What limits the number of MOSFETs/IGBTs in series or in parallel?

Question

I was looking at HVDC systems out of curiosity and I have a question that I have not exactly found an answer for.

In HVDC converter, because of the high voltage (tens/hundreds of kV) they have to use so-called IGBT valves, which are multiple devices in series in order to increase the voltage blocking capability.

Why is there a limit to how many devices can be used in series? Is it because the conduction losses are too great? Is this also why they can't use let's say 10 MOSFETs in series instead of a big IGBT?

I would also like to know how the parallelism is affected in this case. I know that for MOSFETs this isn't a big problem, as they conduct less at a higher temperature, thus no positive feedback leading to the destruction of the device. I'm not necessarily looking for a straight answer, I am happy to research this myself if I am pointed in the right direction.

Answer 1

HVDC design engineer by trade here. You don't want to use MOSFETs for the simple reason of cost (chip area) per A of current and since MLC concept was published, everyone is switching at low speed. Paralleling both IGBTs and MOSFETs becomes a challange at kA levels due to parasitics and fault cases. Series connecting is a major challange. Don't do it unless you absolutley have to and be prepared for several issues with high losses in snubbers, avalanche and so on.

My question is why is there a limit to how many devices can be used in series?

Losses in snubbers, timing requirement on your gate drive, avalanche capability or ns response time in local gate drives with collector voltage slope regulation.

I would also like to know how the parallelism is affected in this case.

Your simple small MOSFET SMPS on your bench will usually work fine with parallelled MOSFETs as long as you have individual gate resistors. Moving up into kV and kA territory, parasitics will have far more to say and all work against you. Matched chips from the same wafer is something you won't be able to do at home and the kV of withstand will make sure you have a long way to go from your gate driver to each chip too, so you are in for a challange.

Bottom line: Don't series nor parallel any transistors unless you absolutley have to!

Answer 2

My question is why is there a limit to how many devices can be used in series? Is it because the conduction losses are too great?

My (limited) knowledge is in the use of thyristors for use in giga watt DC links as supplied by GE, Siemens, ABB etc. I'm not expert but I do have some knowledge. So my answer is about using thyristors but, the same would apply with MOSFETs and IGBTs (except those devices need series diodes when handling the incoming AC).

Let's take an example of the Infineon T2871N80TOH. It is rated at maximum voltages of 8 kV and max current of about 5000 amps. When conducting it has a forward volt drop of about 3 volts at 5000 amps. This means a power dissipation of 4.5 kW and, if (say) 100 are used in series to achieve a peak handling voltage of 800 kV then the power lost is about 0.5 MW.

Compare that with the throughput power of such a converter. Let's say it controls an RMS voltage of 400,000 volts and average current (say) 2000 amps RMS. That is a load power of 800 MW.

So does wasting 0.5 MW look such a bad deal when controlling a load of 800 MW? Does it look like there is some problem with having more series thyristors if the required HVDC link voltage needed to be (say) doubled? It doesn't to me.

Answer 3

if you ever have to parallel transistors of any kind, you need to make sure they share the current equaly - because the point of paralleling is that one transistor wouldn't have managed the whole current alone.

on the other end, if you ever serialy connect transistors you need to make sure they share the voltage equaly - because te whole point of serial connection is that one transistor wouldn't have managed the whole blocking voltage alone.

now, sharing current and sharing voltage is a different story - especially during transients. it's switch mode so transients are a direct result of the concept : turn ON and turn OFF

Balance has to be mantained even during transistor turn ON and turn OFF transients, but the margin for current is vastly higher than for voltage. Most IGBT can accept transient currents that are well over 50x their DC max rating - and this is specified in their datasheet. MOSFET do around 2x. Voltage wise if you exceed the DC max rating you're most of the time rigth away outside the spec (GaN is an exception), and on top of that at +20% it usualy cause destruction of the device.

to makes things worse the non linear capacitance of the channel (high during conduction, low at full blocking voltage) will work against serial connected transistors: if eg. during turn OFF a transistor happens to have higher voltage than it's peers then it will also experience higher dv/dt due to the capacitance dropping with voltage. it will in turns increase its voltage further, decreasing capacitance even more - a positive feedback.

it makes serial connection of any type of transistors very challenging because they all need to be tigthly balanced even during transient; with a mechanisum that amplifies any missmatch of balancing. You end up having to add a lot of margin to your design, building a chain of transistors with a total blocking voltage that is more than twice the DC blocking voltage. To take on Andy Aka example, a 100x8kV IGBT stack is only able to block DC voltage of 400kV - far from the 800kV sum one would have expected. Some active circuit can be used to dynamically control dv/dt during transients and ease the issue, but it makes driving very complex because it's all about matching many floating voltages during a transient. "controling" dv/dt is a way to say reducing dv/dt to the lowest of all transistors - which increase switching losses. In my opinion that's one of the reason why engineers embraced the high complexity of Modular Multi Level Converter (MMLC) - more failure points, but individualy the cells transistors are easier to drive and protect.

in short:

• serial connection of transistors (JFET; MOSFET, IGBT, BJT, MESFET, ...) is a risky business rigth from the start. Transistors have low over voltage tolerance (even for transients) and their output characteristics worsen balancing.
• parrallel connection of transistors is relatively easy as long as they have negative temperature coeficients (MOSFET, JFET, MESFET) but very difficult with positive temperature coeficient (BJT, IGBT, ...). Recent work on IGBT has led to negative coeficient IGBTs, easing the issue.

In all case it consider that parasitics has been keep to a minimum or/and matched - which is also a good amount of work.