IGBT in Linear Mode

Since an IGBT is a 4-layer device one would think that "linear mode" might be like tip-toeing to the edge of a cliff without a safety rope ;-) ...Jim Thompson

--
| James E.Thompson, CTO                            |    mens     |
| Analog Innovations, Inc.                         |     et      |
| Analog/Mixed-Signal ASIC's and Discrete Systems  |    manus    |
| Phoenix, Arizona  85048    Skype: Contacts Only  |             |
| Voice:(480)460-2350  Fax: Available upon request |  Brass Rat  |
| E-mail Icon at http://www.analog-innovations.com |    1962     |
             
I love to cook with wine.     Sometimes I even put it in the food.

Section 6:

"1. Operation in short circuit. The current in the IGBT is limited by its gate voltage and transconductance and can reach values well in excess of 10 times its continuous rating. The level of hole current that flows underneath the N+ source contact can cause a drop across r=92b, large enough to turn on the NPN parasitic bipolar with possible latching. This is normally prevented by a reduction in r=92b, as mentioned in the previous section or by a reduction of the total device transconductance. Since this second technique increases conduction losses and reduces switching speed, two families of IGBTs have been made available by IR, one optimized for low conduction losses, the other for short circuit operation, as indicated in Section

2. "2. Inductive turn-off, sometimes referred to as "clamped IL." In an inductive turn-off the voltage swings from a few volts to the supply voltage with constant current and with no channel current. These conditions are different from those described in the previous section in so far as the load current is totally made up of holes flowing through r=92b. For this reason some manufacturers suggest the use of gate drive resistors to slow down the turn-off dv/dt and maintain some level of electron current, thereby avoiding a potential "dynamic latching" condition. IGBTs from International Rectifier can be operated at their maximum switching speed without any problem. Reasons to limit the switching speed should be external to the device (e.g., overshoots due to stray inductance), rather than internal.

"3. Operation as a linear amplifier. Linear operation exercises the SOA of the IGBT in a combination of the two modes described above. No detailed characterization of IGBTs as linear amplifiers has been carried out by IR, given the limited use of IGBTs in this type of application."

IOW, they didn't think it'd work so they didn't try it.

OTOH:

Section 3.2, "Reducing dV/dt at turn-off" et seq has some tricks to give at least some of that cliff slightly less than infinite slope.

Hey Tim, what's the required BW and signal fidelity? Why fight IGBTs' inability to do "real" linear operation; could PPM with averaging in the load work in your app?

Mark L. Fergerson

Don't know about IGBTs, but lots of high-power-rated mosfets will blow up at relatively low power dissipations in linear mode. We had to test a lot of them.

ftp://66.117.156.8/ExFets.jpg

John

I've just begun the dance with the customer, so I'm not sure of the required bandwidths, etc, although 10kHz is probably sufficient. PPM wouldn't work in this case (alas) -- it really is something where the pass device 'wants' to be working in linear mode.

I'm not sure why they're interested in using IGBT -- I'll have to ask when the occasion comes up.

--

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Do you need to implement control loops in software?
"Applied Control Theory for Embedded Systems" was written for you.
See details at http://www.wescottdesign.com/actfes/actfes.html

The risk is undoubtedly there for IGBTs, as well. They're designed for high voltage * low current, or high current * low voltage, not middle^2.

Were they blowing up at the expected power dissipation, or do they tend to blow up earlier than their maximum junction temperature & thermal conductivity would indicate?

--

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Do you need to implement control loops in software?
"Applied Control Theory for Embedded Systems" was written for you.
See details at http://www.wescottdesign.com/actfes/actfes.html

We tested a bunch of mosfets rated for 300 watts continuous dissipation. They were bolted to a copper block, and pulsed at various power dissipations for 0.1 seconds at very low duty cycle. The application is small-bore MRI gradient drivers. I think we tested them at 200 volts D-S, can't recall exactly.

Most of the ones we tried exploded at well below 300 watts. Power mosfets are apparently optimized for switchmode use, and get something like a second breakdown effect if made to dissipate lots of power at high voltages.

Ixys makes some parts rated for high-power linear use.

John

Earlier, AIUI. A secondary breakdown/SOA (localized heating) sort of thing.

FETs should always be ok when not exceeding the SOA. Did you go outside of the SOA? 100msec is usually the lowest run or already considered DC.

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

Since they were rated for 300 watts continuous, we clearly weren't out of the SOA at less than 300 watts.

