The significance of switching losses on commuter Wh/km?

[bunya]

Hi,
I have done a little homework on the intricate nature of controller design but I still know very little... I would like to hear/read some experienced opinions about switching losses and how significant they are. The particular case study is looking at the new IRFP7718 and comparing it to the IRFP4368, both are 75V, 195A package limited and have similar Rdson of 1.8mOhms. The key difference is in the switching time, capacitance and gate charge of the FETS, I think this is mostly switching related, hence the new topic title

I would like to use this FET in an older style cyclone controller to replace the To-220 that currently reside there (TO-247 compatible). I believe it only switches the high side (upper) at 20kHz.

Ta,
Jake

 

[fellow]

If you want to upgrade TO-220 package, use the TI csd19506kcs (80V, 2.0mOhm, TO-220 package, 150A package limited). Other choices are Fairchild FDP027N08B (80V, 2.2mOhm, TO-220 package, 120A package limited) and TI csd19536kcs (100V, 2.3 mOhm, TO-220 package, 150A package limited). Those "package limited", Rds_on and V_ds figures are off course fantasy and measured at favorable temperatures and conditions regardless of what package/format. Keep a safe margin at all times. 100V 6-fet hack here, fets are TO-220 package:

The mosfet you are refering to is in bigger TO-247AC format. Not that it can't be done, but...

A pair of TO-247 to the left, and TO-220's on the right side. Note: Not the same scale as the first picture.

 

[bunya]

I should have specified more:
The cyclone controller has space/holes in the PCB for both TO-247 and TO-220 and I currently have unknown TO-220 FETs.
Losses from Controller switching are complicated, I have searched through the forum/web and found there are 3 main areas of loss: conduction (Rsdon), switching (Qg, rise and fall time) and diode (semiconductor junction forward voltage? :? )

I just want to know about the difference in switching losses from a FET of similar Rdson but a different Qg, as in the case of the IRFP4368 vs IRFP7718.

In essence, how significant are switching losses on a Wh/km basis (for example median draw of 500W), I know at full throttle they are very small, but what about partial throttle when they are switched at 20kHz.

If it insignificant ie <1%, I will go with the IRFP7718 as they are significantly cheaper.

Regards

 

[fellow]

If you have both holes, than it's much more simpler . You can take a look at the Qg part in the data sheet and compare. Less Qg usually means less switching losses and higher switching frequency. Ykick repoted big difference between 4110 and 3077 fets, so there is place for a gain there. Unfortunatly I don't have any measurements of Rds_on vs Qg part to back up.

I know that there is a big difference in between noname mosfets and 4110, as 4110 runs cooler (but not as cool as I expected). When used reasonably, there is a little difference. When you push their limits, difference is clear.

Here is Ykick quoted:

They (3077) don't turn as much energy into heat like the 4110 FETs do but you might not even be able to tell difference. I've been using them in the last 4-5 Lyen controllers and 3077 FETs appear to deliver 5-10% more range efficiency compared to identical controller and parameters using 4110.

Another problem is determining if the mosfets we are using / comparing to each other are genuine or not...

 

[bunya]

Thanks fellow, that is quite a significant difference, I have just looked at the 3077 and 4110 datasheets and the capacitance/qate charge is quite similar whereas the Rdson is about 30% higher for the 4110. I guess this would suggest that Rdson plays the larger role for Ebike applications, especially at higher power. Conversely it doesn't highlight the significance of gate charge and capacitance on efficiency.

 

[fellow]

Yes, that seem to be the conclusion as the 4110 has lower Qg (150nC) than 3077 (160nC), but 4110 has higher losses due to higher Rds_on. IRFP4368 and IRFP7718 have high Qg values, but lower Rds_on. Both TI mosfets seem to have both values at the lower side, but lower current limit than the To-247 mosfets. Most of the controller gurus here on ES seem to chase Rds_on value and prefer TO-247 pack (if possible), so i guess Qg is more important at extreme switching frequencies not used by our controllers. I'm sure they will chime in (Lebowski comes to mind)...

