DC Load MOSFET Question: PWM or Linear Control?

[timb]

Hey guys, I’m working on a high power Programmable DC Load. I’m trying to spec out the MOSFETs but have a few questions…

I know most MOSFETs aren’t designed to run in the Linear region due to uneven heating of the internal holes. From what I understand, at lower Id they have negative tempcos as well. I suppose most MOSFETs are used in switching applications, where these problems don’t show up.

After doing some comprehensive research, I’ve found there are several MOSFETs designed specifically for a SOA in the Linear Region. This would be ideal for my DC Load; one issue is that they tend to be rather expensive.

So I’ve been thinking, is there any particular reason you can’t just use a normal MOSFET with PWM control? I’ve been specifically looking at IRF’s DirectFET line, which basically puts the die directly under a metal shell, allowing pretty much the most direct two sided cooling you can get.

To summarize, is there any reason running MOSFETs with PWM instead of DC Servo control wouldn’t work for a DC Load?

[bktemp]

Think again: When the mosfet only acts as a switch, where does the power go? The only way this could work is building a stepup converter pumping the voltage into a fixed resistor, so simulating a variable load.

[timb]

When the switch is on, the power goes through the switch, where some of it is turned into heat due to the resistance, the rest goes to ground. The same thing that happens when you run the MOSFET in linear mode. The difference is that in linear mode you’re controlling the power flow by how much the switch is open. In switching mode you’d be opening the switch fully each time, but controlling how long it’s on.

If you switch at a suitably fast speed and use a few large capacitors to smooth the ripple out…

Or am I thinking about this from the wrong perspective?

[Richard Head]

I'm working on a similar project (1600W) and am using very expensive IXYS MOSFETS (SOT-227 package) that are intended for linear operation. In my circuit each MOSFET has it's own current sense resistor and feedback loop. This ensures equal current sharing between devices to within 1%. The linear MOSFETs ensure that they are reliable in linear operation.
Note that the SOA of switching MOSFETs indicates that they are suitable for linear operation and are dissipation limited only. This is incorrect and the semiconductor manufacturers have admitted this. They are not reliable when operated within the SOA at high Vds and moderate current hence the emergence of special linear MOSFETs.
Before I opted for the linear MOSFETs I looked into using switching MOSFETs and PWM to discharge a capacitor bank via a high power load resistor. It can be done using PWM and probably has some advantages over the linear option. I opted to go the linear route as there was no noise produced (serious considerstion for this application) and cost wasn't too critical.

Dick

[bktemp]

When the switch is on, the power goes through the switch, where some of it is turned into heat due to the resistance, the rest goes to ground. The same thing that happens when you run the MOSFET in linear mode. The difference is that in linear mode you’re controlling the power flow by how much the switch is open. In switching mode you’d be opening the switch fully each time, but controlling how long it’s on.

If you switch at a suitably fast speed and use a few large capacitors to smooth the ripple out…

Or am I thinking about this from the wrong perspective?

You mean using the Rdson as the load resistor?
The current flowing will be extremely high, also the ripple current. At high voltages the current will be so high that the mosfet die melts almost immediatly. Just select a mosfet, use your input voltage and calculate the peak power dissipation, it will be in the kW region.
For low voltages it could work, but there is no benefit over linear operation: Some mosfet datasheets show SOA curves for DC. At low voltages, you can use the full power dissipation capability.
At higher voltage, the SOA curve gets smaller and your method even increases the current. It will be counterproductive.

[timb]

I'm working on a similar project (1600W) and am using very expensive IXYS MOSFETS (SOT-227 package) that are intended for linear operation. In my circuit each MOSFET has it's own current sense resistor and feedback loop. This ensures equal current sharing between devices to within 1%. The linear MOSFETs ensure that they are reliable in linear operation.
Note that the SOA of switching MOSFETs indicates that they are suitable for linear operation and are dissipation limited only. This is incorrect and the semiconductor manufacturers have admitted this. They are not reliable when operated within the SOA at high Vds and moderate current hence the emergence of special linear MOSFETs.
Before I opted for the linear MOSFETs I looked into using switching MOSFETs and PWM to discharge a capacitor bank via a high power load resistor. It can be done using PWM and probably has some advantages over the linear option. I opted to go the linear route as there was no noise produced (serious considerstion for this application) and cost wasn't too critical.

Dick

Interesting idea on discharging the capacitor bank into a large load resistor. And yeah, the IXYS parts is what I was looking at as well. I’ve read some of their app notes, along with AN’s from MicroSemi.

My original plan was to monitor low value current shunts on each FET with INA213s. The idea was to use a TI C2000 processor to control the feedback to the controlling op-amps.

What have you found in terms of heatsinks for your project? I’m still looking, but Ohmite’s C40 Series looks nice. (You can double them up and attach a fan directly to it. Another option would be something like this from eBay. The idea being you could tap sides of the case and mount the FETs directly.

[dannyf]

is there any particular reason you can’t just use a normal MOSFET with PWM control?

No. There are gazillions of them doing precisely that.

There must be something unique about your application / understanding / terminology that leads you to that particular conclusion.

[DanielS]

Think again: When the mosfet only acts as a switch, where does the power go? The only way this could work is building a stepup converter pumping the voltage into a fixed resistor, so simulating a variable load.

Even better: step-up the voltage and feed it back to the source's input to reduce your power supply testing power bill so you only pay for conversion losses... you save 70-80% of your energy costs compared to dumping all that power into resistors!

[David Hess]

What kind of derating would be necessary on MOSFETs intended for switching applications to use them in linear mode with appropriate ballasting?

Do any of the manufactures provide this information?

