mosfet or bipolar?

[grouchobyte]

I am currently doing a design project and need some advice. Good thing I am being paid for this 'cause it ain't always fun.

Its a transconductance amplifier capable of +/- 100 Amps with 10 volts of maximum compliance. Input supply voltage needs to be minimally above or below the load voltage to keep power dissipation low ( like a a volt or two). Bandwidth is DC- 1 Mhz ( yes, you read it right, DC) with decent step response( minimal under and overshoot). Capable of driving precision shunts from uohms to several ohms. System-wide in the high current ranges there are multiple positive and negative cells connected at their collective  outputs( no pun intended) with great attention paid  in keeping parasitics in check especially inductance ( L*di/dt).

Each current cell is capable of producing +/- 5 amps and must have low output offset (.02%  of range ) or nullable offset with low drift coefficients to maintain system stability requirements. There are other mind-numbing requirements but I wont bore you with them ( those power line metrology folks can be a tad  )

The basic topology has many variants and each one has it drawbacks and benefits. I have achieved great (simulation) results with a high speed CFB THS3095 driving a MOSFET in a tight control loop and have excellent linearity, step response, nullable offset with acceptable drift and decent phase margin with excellent stability over a decade of current. Lower current ranges are achieved with a bunch of paralleled opamps. The problem is with the mirrored version of that cell design in the negative current cell design

Using a bipolar or bipolars as the pass element(s) is certainly more stable then messing with MOSFETs but the high drive requirements and loop response suffer, plus I need to current sense in the load side which is problematic for numerous reasons. MOSFETS are great for switching but for linear control in a fast transconductance application I am on the fence about which is better.

Question: Mosfet or Bipolar?

[T3sl4co1l]

Bipolar. Keep the drive circuit impedance low, because the impedance isn't magically high just because you threw a little hFE at it.   Use a Darlington topology (think: more buffering!) if necessary.

If it's for testing, really, don't worry about power... throw huge heatsinks at it.  Accuracy first!

Tim

[David Hess]

I built something similar in the past using high power current mirrors but the current level was only 5 amps or so and the voltage was much higher.

For this I would probably get myself into trouble using a common source power MOSFET to drive a common emitter bipolar transistor (or maybe parallel transistors with emitter ballasts?) with both a base-emitter shunt resistor to control the bipolar turn-on and single source/emitter resistor to control the overall transconductance.  Of course then it has to be mirrored for bipolar operation.

5 amps per cell seems low given the mechanical complexity of assembling more cells while maintaining low parasitics.

[Marco]

Its a transconductance amplifier capable of +/- 100 Amps with 10 volts of maximum compliance. Input supply voltage needs to be minimally above or below the load voltage to keep power dissipation low ( like a a volt or two).

No idea about the transistor, but I'm curious about these requirements ... how do you plan to do 10V compliance and tight (tracking?) supply rails around the output voltage at the same time?

[David Hess]

No idea about the transistor, but I'm curious about these requirements ... how do you plan to do 10V compliance and tight (tracking?) supply rails around the output voltage at the same time?

I took that to mean that the *output* compliance needs to be +/- 10 volts.  The fixed power supply voltages need to be higher to account for dropout voltage of the output stage.  So minimizing the dropout voltage of the output stage by design will minimize the needed power supply voltage and minimize the power dissipation which is a limiting factor.

[grouchobyte]

Its a transconductance amplifier capable of +/- 100 Amps with 10 volts of maximum compliance. Input supply voltage needs to be minimally above or below the load voltage to keep power dissipation low ( like a a volt or two).

No idea about the transistor, but I'm curious about these requirements ... how do you plan to do 10V compliance and tight (tracking?) supply rails around the output voltage at the same time?