 

[Punx0r]

3077 is generally regarded as running cooler than 4110. That seems fairly typical of FETs with differing voltage ratings (higher rating, more losses for a given current).

IIRC a reasonable brushless controller should be high-90's % efficiency, so a modest gain in FET efficiency is unlikely to have a noticeable effect on your range/consumption. The main concern with FETs is not overall efficiency but preventing their own overheating/destruction, which considering their small size doesn't take much energy.

 

[Njay]

From a theoretical point of view, you can approximate switching losses of one FET in an inductive switching application, by excess, by (this approximation is actually based on the idealized switching diagram as in fellow's post)

[pre]Pd(sw) = 0.5 x Vds x Id x (tr + tf) x Fsw[/pre]
So, in a controller where you don't change the current limits, the power voltage nor the switching frequency (which is what happens when you only swap FETs, as far as I can see), switching losses are proportional only to switching time, tr and tf. tr and tf are the time it takes for the FET to switch from on to off and vice-versa, and that is controlled by the FET driver and the FET's required gate charge. Since, in the same controller, the FET driving capabilities don't change (as the driver isn't changed, when you just swap the FETs), it all depends on the FET's gate charge requirements.

When you swap an FET by another with lower gate charge requirements, the new FET will switch faster (reduction of tr and tf), thus having lower switching losses. However, by switching faster, currents also switch faster and that means higher voltage transients at the controller's rails and other places in the power stage. At some transient level point, your FET, or some other component in the power stage will die, either immediately or after some time.

 

[bunya]

Thanks, I can see that path you've taken to reduce it down to gate charge.

I think I over simplified tr and tf when comparing the datasheet between the irfp4368 and irfp7718 FET. Because although the IRFP7718 has a higher total gate charge of 552nC compared to the 380nC of the IRFP4368, the tr and tf values are 164ns and 160ns respectively for the 7718 where as the 4368 has tr and tf od 220 and 260 respectively. What I forgot to look at was the Vdd Id Rg which are different between the two data sheets.

Overall the difference between the switching losses compared as a percentage at 100% duty cycle.
IRFP4368 at 49V 195A as per datasheet.
P_loss_switch = 0.5 x 49 x 195 x (220+260)x10^-9 x 20000 = 45.9W
%loss_switching = 0.09%

IRFP7718 at 38V and 100A as per datasheet
P_loss_switch = 0.5 x 38 x 100 x (164+160)x10^-9 x 20000 = 12.3W
%loss_switching = 0.32%

Although it is difficult to compare between the two FETs here because they are operating at different voltage and current it can be seen that switching losses are relatively small for both FETs. I can say then that for practical purposes the difference here is relatively small although for the 7718 FET at a lower duty cycle of say 10% the losses would be approximately 3% or maybe higher (assumption below).

From a theory standpoint. I think the inductive coils will buffer the inrush of current at "turn on" and thus reduce switching losses associated with tr. This is speculation though. This inductance may well result in a voltage spike at turn off that negates the reduction at turn on (snubber caps).

In regards to my assumption that losses would be 3% at 10% duty cycle for the 7718 FET, I have a question about current through the stator coils at different PWM duty cycles:
At lower RPM Bemf voltage is low and a larger current can flow. Is this instantaneous current equal for different PWM duty cycles at the same low rpm, or does the inductance in the stator coils restrict peak instantaneous current at low duty cycles? ie At lower duty cycles does the this current never reach steady state condition before being switched off again? Perhaps I should just get a scope and have a play and work this out for myself (more practical learning, fascinating stuff! ).

If so this adds another dimension to the calculation of switching losses. I would be happy to extend this post if there is enough interest and try a small case study.

 

[Alan B]

The bottom line here is how much heat does the controller make? That's the only loss you can reduce. If your controller gets hot, there is something to improve. If not, then what can you save?

