WARNING graphic nerd porn! Si vs SiC MOSFET losses compared

[zombiess]

This is a simulated comparison I performed on Si vs SiC MOSFETs for a buck inverter I am designing. I thought that others might enjoy some of the info I got from performing these. The original goal was to simulate the junction temperature of the MOSFET after calculating it's losses. This is a little trickier than it sounds as it has to be through an iterative process. Once a Tj is found, the simulation must run again to calculate the losses at the higher junction temp, which then needs to be run again. I found through trial and error that 10 iterations was a good amount to achieve a stable Tj.

Many of you have heard me say that there is more to losses than the conduction loss. Conduction loss is determined by the RDSon of the device.

Choosing a device for your inverter depends a lot on what ranges it will operate in. Bus voltage, switching freq and how dynamic the load can alter your choice in device to meet a goal.

All of the heat sinks were setup at the same and manufacturers spec sheet values were used. Many of the losses in the device vary with current and junction temp. To account for this, 2nd and 3rd order polynomials were used instead of linear fitting to maintain accuracy. The formulas were calculated by performing a curve fit from the datasheets values. Not all the manufacturers datasheets have the same parameters, so this comparison was customized to these devices to make the outcome as accurate as possible while trying to avoid assumptions.

Only the main losses were simulated, switching energy loss, conduction loss, diode forward drop and diode reverse recovery.

Feel free to ask questions and I'll do my best to try and answer them. If you would like to see a comparison of these two devices performed, post up what settings you would like and why. I'll try to get the results posted so we can discuss them further. Unfortunately I do not have much time to compare different devices than what I have here due to the time to enter the data into the simulation and the datasheet research to get it as accurate as possible.

I'm not sure what the accuracy is of this to real life. I would hope it is withing 10%.

And now for the graphic nerd porn I promised :lol:

REMEMBER when reading these graphs that the losses were calculated at EACH PWM PULSE on a 1Hz sine wave. That is why there is a sinusoidal shape to the losses. To get the total, you need to take the integrate the area under each curve, that is what was done in the text showing the total of each curve.

First up SiC at 1A and 10A @ 170 VDC Bus



View attachment 23
View attachment 22




View attachment 18

View attachment 16

View attachment 14

View attachment 12


Si MOSFET 1A and 10A @ 170 VDC Bus





View attachment 5

 

[liveforphysics]

Nice work!

 

[toolbag]

This is really great! I'm amazed at how well the SiC part holds up at high frequencies.

 

[bigmoose]

Great work Zomb! Multivariate modeling always leads to better optimization. From your studies, what conclusions have you come to? Digesting the graphs cold, its a little hard to get your arms around it all.

 

[Arlo1]

The graphs are hard for me to read. I'm color blind.... I need colors that stand out better.

 

[HighHopes]

i've done exactly what you did but with comparinson of Si IGBT/Diode and Si IGBT + SiC Diode. i also compared my losses to bench test to verify the math i was using. what i found was the math method using datasheet curve plotting like you did was more accurate than manufacturer's own published online tools to determine temp rise. this was because the online tools linearlize the datasheet curve where you are not. that's important distinction when your output is AC.

i think you should take your sineusoidal cycle and make an average. temperature rise is an averaging anyway.. when you point a temp laser gun you don't see the temp rise and fall in a sinusoidal wave.. you see ONE number, the average. so i would remove your plots and just post bar graphs of the average.

also, what is interesting conclusion.. at what switching frequency does the average loss become the same as Silicon? i can tell you that for IGBT with SiC diode, it was 40kHz (this is for a 50kW motor drive, 540Vdc bus).

 

[Arlo1]

i can tell you that for IGBT with SiC diode, it was 40kHz (this is for a 50kW motor drive, 540Vdc bus).

What do you mean by this?

 

[HighHopes]

i mean i had an IGBT phase leg module, 3 of them, rated for 1200V, 300A. inside this module was IGBT with anti-parallel diode both were silicon based (similar to your leaf inverter module). the manufacturer also made for me the SAME module with the same IGBT but replaced the anti-parallel diode with silicon carbide diode.

then i ran both modules on the same platform with 540Vdc input, same environment, same cooling. went to 50kW continuous. measured the temperature. this was with 20kHz switching on both module tests. the all silicon module had temperature rise of some value and of course the silicon carbide had lower temperature rise. so i increased the switching frequency of the siliconcarbide module until the temperature was the same as the all silicon module (at the lower temperature). the switching frequency where temp was the same was 40khz. or ... you could run the siliconcarbide module at 20kHz (same as the all silicon one) and be happy with 30% less losses. remember this was just the diode that was siliconcarbide. what zombiess is talking about is a mosfet with the entire thing siliconcarbide so i would expect better than 30% dissipation improvement in his test.

 

[zombiess]

What I found with this simulation which is based off HighHopes previous work, is that the SiC device in this inverter design become advantageous for inverters operating > ~25kHz.

20kHz
Si = 65C @ 20W of losses
SiC = Tj 78C @ 22W of losses

As you can see from the losses, the Si device is a better choice for this inverter if the switching freq is < 20kHz. The Si device even runs cooler making thermal management easier.

Once the switching freq goes > 25kHz, SiC becomes a clear winner.

40kHz
Si = 77C @ 32W of losses
SiC = 85C @ 25W of losses

Some other observations
SiC runs hotter than Si... depending on application.

For a high current application such as a motor drive Si might be a better choice. There are so many variables at play, a study needs to be performed.

SiC Diode Vsd losses are high compared to Si. The SiC diodes have a forward voltage drop of > 3V and some as high as 4V. Thankfully the Qrr is very low on SiC which once again adds to their strength for high switching frequencies.

I learned just how quickly switching losses add up. I've been saying on here for a while that there is a lot more to a MOSFET than the RDSon value that most people focus on. There are many decisions which need to be made before selecting a power device. It's not always as clear cut as it seems on the surface. This seems to follow the trend of MOSFETs in general. Datasheets have valid test info, but you must know how to interpret it. Specsmanship seems to be the name of the game until you get into the higher end manufacturers. It's annoying how few manufacturers publish important specs such as switch energy. Devices can be difficult to compare 1:1 without doing some math that isn't easy to locate and is often just a best guess.