If the shunt load is say 2 ohms and the current is 5 amps measured at the load with an average, envelope or peak detector, I plan to just servo the supply voltage down to a couple of volts above that so for 10 volts of output compliance I would servo to 12v. For a 1 milliohm load with 5 mV of compliance I would servo the input voltage to say 2 volts or whatever my pass element needed. Of course some if not most of the compliance is needed to overcome the parasitic inductance in connecting to the load because all connections have some inductance. I plan to use coax to connect to the load, hopefully w/less than 50- 100nH overall or few volts lost in the connections.

[MagicSmoker]

I believe this is what the military folk might call "an opportunity to excel"...

I wouldn't say BJTs for the power amplifier are a slam dunk, mainly because high current and high f_t tend to be mutually exclusive. Conversely, while there are MOSFETs designed for linear operation (see, for example, IXYS L2 series), slewing such a hugely capacitive load as a big MOSFET's gate at >60V/us is going to take huge currents on its own. Not to mention stabilizing the control loop.. Oy.

So might I recommend "cheating" by using a power op-amp designed for such applications? For example, https://www.apexanalog.com/products/pa107.html.

[Marco]

If the shunt load is say 2 ohms and the current is 5 amps measured at the load with an average, envelope or peak detector, I plan to just servo the supply voltage down to a couple of volts above that

So it's acceptable to have a large signal bandwidth much smaller than a MHz? (With some wacky distortion if you try to exceed it.) Since you mentioned negative currents you'll need 2 servo'd power supplies BTW.

PS. AFAICS the biggest advantage of BJTs is the biasing in push/pull, bias circuit for MOSFET push/pull is a PITA.

[T3sl4co1l]

Making a class G or H or whatever system just sounds like inviting disaster, or at least making more trouble than it's worth.  You get much higher, and much more variable, Ccb at low Vce, so your loop has to be slower to compensate for it anyway.

Face it, you're making a 1kW amplifier.  It's going to get hot.  It's going to be big.  It has to be big!  No, it literally must be big: to get low inductance, you need many transmission lines connected in parallel, or very wide ones.

I don't know of anyone who makes coax that'll be nearly low enough impedance to be worthwhile.  You'll burn your inductance spec with just a few inches of 50 ohm stuff.

Purchase a box of two-sided copper clad.  A hunk of that makes a parallel-plate transmission line with an impedance of a few ohms.  Stack all your transistors along one edge, and connect the load to the other!



Regarding device choice: BJTs have more gain period, and you need all the gain you can get.  You do need high current AND fast devices, but common audio parts will suffice -- they often have fT peaking over 60MHz.

Looks like D44VH10/D45VH10, 2SC5001/2SA1834 (caution, low voltage), 2SC6017/2SA2169 and PHPT60410NY/PHPT60410PY are reasonable candidates.  The SMT ones you'll obviously need more of, unless you're okay with pulsed operation -- which is an interesting way to go, and should be good enough for something like this, really.

Otherwise, using more devices means potential bandwidth savings, as you could use lower current, higher fT devices maybe.

A classic audio part would be Sanken 2SC2921/2SA1215.  A few bucks, but made to dissipate power, and run in a linear amp.  Not as fast as the others.

Offhand, RF transistors that are suitable would be,
MRF1150MB, MRF454, MRF455, MRF429
They're >=10A each, and >$30 each.  You'd need a few dozen, so it's not a cheap idea, but it would be more than fast enough.  Keeping them stable, and supplying fast enough drive, would probably be... interesting.  But hey, I'd accept the challenge!

Tim

[grouchobyte]

Making a class G or H or whatever system just sounds like inviting disaster, or at least making more trouble than it's worth.  You get much higher, and much more variable, Ccb at low Vce, so your loop has to be slower to compensate for it anyway.

Face it, you're making a 1kW amplifier.  It's going to get hot.  It's going to be big.  It has to be big!  No, it literally must be big: to get low inductance, you need many transmission lines connected in parallel, or very wide ones.

I don't know of anyone who makes coax that'll be nearly low enough impedance to be worthwhile.  You'll burn your inductance spec with just a few inches of 50 ohm stuff.