Even if the controller does get warm, the percentage of power delivered to power wasted is small, so the benefit is more to the temperature and life of the controller, not so much a system efficiency improvement (unless your controller is really hot).

 

[Njay]

I think Alan sums it up pretty nicely.

Anyways, your calculations are irrelevant. You would need to measure tr and tf in the controller. Those times from the datasheet are not using your controller's FET drivers, nor current limit, etc etc.

One way to greatly reduce dissipated power on your controller would be to stop using asynchronous rectification and start using synchronous. The difference in power dissipated can be very big, especially at higher currents, as the OFF cycle phase current going through a FET's body diode dissipates lots more than if the FET was turned on.
But doing it is an entirely different story, as it would require a change in the FETs control strategy -> at least the controller's firmware.

The controller limits the motor current, by terminating the PWM's ON cycle phase earlier (effectively reducing the duty cycle).
As far as I'm aware, the typical motors LR constant together with the typical switching (PWM) frequencies in controllers dictates that the current doesn't go down to zero on each PWM cycle, in fact, we don't want it to go, we want it to remain relatively stable (with a few percent ripple only) for reasons of motor torque and controller heat dissipation.

 

[zombiess]

The bottom line here is how much heat does the controller make? That's the only loss you can reduce. If your controller gets hot, there is something to improve. If not, then what can you save?

Even if the controller does get warm, the percentage of power delivered to power wasted is small, so the benefit is more to the temperature and life of the controller, not so much a system efficiency improvement (unless your controller is really hot).

This is great advice right here.

I don't believe you will have to worry a lot about switching losses, but in some cases they can become quite significant, but I doubt you would notice a difference.

Get the heat out the best you can. Since you are swapping out the MOSFETs, polish the mating surface the MOSFET will mount to and install good insulators. Many controllers come with really thick electrical insulators that don't conduct heat very well.

As for choosing a MOSFET, the candidates posted are all pretty close. Since this is relatively low power I don't think it will make much difference. If you ride full throttle a lot, then a MOSFET with a lower RDSon will run cooler and would be my first choice. I think most controllers run 10-16kHz, at least most of the ones I've seen are in this range.


Some quick guidelines about MOSFET losses.
Low duty cycle - dominant losses - diode losses (Vsd AND Qrr!), switching losses then conduction losses.

Mid duty cycle - dominant losses = depends on MOSFET, this is the region where you need to break out the math to figure out which is the dominant one.

High duty cycle - Dominant = Conduction losses, switching, diode

Qg is best used to estimate the amount of gate driver current required to get the desired switching time. Bigger Qg = more difficult for the gate driver to drive which could mean slower switching times. Most of the cheap controllers have really poor gate drivers, so if you switch out FETs I would consider the Qg values if they are significantly different, say a stock 100 nC Qg and replacing it with a 500 nC Qg FET. This might cause slower switch times and result in much higher losses.

I'll try to post some graphs I made which show all of the losses at different switching frequencies and power levels. I believe they nicely illustrate the different types of losses which can be expected. I did a comparison of a high(ish) RDSon 1200V SiC MOSFET vs a low RDSon 600V MOSFET. It's pretty eye opening, at least to me. I'll probably start a new thread on this.

 

[Punx0r]

Please do

 

[bunya]

Thanks for the input, the controller runs quite cool in 15C/60F ambient temp so I should just keep riding, especially after realising I'm picking at straws
I think there is plenty of headroom from a thermal perspective anyway, the case is very large and the walls are thick ally so I could just push the FETs harder and see if they generate a noticeable amount of heat. When I get around to putting the bike on the scope I'll take some screen shots and upload them here in case anyone would like to see the waveform etc.
The quick guide to FETs losses was very helpful, i will check the gate driver by measuring tr and tf to see if it can supply enough juice for acceptable switching times for the larger FETs.

 

[zombiess]

Please do

Started my new thread
http://endless-sphere.com/forums/viewtopic.php?f=30&t=70503