fet suppliers

[cadstarsucks]

i think cds might have misunderstood my intentions to use 2 of these to replace 2 fets with double the Rds-on. not 1 for 2, and that may have been the basis for his argument that 2 cheaper fes were better.

most arguments are based on misunderstandings.

True and true.

I was under the impression that there were going to be unpopulated locations. That then makes sense. I just wish we could get down to that for 300V and 33mOhm

Dan

 

[bobmcree]

i think when you get up to 300v bipolar still excels. as cds knows, you can switch 100A with a transistor and have less than a volt across it, for <100w dissipation. for a fet to have the same dissipation at 100A it would need to be <10 milliohms and there are no 300v fets anywhere near that.

 

[cadstarsucks]

i think when you get up to 300v bipolar still excels. as cds knows, you can switch 100A with a transistor and have less than a volt across it, for <100w dissipation. for a fet to have the same dissipation at 100A it would need to be <10 milliohms and there are no 300v fets anywhere near that.

Actually we suffer with 2V on IGBTs... Just found a 280V @ 41m FET for $2.90 That gets us down to about 1V and high speed!

Dan

 

[bobmcree]

the old standby MJ10016 darlingtons i was using 35 years ago in the power supplies for mri gradient coils have a saturation voltage of about 2v @20A and will stand off 500v, so i am surprised you cannot do better than that with what is available now, but i admit i have been out of the bipolar high power game awhile.

 

[cadstarsucks]

the old standby MJ10016 darlingtons i was using 35 years ago in the power supplies for mri gradient coils have a saturation voltage of about 2v @20A and will stand off 500v, so i am surprised you cannot do better than that with what is available now, but i admit i have been out of the bipolar high power game awhile.

Darlingtons are slow and that one would require an amp of base drive on top of it. We usually use IGBTs but FETs are catching up now at US line voltage levels at the same price so we are looking at them again now.

Dan

 

[fechter]

They've made steady improvements in FETs over the years. IGBTs are still the in thing for over 150v. That 2v drop is a killer for heat production. I guess that's why most electric car controllers are water cooled. I can't wait until they get perfect transistors perfected.

 

[bobmcree]

the old standby MJ10016 darlingtons i was using 35 years ago in the power supplies for mri gradient coils have a saturation voltage of about 2v @20A and will stand off 500v, so i am surprised you cannot do better than that with what is available now, but i admit i have been out of the bipolar high power game awhile.

Darlingtons are slow and that one would require an amp of base drive on top of it. We usually use IGBTs but FETs are catching up now at US line voltage levels at the same price so we are looking at them again now.

Dan

i agree the beta is low, but with a microsecond inductive switching time they were the best thing available at the time, and are fast enough for a switcher running at 100 kHz easily.

 

[eP]

the old standby MJ10016 darlingtons i was using 35 years ago in the power supplies for mri gradient coils have a saturation voltage of about 2v @20A and will stand off 500v, so i am surprised you cannot do better than that with what is available now, but i admit i have been out of the bipolar high power game awhile.

Darlingtons are slow and that one would require an amp of base drive on top of it. We usually use IGBTs but FETs are catching up now at US line voltage levels at the same price so we are looking at them again now.

Dan

i agree the beta is low, but with a microsecond inductive switching time they were the best thing available at the time, and are fast enough for a switcher running at 100 kHz easily.

So what would be the switching loses at 100 kHz and microsecond switching time ?

 

[cadstarsucks]

So what would be the switching loses at 100 kHz and microsecond switching time ?

That involves the drain source voltage, peak current, rise time, frequency, fall time, gate drive voltage, and total gate charge.

Vd-s*Ipk+Tr*F/2+Tf*F/2+Vg*Qg(tot)*F, approximately.

Dan

 

[eP]

So what would be the switching loses at 100 kHz and microsecond switching time ?

That involves the drain source voltage, peak current, rise time, frequency, fall time, gate drive voltage, and total gate charge.

Vd-s*Ipk+Tr*F/2+Tf*F/2+Vg*Qg(tot)*F, approximately.

Dan

Would you like to use that formula to show us a simple case ?
What would be the loses as a percentage of swiched power ?

 

[bobmcree]

first, the formula given is not applicable to a bipolar transistor like the mj10016 i mentioned.

second, i never suggested this part for a new design. i merely compared the 2v drop of a 35 year old part to those today

third, what the heck does this have to do with reasons not to buy our fets from irtronix?

fourth, i may have "shot myself in the foot" by telling you all about the low price i was getting, but that's ok. irtronix now has them online at 3.20 which is a 20% increase, but that is a single piece price and i expect i can do better on a couple of hundred. fet prices fluctuate with supply and demand, and apparently they saw the demand go up when folks read my post and ordered all they had but the last 8. they will have 2k more soon, and they are now listed on the website. i am glad to let people who might be on a tight budget get a better price.

i am raising the price to replace all 12 fets in a working controller or one with just some blown fets from $80 to $90 because i will have to pay a higher price for the fets than last time justin is going to send people to me who ask him to replace the stock 4710s with 4110s.

fifth, justin at ebikes.ca paid more than this for the ones he has in stock, and his price is quite reasonable considering he paid digi-key prices.

sixth, have a great weekend all, i will not be around for a few days as i am going fishing.

btw, did you hear the one about the chinese guy who called to buy some fets and the price had gone up, so he asked why? the salesman told him "fluctuations in market demand" the asian replied "fluck you americans too!!"