Purchase a box of two-sided copper clad.  A hunk of that makes a parallel-plate transmission line with an impedance of a few ohms.  Stack all your transistors along one edge, and connect the load to the other!



Regarding device choice: BJTs have more gain period, and you need all the gain you can get.  You do need high current AND fast devices, but common audio parts will suffice -- they often have fT peaking over 60MHz.

Looks like D44VH10/D45VH10, 2SC5001/2SA1834 (caution, low voltage), 2SC6017/2SA2169 and PHPT60410NY/PHPT60410PY are reasonable candidates.  The SMT ones you'll obviously need more of, unless you're okay with pulsed operation -- which is an interesting way to go, and should be good enough for something like this, really.

Otherwise, using more devices means potential bandwidth savings, as you could use lower current, higher fT devices maybe.

A classic audio part would be Sanken 2SC2921/2SA1215.  A few bucks, but made to dissipate power, and run in a linear amp.  Not as fast as the others.

Offhand, RF transistors that are suitable would be,
MRF1150MB, MRF454, MRF455, MRF429
They're >=10A each, and >$30 each.  You'd need a few dozen, so it's not a cheap idea, but it would be more than fast enough.  Keeping them stable, and supplying fast enough drive, would probably be... interesting.  But hey, I'd accept the challenge!

Tim

Tim

You are one smart Analog Engineer..  ....and I dont hand out compliments easily.  I have been designing analog circuits professionally for over 35 years and your comments are dead on. Thanks for your sage advice.....much of which I am already looking into or realize, as darth vadar would say....accepting my destiny.

The d44vh10 and d45vh11 are the preferred bipolars ( you zeroed in on the same part i did...amazing)
I have abandoned the cfb drive and mosfets, not enough phase margin, too much offset and too much of a tendency to oscillate.

Yup, the low inductance connections are going to be challenging if not impossible. Good thing I spent 3 years working as a signal integrity engineer at Tektronix.....

Tim, your comments really impress me. Perhaps we can PM offline and/or collaborate some more. We certainly think the same way. I do this for money and I know you do too.

@grouchobyte

[T3sl4co1l]

Sure, no problem

Tim

[Kleinstein]

Keeping the inductance low might be difficult. So think about a relatively compact design and maybe water cooling.
Class G or H is not helping much at such a low voltage. So avoid the extra trouble that comes with it, even if this in only having a larger number of transistors and thus a larger layout.

[danadak]

If you look at power darlingtons, say an 8A device, Ft ~ 4 Mhz, so at 1 Mhz,
lets say you want to settle in 10% of the time, 100 nS settling to some spec,
say 1%, pretty challenged with a 4 Mhz Ft. Could investigate SiC based bipolar
darlington, not sure.

Then of course you have to live with its lousy compliance impact, ~ 2V Vcesat.

My inclination would to work out the MOSFET case. And as prior poster stated,
less cells overall. Still a challenge due to Miller. High power LDMOS ? But freq
range an issue.


Regards, Dana.

[Marco]

Making a class G or H or whatever system just sounds like inviting disaster, or at least making more trouble than it's worth.

Depends on what you think the entertainment value is worth

It's an interesting challenge. How about a shunting approach? Make a bipolar buck based current source to deliver an approximation of the necessary current and have the amplifier shape it into the desired current? Might need to go digital so you can put some delay between the buck and the amplifier.

[grouchobyte]

Thanks for all your comments and suggestions.....all good!
The deadline is fast approaching and I am contempating the various approaches. The real compromise will proably end up being something like this

bipolar pass element(s) driven my a small mosfet.
Multiple pass elements in parallel to share the load
More like 2-3 volts across each cell instead of only say a volt
Servoing the supply voltages(+/-) based on peak detection or envelope detection on the output
Current sensing in the pass element input side to optimize the control loop speed
Multiple low Z  copper paths in parallel as Tim W suggests to minimize inductance effects with paralleled cells

May the force be with me.
-yoda