I have seen some mosfet SOA curves that included voltage effects. I recall seeing a fet whose SOA was reduced by about 4:1 at higher voltages.

For the fets we selected, we wound up estimating a thermal model. In the amplifier,

ftp://jjlarkin.lmi.net/Amp.jpg

we have a uP that digitizes rail voltages (up to +-200), amp output voltage, fet currents, and heatsink temperature. It runs a realtime simulation of junction temperatures and shuts things down at an estimated 140C Tj. That lets us push the fets pretty hard.

But seriously, switchmode fets get fragile at higher voltages. The SOAs probably assume they are running close to saturation. [1]

John

[1] I use "saturation" in the bipolar sense, low Vds, ohmic region. Some people use the word to mean operating fets in the higher voltage, constant-current region.

Well,since there is an abundant lack of that information, and you seem to have a have more than a passing interest, then: DIY.

Testing semiconductors to destruction is tedious and expensive. Especially so for big semiconductors.

I was doing that on some Claire mosfet SSRs and discovered that I could tease them to just-before-destruction, by looking at accelerating drain current, probably the thermal precursor to some second breakdown sort of thing. Once I got that sort of calibrated, I could take a lot of data without incinerating a lot of chips.

John

Waaayyyy OT, but yes, I've done that, hung my feet over and climbed down the wall. 5,000 feet vertical.

-- Cheers, James Arthur

I had a good friend who would dance on the edge of cliffs. I think it was as much to hear the words "Peter, No!" as for the visceral thrill.

Somehow, I'm not inclined to do that with my customer's money unless there are clear advantages to success, and the customer's signed up to what I'm doing.

--

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Do you need to implement control loops in software?
"Applied Control Theory for Embedded Systems" was written for you.
See details at http://www.wescottdesign.com/actfes/actfes.html

The specified dissipation figure usually means that you are keeping the die below its maximum temperature by keeping the case at 25 C.

When pulsing, then the transient thermal behaviour needs to be taken into account. There's usually a derating curve for that as well. Spice can model the die temperature for that too if the mfr. gives you the thermal circuit.

Small die die quicker.

boB

Yabbut, the transient cooling is _better_ than the steady state cooling, because the heat is flowing into surroundings that are cooler than they would be in steady-state.

Cheers

Phil Hobbs

--
Dr Philip C D Hobbs
Principal
ElectroOptical Innovations
55 Orchard Rd
Briarcliff Manor NY 10510
845-480-2058

email: hobbs (atsign) electrooptical (period) net
http://electrooptical.net

Sure. A copper block does that pretty well, for a tenth of a second.

But the point I was making is that most power mosfets are designed to work in switchmode, and can blow up at way under rated power dissipation when used in linear applications, namely dissipating power with higher Vds.

We were blowing up "300 watt" mosfets at 150 watts, in 50 milliseconds, bolted to a copper block.

John

At least when I worked with them, it was STRONGLY discouraged. The problem was that while FETs have negative thermal coefficient, so the current tends to share over the area of the transistor, the IGBT has a positive coefficient, so current tends to "hog" to the hottest area of the transistor, leading to rapid damage. The only way to fight this is to saturate the transistor quickly and hard.

Jon

300W is most likely a marketing spec. But no matter what, a power rating has nothing to do with SOA.

A FET can be perfectly ok at 3V/100A (300W) but explode with gusto at

50V/20A (100W). The SOA diagram will tell you. Like this one in the link Lasse brought:

They all drop off in the SOA at higher voltage.

Nice blue theme color there :-)

It's all spelled out in the SOA curves. Don't push you luck there or it'll go ... tssk ... *PHOOMP*

Thing is, if you are just a wee bit into forbidden SOA turf it may all work fine for many months until one of the FETs really gets sick of it.

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

I haven't needed to do it in quite a while, but you CAN hook a behavioral device into your simulation to automatically plot SOA curves. I've even been known to have it also plot the spec as well ;-) ...Jim Thompson

--
| James E.Thompson, CTO                            |    mens     |
| Analog Innovations, Inc.                         |     et      |
| Analog/Mixed-Signal ASIC's and Discrete Systems  |    manus    |
| Phoenix, Arizona  85048    Skype: Contacts Only  |             |
| Voice:(480)460-2350  Fax: Available upon request |  Brass Rat  |
| E-mail Icon at http://www.analog-innovations.com |    1962     |
             
I love to cook with wine.     Sometimes I even put it in the food.