Why simulation is better for estimating Si MOSFET switching losses

[@levnemets2148]

Hello Prof. I totally agree with your opinion regarding simulation. The only problem is how accurate MOSFET's simulation models are. I have tested some models, and I can say that not all models can be trusted. For example, INFINEON models correspond very well to the datasheet, but Vishay do not at all, that is very annoying. In your video you created thermal network for MOSFET, but INFINEON has the models with internally included thermal network, which allows automatically adjust MOSFET parameters depending of Tj

[@sambenyaakov]

Good points. Indeed, not all the models are good. Hopefully, they will improve with time. AS for the advanced infineon model, we found that it suffers from convergence problems. WE used the older model and corrected for dependence of Rds(on) on Tj.

[@levnemets2148]

@@sambenyaakovIs it possible to dynamically pass back to MOSFET model information of Tj obtained from external thermal network during simulation?

[@sambenyaakov]

@@levnemets2148 Yo can put a series resistor which is a function of Tj to make Rds(on) Tj dependent.

[@alexpaciniat]

@@sambenyaakov Be aware that most Infineon models are validated on Simetrix, and they might have issues with other simulators (eg. LTSpice, QSpice, ...). [All statements and opinions are my own and unrelated to my current employer.]

[@sambenyaakov]

@@alexpaciniat Indeed, the advanced Infineo mode are almost useless on LTspice.

[@avin6602]

Thank you! if considering the parasitic indictnace of the MOSFET it turn out the Vds of the HS and LS starting to change imidiatly with the changing of currnt (not only when reachnig the main inductor currnt {load currnt}) and this is also can change the swithching losses quite a bit. Same for the Vds parasitic inductanceof the PCB.

[@sambenyaakov]

Indeed, see kzbin.info/www/bejne/fWGQg3aIq8lsprs. This agin shows why simulation is better.

[@nurahmedomar]

@7:12, should P=Id*Vds(t) also includes conduction loss in your simulation? The load current is SINE, when M1 is fully conducting, Vds is not totally zero, and P=Id*Vds includes both switching loss and conduction loss for the whole period. Maybe I'm wrong, can you elaborate on this? Thanks.

[@sambenyaakov]

Yes, P=Id(t)*Vds(t) includes conduction losses to get the total losses for estimating the junction temperature. It is easy to get the switching losses by themselves.

[@2meters2]

I have a question related to switch-on power losses as mentioned in data sheets. In many data sheets, there is a mention of "Turn-On Switching Energy (SiC Diode FWD)" and "Turn-On Switching Energy (Body Diode FWD)". For example the C3M0040120D. These numbers (especially for the "Body Diode FWD" case) are very high, in the order of mJ, which would render them useless even for PWM frequencies of 40-60 kHz. Could you comment on what the value is of these "switch on" energy loss numbers for estimating the losses if we are just building a half-bridge, buck power converter ? Thank you !

[@sambenyaakov]

Good question. Indeed the switching energy is high but for the Sic you mention this is for a 800V voltage and high currents. Comparing to IGBT the loss energy is aout 3-4 time lower. You can't compare it to GaN since they are limited to 650V. Unfortunately, this is the situation today.

[@alexpaciniat]

​@@sambenyaakov GaN vs SiC, on a similar voltage class, I have seen to have around 3 to 4 times lower switching losses in hard switched applications also due to the zero Qrr, which is also strongly temperature dependent. [All statements and opinions are my own and unrelated to my current employer.]

[@sambenyaakov]

@@alexpaciniat GaN has definitely lower losses (and eventually might be more economical than SiC). But all my attempts (and attempts by others) to apply GaNs in high power applications in which SiCs are now used (e.g. EVs) failed. Have you any idea why?

[@alexpaciniat]

​@@sambenyaakov I am actually more interested in hearing your opinions on that :) There is probably more than a reason, depending on the application. To be fair, in my previous role I was involved in the definition of the next generation of GaN devices (in Infineon), so I cannot comment too much. But in general, a properly designed (pareto-optimal) GaN based converter often dominates a pareto optimal SiC equivalent in terms of performance and power density. Then [system] cost (obviously), reliability, ease of use, peak current, Rdson vs temperature, available voltage classes, etc..., are all factors to be weighted in in a final system.

[@sambenyaakov]

@@alexpaciniat ​My wild guess is that large GaN dies are prone to breakdowns, perhaps due to avalanches as a result of uneven spread of current. But this is just a conjecture.

[@eduardinification]

Hi Prof. There is simulation SW that rely on turn-on and turn-off energy related data (among other parameters) to calculate power losses of switching devices. The data is temperature parametriced and some manufacturers provide the model for such calculations (a part of the SPICE model), and it seems that the power loss calculation is accurate even though is using a ideal switch for simulation. Do you think that is better to use this kind of SW instead of SPICE model based?

[@sambenyaakov]

Hi, 1.We add temp effect 2. Newer models of MOSFETS already have a built in temperture dependene of Rds(on) as well as therma netwirk foe Tj. 3. I have ne experince with loss calculation SW. I would assume that simulation is more acurate especially if you use the newer MOSFET models.

[@mhanssler]

Hi professor, would this contribute to premature failures? I.e. case temperature looks ok but the losses are causing an unseen issue?

[@sambenyaakov]

tanks for commet. The (t) belongs to all. P=Id(t)*Vds(t). Yes, it does include conduction losses. The purpose is to get the total loss for estimating the junction temperature. Easy to separate between conduction and switching lossed.

[@omridavidi8885]

תודה רבה מעניין מאוד

[@sambenyaakov]

👍תודה

[@tamaseduard5145]

👍🙏❤

[@sambenyaakov]

👍😊🙏