How to understand switching frequency

How to understand switching frequency in IGBT spec

So, I am looking at this datasheet: **broken link removed** , and I want to answer a simple question: what frequency can I switch this at? My problem is that the datasheet doesn't tell me this directly. I know that it depends on a lot of factors, like Vce, Ic, etc. Am I supposed to use the parasitic capacitances, charges, and rise/fall times to predict the operating frequency range?

I see some specs related to f = 1MHz, but I know that is too high for most IGBTs to operate at, so why do they use only 1MHz?

They use 1MHz to determine the internal capacitances (that is when the device is in the off-state and remains in off-state during measurements). Many semiconductor manufacturers specify capacitances at 1 MHz.

When you look to the rise time graph (figure 14), using this device at 1 MHz (T=1000ns) as a switch will not be efficient. Rise time is in the 200 ns range. Also the reverse recovery time for the parallel diode is almost 1000 ns.

When using about kHz in combination with a (quasi) resonant topology, you can have switching loss in same range as conduction loss. In case of a hard-switching topology (buck, flyback, etc), maximum usefull switching frequency will be much lower to have reasonable efficiency.

Thank you for the information. I do plan to use it in a ZVS QRC. I still do not know a direct way of determining what switching frequency I can use it at. I understand your reasoning, that the rise/fall times must be much lower than the switching period, but this does not give me an explicit switching frequency to go by, but rather an estimated/educated guess. Is it as simple as the rise/fall time specs that I should go by?

I don't think it is that simple. In my opinion, you should fully evaluate all losses and see whether switching loss is acceptable. When to high, you need to reduce switching frequency (larger magnetics?) or use faster components.

If you plan to simulate your circuit, you can put a current measuring probe in the simulation. With a pspice multiplier you can multiply Ic and Vce to see the instant power loss versus time. From this graph you can see where the losses are within one switching cycle. If you integrate the instant power loss versus time (you may use a low pass filter) you will get the average total power loss in the switch.

Besides the actual loss, there can be more factors. When using a bimosfet, one device could do your job saving you the problems with series circuits.

WimRFP said:

Besides the actual loss, there can be more factors. When using a bimosfet, one device could do your job saving you the problems with series circuits.

Click to expand...

This comment threw me on a long googling research session yesterday. I didn't exactly know what bimosfet was, but I figured it was just one of those technical designs that marketing likes to turn into a gimmick. When you say series circuits, are you talking about putting MOSFETs in series to share the hold off voltage? Is that why you say a bimosfet will save me the problems from putting MOSFETs in series, because its inherently higher hold off voltage?

AlienCircuits said:

This comment threw me on a long googling research session yesterday. I didn't exactly know what bimosfet was, but I figured it was just one of those technical designs that marketing likes to turn into a gimmick. When you say series circuits, are you talking about putting MOSFETs in series to share the hold off voltage? Is that why you say a bimosfet will save me the problems from putting MOSFETs in series, because its inherently higher hold off voltage?

Click to expand...

Yes, you are right, it may save you from putting mosfets in series to handle the working voltage.

As soon as SiC bipolar transistors (Vce> 2.5kV) are commercially available, I would love to do some experiments with them, as they have really low on-state voltage.

WimRFP said:

Yes, you are right, it may save you from putting mosfets in series to handle the working voltage.

As soon as SiC bipolar transistors (Vce> 2.5kV) are commercially available, I would love to do some experiments with them, as they have really low on-state voltage.

Click to expand...

Have you used series MOSFETs before? Is that how people switched high voltages with solid state switches before these high Vds/Vce rated switches were made? I only found a few IEEE articles about it where they used series resistors in parallel with each FET, and charge pumps on the FET gates.

I once did 2 mosfets in series with inductive gate drive. Series circuits of power switches is common practice in HVDC (High Voltage DC) distribution systems, but I don't have experience with such voltage and power levels.

WimRFP said:

I once did 2 mosfets in series with inductive gate drive. Series circuits of power switches is common practice in HVDC (High Voltage DC) distribution systems, but I don't have experience with such voltage and power levels.

Click to expand...

That is cool. I would like to try making a series HV mosfet switch just to have the experience just in case series switches ever becomes my only option in a project, but I don't have time. I am going with the bimosfet if I can get the stupid spice file to work in LTspice

I tried a SiC diode in spice because the reverse recovery time is supposedly 0, but I think it had really large junction capacitances because it was preventing my MOSFET from turning on correctly. I did not know that they are making SiC BJTs, that's pretty exciting

Resonant switching high voltage IGBT circuits are usually working below 10 kHz based on switching loss considerations, hard switching circuits in a low kHz range, just to shine a light on state-of-the-art. You didn't however tell about your switching frequency constraints, so it's hard to determine if the considerable effort for series connected low voltage devices pays somehow.

You are right, SIC diodes (schottky diodes in general) have larger capacitance then similar rated fast recovery diodes. Though the recovery proces in schottky diodes (SIC included) may be negligible, the larger junction capacitance results in a reverse charge, hence a reverse current spike.