[diyaudio]

Hey guys, I’m working on a high power Programmable DC Load. I’m trying to spec out the MOSFETs but have a few questions…

I know most MOSFETs aren’t designed to run in the Linear region due to uneven heating of the internal holes. From what I understand, at lower Id they have negative tempcos as well. I suppose most MOSFETs are used in switching applications, where these problems don’t show up.

After doing some comprehensive research, I’ve found there are several MOSFETs designed specifically for a SOA in the Linear Region. This would be ideal for my DC Load; one issue is that they tend to be rather expensive.

So I’ve been thinking, is there any particular reason you can’t just use a normal MOSFET with PWM control? I’ve been specifically looking at IRF’s DirectFET line, which basically puts the die directly under a metal shell, allowing pretty much the most direct two sided cooling you can get.

To summarize, is there any reason running MOSFETs with PWM instead of DC Servo control wouldn’t work for a DC Load?

IRF’s DirectFET is used specifically for high speed applications, used to minimise package parasitics what is so high speed about your DC load?

[tom66]

Even better: step-up the voltage and feed it back to the source's input to reduce your power supply testing power bill so you only pay for conversion losses... you save 70-80% of your energy costs compared to dumping all that power into resistors!

Even better, feed the output of your boost converter into the input so you can generate infinite power**!

[nctnico]

I'm also working on a dummy load and choose to use resistors and switch these into various combinations to create a wide range of resistances. The advantage is no control loop and resistors may run much hotter than transistors so you can get away with a smaller heatsink. I've build a similar dummy load for a customer and it works pretty well.

If you really want to build a linear dummy load I'd always choose transistors. The collector of a transistor is a natural current source (or sink in this case) so you only need a slow control loop to take care of thermal runaway. In most cases you want to set a dummy load to a certain current.

[Pedram]

IRFPx50 (like IRFP150 ) works well on semi-high current loads... refer to http://www.eevblog.com/2012/05/23/eevblog-281-bk-precision-8500-electronic-load-teardown/

[dannyf]

What kind of derating would be necessary on MOSFETs intended for switching applications to use them in linear mode with appropriate ballasting?

Most SOA or temperature-based derating.

The role of thumb is that they tend to dissipate no more than 1/3 of what their package is rated for: 10w for to220 and 30w for to247/3p.

[diyaudio]

IRFPx50 (like IRFP150 ) works well on semi-high current loads... refer to http://www.eevblog.com/2012/05/23/eevblog-281-bk-precision-8500-electronic-load-teardown/

Spot on.

I've noticed they parallel them, I still dont get why those TL074 quad are arranged that way.

[Pedram]

IRFPx50 (like IRFP150 ) works well on semi-high current loads... refer to http://www.eevblog.com/2012/05/23/eevblog-281-bk-precision-8500-electronic-load-teardown/

Spot on.

I've noticed they parallel them, I still dont get why those TL074 quad are arranged that way.

i'm not sure about BK's one but this agilent one use separate resistor and 2 opamps for each mosfet
http://cp.literature.agilent.com/litweb/pdf/5951-2828.pdf ( page 90 )

[diyaudio]

IRFPx50 (like IRFP150 ) works well on semi-high current loads... refer to http://www.eevblog.com/2012/05/23/eevblog-281-bk-precision-8500-electronic-load-teardown/

Spot on.

I've noticed they parallel them, I still dont get why those TL074 quad are arranged that way.

i'm not sure about BK's one but this agilent one use separate resistor and 2 opamps for each mosfet
http://cp.literature.agilent.com/litweb/pdf/5951-2828.pdf ( page 90 )

I have this manual its a good reference document.

[timb]

So the BK Precision unit is using those resistors for balancing *and* current sense? Interesting, that's what I was suspecting after watching a tear down video of it. (Gerry's I think; Dave didn't go into much detail about them in his.)

Is there a minimum size that should be used for balancing? I'm planning on using a 0R002 resistor on the low side of each FET.


Sent from my iPhone

[tom66]

0.002 ohm is 18 times smaller than IRFP150N on resistance, how much difference will it make? The lead resistance is probably at least that much.

[diyaudio]

0.002 ohm is 18 times smaller than IRFP150N on resistance, how much difference will it make? The lead resistance is probably at least that much.

Note the RDS on is only relevant in switching applications. however taking current samples from shunts at that ohmic region is another story, current shunt theory should unveil the answer.   

[timb]

Sorry guys, I meant 0.020, not 0.002! Stupid phone. That said, I've gotten very accurate current readings with an INA215 and 0.005 shunt before, so 0.002 would be possible assuming the right layout.


Sent from my iPhone

[dannyf]

0.002 ohm is 18 times smaller than IRFP150N on resistance, how much difference will it make?

The size of the current sensing resistor has no relation to the Rds. As long as the current sense amplifier has sufficient gain, the current sensing resistor can be as low as you want.

[Mr Carlsons Lab]

0.002 ohm is 18 times smaller than IRFP150N on resistance, how much difference will it make?

The size of the current sensing resistor has no relation to the Rds. As long as the current sense amplifier has sufficient gain, the current sensing resistor can be as low as you want.

You may need this amp for that low resistance though

[timb]

Right, I understand that. I'm asking if I'll need additional resistance to balance the FETs. (I'm talking about running in Linear mode here, not PWM.)


Sent from my iPhone

[DanielS]

Right, I understand that. I'm asking if I'll need additional resistance to balance the FETs. (I'm talking about running in Linear mode here, not PWM.)

If you want to parallel multiple MOSFETs, a source resistance between the source and sense resistor would make the circuit less sensitive to slightly mismatched devices.

Or you could give each MOSFET its own error amplifier/integrator and make each channel track 1/Nth of the desired current.