 

[eP]

third, what the heck does this have to do with reasons not to buy our fets from irtronix?

Lets see the case and you will see it have.

 

[Jozzer]

Happy fishing Bob

 

[cadstarsucks]

first, the formula given is not applicable to a bipolar transistor like the mj10016 i mentioned.

second, i never suggested this part for a new design. i merely compared the 2v drop of a 35 year old part to those today

I was answering a question, nothing more, and I figured that the interest would be in dissipation in a FET as opposed to the antiquated bipolar or the IGBTs.

It had the beneficial side effect of sending me off to look into FETs for 110VAC supplies. We ended up finding a few at 250V and <50m.

 

[cadstarsucks]

fourth, i may have "shot myself in the foot" by telling you all about the low price i was getting, but that's ok. irtronix now has them online at 3.20 which is a 20% increase, but that is a single piece price and i expect i can do better on a couple of hundred. fet prices fluctuate with supply and demand, and apparently they saw the demand go up when folks read my post and ordered all they had but the last 8. they will have 2k more soon, and they are now listed on the website. i am glad to let people who might be on a tight budget get a better price.

Which, if you reread the thread, was all I was saying in the first place: that is not a repeatable price.

Dan

 

[eP]

first, the formula given is not applicable to a bipolar transistor like the mj10016 i mentioned.

second, i never suggested this part for a new design. i merely compared the 2v drop of a 35 year old part to those today

I was answering a question, nothing more, and I figured that the interest would be in dissipation in a FET as opposed to the antiquated bipolar or the IGBTs.

It had the beneficial side effect of sending me off to look into FETs for 110VAC supplies. We ended up finding a few at 250V and <50m.

So would you like to apply your formula to that case: VDD=85V ID=20A (for irfb4110) or 2x10A for(2xSTP120NF10 ) f=50kHz ?

 

[bobmcree]

i just checked my mail one last time before heading for the river.

does anyone else wonder why someone who does not even own an ebike and has not expressed any desire to do so would spend so much time arguing on this forum?

i thought the purpose of this forum was to share our ideas to further our enjoyment of our electric bikes and other vehicles. i guess some other people have their own agenda.

the price i quoted WAS repeatable to the people who jumped on it and bought all that were in stock. i could have been stingy and kept it to myself, but i thought we were here to help one another.

if we are talking about production quantities i expect i can get a better price from irtronix by purchasing 100 which will let me rebuild the next 8 controllers. a 20% premium on single piece price over 100 pc price is certainly not uncommon.

anyway, the steelhead are running, my hand-tyed flies are ready, and i am off...

 

[TylerDurden]

the old standby MJ10016 darlingtons i was using 35 years ago in the power supplies for mri gradient coils have a saturation voltage of about 2v @20A and will stand off 500v, so i am surprised you cannot do better than that with what is available now, but i admit i have been out of the bipolar high power game awhile.

Darlingtons are slow and that one would require an amp of base drive on top of it. We usually use IGBTs but FETs are catching up now at US line voltage levels at the same price so we are looking at them again now.

Dan

i agree the beta is low, but with a microsecond inductive switching time they were the best thing available at the time, and are fast enough for a switcher running at 100 kHz easily.

So what would be the switching loses at 100 kHz and microsecond switching time ?

Hmmmm.....

I know nothing about electronics but let's see if I get close on logic:

If a microsecond is .1x one cycle of 100KHz and we're switching On-Off, that could be 20% right there... plus rise/fall time and the rest...

Wild-ass-guess: >25%

 

[cadstarsucks]

So would you like to apply your formula to that case: VDD=85V ID=20A (for irfb4110) or 2x10A for(2xSTP120NF10 ) f=50kHz ?

Ok...

Vd-s*Ipk+Tr*F/2+Tf*F/2+Vg*Qg(tot)*F, approximately.

Oops, forgot the duty cycle on the Vd-s*Ipk product and the Vs voltage and currents on the rise and fall factors, sorry. I was rushing through it while at work. ... Shoot forgot the SQR on the Ipk as well.

You are defining:
Ipk as 20A
F = 50KHz

From the irfb4110:
Tr = 67nS
Tf = 88nS
Vgs = 10V (we will assume that the controller is using the spec test condition, though it could be higher)
Qg = 150nC

For the sake of argument we will assume the motor is going at it's normal 42V speed making the motor voltage 42V with a 50% duty cycle (which will divide the voltage in half) and that the motor normally has 20A in the windings at this point (full load)

applying the formula we get:

0.005(Rd-s)*20A²*0.50 + 85V*20A*67nS*50KHz/2 + 85V*20A*88nS*50KHz/2 + 10V*150nC*50KHz

1W +2.85W +3.75W +.075W

Unless I've missed something else this should be fairly good for a DCB motor. And while the motor will be seeing 20A the battery will only be seeing 10A.