In an application where I used fast IGBTs with SIC diodes (ZCS topology MF power source, fixed frequency) I had to redesign the snubber networks to reduce some switching transients. The overall recovery charge for the diodes was higher then expected.

If you can delay your project, maybe some other nice HV components will be on the market within some years.

FvM said:

Resonant switching high voltage IGBT circuits are usually working below 10 kHz based on switching loss considerations, hard switching circuits in a low kHz range, just to shine a light on state-of-the-art. You didn't however tell about your switching frequency constraints, so it's hard to determine if the considerable effort for series connected low voltage devices pays somehow.

Click to expand...

Thank you for shining light, that is definitely something that I am happy to get.

The design I have been working on is to step 1000V down to 800V, with only about 150-200uA load current. I have been using a MOSFET to switch at 20-40kHz, and I can use a .22H inductor and operate in discontinuous mode.

I will probably be lowering my switching frequency much lower, and then increasing my output capacitance to counteract the ripple if I use this bimosfet. My problem has been that the MOSFETs I have been using in simulations have really high parasitics since they are made for high power applications, rather than just high voltage. I could not make a driver circuit that fixed this issue. From that step, I began researching quasi resonant soft switching and am focusing on using a ZVS so that my switch can switch faster without those parasitics getting in the way so much. The problem with that is I found the peak voltage across my switch will be, at a minimum, twice the input voltage to maintain ZVS conditions, and that is why I began looking at this part since no MOSFETs can operate > 2000V Vds.

The 200 uA load current range changes a lot. Just as a quick guess, any available semiconductor device will have a much larger chip area than needed for the application, involve respective capacitances and a certain amount of non-revcoverable charge stored in it.

Small signal IGBT aren't available at all, so I would primarly think about small signal MOSFET, possible series connected. They are are made with up to at least 800 V Vds rating.

So perhaps use 2 small signal MOSFETs in series and completely avoid the parasitic issues which makes ZVS no longer a focus too. Is that what you're suggesting?

ZVS may be still reasonable to reduce losses, but not necessarily required.

200V across the component with 200uA is only 40mW. Why can't you use linear solution? BFC60 or other small signal HV mosfet?

Yes, I think that, because the current is so low, a resonant switch will not give me much to pay off the cost of needing twice as much Vds hold off voltage and additional parts if I used a resonant. I was mainly interested in the resonant switch to enable my huge parasitic (like you said, large chip area) MOSFET to actually switch. I will try to learn series MOSFETs now, and see where this gets me.

Am I correct that the basic idea with series MOSFETs is to place resistors across the Vds, and to charge their gates with a diode-capacitor pump? That is what I saw in some articles, but I was curious if that is just 1 way to do it, or the standard.

- - - Updated - - -

WimRFP said:

200V across the component with 200uA is only 40mW. Why can't you use linear solution? BFC60 or other small signal HV mosfet?

Click to expand...

Well, mostly because I am not very experienced with my understanding of FETs used as amplifers. I only knew enough to pass my class. Wouldn't I need to bias the FET to be in the saturation region with my unregulated 1000V input voltage? And also, I would need to keep it in that region for the whole swing of 200V, and hope that I can model it enough to make an accurate control system. I will look into this actually, as it seems quite practical.

Someone suggested that I use a series of 200V zeners that are biased by a FET at ground, but I was concerned about how good of a regulator that would be.

If there is some budget available, you may use Analog Devices ADUM isolated Mosfet drivers to drive "floating" mosfets.

Analog design isn't the easiest route (but very nice). It will give a clean output. If you don't want to design the interface, you may use an (expensive?) isolation amplifier.

Complete other route is to build two similar half bridge oscillators (that are in series with each other so you get good voltage sharing). The advantage of classical oscillators is there relative low EMI generation and energy in component's capacitances is partly recycled. Both oscillators may share same secondary coil with rectifier to generate the 800V. You have to find some means to regulate the output voltage.

Complete other solution is the drop the 1000V input and use a low votlage supply with converter generating the 800Vdc.

I think the most straight forward design will be to use a linear regulator now, considering your comment that it would only need to dissipate 40mW. Thank you guys for giving me a little dose of reality

Like I said, I have little experience using transistors in their amplifier region, and have only really needed to use them as switches before. I'm starting to brush up in that area and will be attempting to design a series pass linear regulator. I'm glad I learned a lot about bucks tho, cause I will probably need to do a SMPS one day.

I agree with WimRFP that a linear solution would be better. For such a small power level, there's really no sense at all in a switching solution. In fact you'll likely get much better efficiency with a linear regulator. And it should be much simpler. Even if you want to "practice" building a SMPS, this application sounds like a bad way to go about doing that.

For a linear regulator, you'll likely want a way to decrease the max voltage stress on the transistors (during start up the regulator will see 1000V from input to output, so you have to design for that). Putting multiple transistors in series can work, but then you need to make sure they share voltage somehow, which can be difficult. For a simpler solution, I would put a clamping zener in parallel with one transistor, and put a series resistor on the source. That way during startup, the voltage across the transistor will be clamped to something reasonable, and the initial voltage surge will be handled by the resistor.