, so the numbers a builder needs are not in the sheet, and the way the hobby has filled that gap is by writing down what machines did. The result is not a specification. It is a record: a frequency, a peak current, a part number and a name.

What the record can do is bound the question. What it cannot do is answer it for your machine, and the reason is in the tables themselves.

What you decide

Whether the record has a boundary in it

Take the obvious test first. If a switch fails because it is asked to turn off too much current too fast, then the rate at turn-off should separate the two tables. That rate is the slope of the resonant current as it crosses zero:

dI/dt = I_pk * 2*pi*f

Run it across all thirty five rows [derived, on his frequencies and currents]:

                      dI/dt at turn-off, A/us
proven, n = 32        min 220   median 377   max 1319
failed, n = 3         1232      528      220

The failures land inside the proven range rather than above it. The worst of them, a CM600HA-24H at 56 kHz and 3500 A, sits below the highest surviving row, an FZ1200R12KL4C at 35 kHz and 6000 A. The mildest failure is at exactly the lowest rate anybody proved.

So the rate is not the boundary. Something else decides, and the table has three cases that isolate it, because three part numbers appear in both lists.

The three parts that are on both lists

CM600HA-24H. Proven at 34 kHz and 2500 A by fh89 on the forum, failed at 56 kHz and 3500 A by Kizmo. Both terms were raised at once, so this pair says only that there is a ceiling somewhere between them. It is the least informative of the three and the easiest to over-read.

SKM400GB124. Proven at 42 kHz and 1300 A by Finn Hammer, failed at 42 kHz and 2000 A by Finn Hammer. Same builder, same machine class, same frequency, and the current is the only term that moved, by a factor of 1.54. That looks like a current ceiling until you read the failure row's own wording, which is the only row in either table to carry a repetition rate: it reads "2000A / 500 BPS". Five hundred bangs a second is a duty figure, not a peak figure, and heat that does not leave between bangs is a different failure from a switch that cannot turn off. The row does not say which of the two it was, and neither does this page.

IXGN60N60C2D1. This is the useful one, and it takes four rows to read.

proven    70 kHz    500 A   IXYS IXGN60N60C2D1 and SKM200GB123   Mads Barnkob
proven   150 kHz    500 A   IXGN60N60C2D1                        Alex Yuan
proven   327 kHz    300 A   IXYS IXGN60N60C2D1                   Mads Barnkob
FAILED    70 kHz    500 A   IXYS IXGN60N60C2D1                   Mads Barnkob

The first and the last are the same builder, the same part, the same frequency and the same current, one in each table. The proven row names two parts on one line and the failure row names one of them, which is the reading that makes both true: at that operating point he ran an IXYS and a Semikron, and the Semikron held.

Now put Alex Yuan's row beside it. The same IXYS part, the same 500 A, at more than twice the frequency, and it held. That is 471 A/us against the 220 A/us that failed [derived, both rows]. And the third row is the same builder again, running that part at 327 kHz, which is 616 A/us at a lower current.

Three rows for one part, and no ordering of frequency or rate puts the failure where it belongs. What is left is everything the table does not record: the layout, the gate drive, the cooling, and how many hours it ran before it stopped being in the proven list.

Which way the record slopes

The thirty two surviving rows do have a shape, and it is the one the physics predicts even though the failures do not sort by it. Low frequency comes with high current and high frequency with low: 6000 A appears at 35 kHz, 300 A at 327 kHz. That is the same trade this site works from the other end on the bridge page, and the record agrees with it as a tendency while refusing to give it a number.

Read the slope, then, and not the individual row. A part that appears at 300 A and 270 kHz is telling you about a class of machine, not offering you a ceiling to design against.

What will get you

Reading a row as a rating. Everything above. The word in the table is proven, and what it proves is that a machine existed, once, in somebody's room.

Taking the mildest failure as the safe limit. The lowest failing rate on the page is also the lowest proving rate on the page, and both belong to the same builder. A number that appears on both sides of a boundary is not a boundary.

Assuming a brick that survives a higher frequency will survive a lower one at more current. The IXYS rows are the counterexample: the same part is on the record at 327 kHz and at 150 kHz, and the failure is at 70.

Missing that one failure row is a duty figure. SKM400GB124 at 500 BPS is the only entry in either table with a repetition rate attached, and repetition rate is what decides whether heat leaves between bangs. Compare it against the proven row from the same builder as a duty question, not a peak question, and a pulse rating is a temperature is where that half is worked.

The numbers

rows, proven          32       rows, failed             3
parts on both lists    3       builders on both lists   2
dI/dt proven          220 to 1319 A/us      [derived]
dI/dt failed          220, 528, 1232 A/us   [derived]
highest current       6000 A at 35 kHz, FZ1200R12KL4C, zilipoper
highest frequency     327 kHz at 300 A, IXGN60N60C2D1, Mads Barnkob

Every frequency and current above is his; every rate is I·2πf on his figures and is marked as ours. The table is a self-report with no method attached: no row states how long the machine ran, at what bus voltage, or with what gate drive, and two rows name a part without naming the manufacturer.

What goes wrong

A part on the proven list dies on your bench. Nothing in the record was promised to you. The row you copied held for one layout, one gate drive and one cooling arrangement, and the IXYS case above shows two builders getting opposite results from the same part at the same current.

The arithmetic says one frequency and the record says another. Both can be right. The calculated ceiling is a thermal statement about a junction, and the record is a statement about machines that existed. When they disagree, the question is which limit you are actually near, and the answer is usually neither the one you computed nor the one you copied.

Where next

IGBT or MOSFET is where the device gets chosen rather than looked up, including what a TO-247 dependably takes and what paralleling actually buys.

A pulse rating is a temperature works the duty half, which is the term the one annotated failure row turns on.

The bridge and what kills it holds the voltage side of the same question, where the record is thinner and the arithmetic is firmer.