In a BLDC things get a little weird... In 3PH each FET is only on about a third of the time and then PWMed. While the overall motor might be seeing 20A it is not in any one winding.

I fear this has gotten a bit muddied and we can continue if you need it clarified. As you can see it is actually the switching losses that dominate and would be doubled if the frequency was higher.

Dan

 

[eP]

I think we can continue right now.

 

[eP]

Hmmmm.....

I know nothing about electronics but let's see if I get close on logic:

If a microsecond is .1x one cycle of 100KHz and we're switching On-Off, that could be 20% right there... plus rise/fall time and the rest...

Wild-ass-guess: >25%

Thanks TD
That is where i want to go.

Microsecond maybe is ok for signal switching at 100 kHz freq. but is unacceptable long for power switching for the sake of high loses.

So if we want fairly compare different fets configs we need consider switing loses too.

So i'm waiting for cadstarsucks continue.

 

[cadstarsucks]

Thanks TD
That is where i want to go.

Microsecond maybe is ok for signal switching at 100 kHz freq. but is unacceptable long for power switching for the sake of high loses.

So if we want fairly compare different fets configs we need consider switing loses too.

So i'm waiting for cadstarsucks continue.

Ok...Lets see if I can make it clear since I have a little more time for the moment...

Ploss = Pon + Pr + Pf + Pg

=I²Rds*dutycycle + Vds*Ids*Tr*F/2 + Vds*Ids*Tf*F/2 + Vg*Qgtotal*F

The duty cycle does not effect the current, the motor load does. But, the duty cycle does effect the motor voltage, CEMF, which is proportional to speed.

Ploss = 20²*0.005*0.5 + 85*20*67nS*50Khz/2 +85*20*88nS*50KHz/2 + 10V*150nC*50KHz

= 1 + 2.85 + 3.75 + 0.075
=7.68W

When the switch is off the motor current is either going through a diode or another FET so you can expect this much loss again on a FET or even more on a diode, something on the order of 5W conduction losses for a diode (0.5V*20A*0.5) plus whatever the switching losses are for the diode. I think most controllers use FETs as "synchronous" rectifiers.

Again this is not going into the complexities of BLDC, where the switching losses would be applied to all six FETs and the better controllers would be using the FETs as rectifiers as well.

As you can see the overall dissipation of a controller can be fairly high.

Dan

 

[eP]

Thanks Dan

Before we go to the BLDC complexities would you like to show us similiar calculations for 2 fets equivalent case ?

What would be (how long will be) tr and tf times when the driver will drive 2 fets (STP120NF10) at once instead one irfb4110 ?

 

[fechter]

When the switch is off the motor current is either going through a diode or another FET so you can expect this much loss again on a FET or even more on a diode, something on the order of 5W conduction losses for a diode (0.5V*20A*0.5) plus whatever the switching losses are for the diode. I think most controllers use FETs as "synchronous" rectifiers.

Again this is not going into the complexities of BLDC, where the switching losses would be applied to all six FETs and the better controllers would be using the FETs as rectifiers as well.

Dan

That would make sense, but all the ones I've seen just use the body diodes in the FETs and don't do synchronous rectification. Cheap design.

Despite this, the overall dissipation on a Xlyte controller with 4110's is very low. I think about half the heat is from the linear voltage regulator.

With the stock 4710's, it runs much hotter, all other things being equal. This leads me to think that the on resistance is the most significant sorce of loss. Switching loss seem to be very small. Conduction loss during freewheel would be high, but the duty cycle for that is quite small, so overall it doesn't seem to be a big factor.

 

[cadstarsucks]

That would make sense, but all the ones I've seen just use the body diodes in the FETs and don't do synchronous rectification. Cheap design.

Despite this, the overall dissipation on a Xlyte controller with 4110's is very low. I think about half the heat is from the linear voltage regulator.

With the stock 4710's, it runs much hotter, all other things being equal. This leads me to think that the on resistance is the most significant sorce of loss. Switching loss seem to be very small. Conduction loss during freewheel would be high, but the duty cycle for that is quite small, so overall it doesn't seem to be a big factor.

We are in a gray area... If the controller is PWMing to keep the current down the switching losses will be there and related to PWM frequency. Conduction losses will be based on the duty cycle and diode losses will be 1-duty cycle.

If you are going slow with a heavy load the diodes will be at 0.8V*20A*0.1= 1.6W for the 4710 and 0.6*20*0.1=1.2W for the 4110 plus switching losses which are about half in the 4110. How hot it actually gets depends on how good the thermal design is, air temperature, surface area, and air flow.

Dan