Do I care about Vos if I have dual rail supply?

[Bootalito]

If I have an op amp that has a split rail do I really care about offset voltage in order to get the output voltage to 0.0?
Context: I'm designing a fairly schmick electronic load (EasyEDA.com link to full schematic.  I'm powering the op amps doing the business from a center tapped transformer with a nice clean low ripple linear regulated +15V and -15V.  I'm trying to choose an op amp that will individually control each of the 8 mosfets.  Teardowns of other electronic load have yielded that NE5532's (high slew rate , 0.5mV offset, high performance op amp) and TL071's (Jfet input, 5mV offset, high performance op amp) work in a commercial product.  Scullcom hobby electronics used an AD8630 (expensive, precision op amp, 1uV offset, slow (relativly) slew rate( 1V/us)).  I want to keep costs down as I'd like to make this fully open source and have other people be able to afford the parts, so I want to use an NE5532 or a TL071, but I'm worried about getting down to 0 Vgs on the mosfets so that they are OFF...completely.
If I'm powering these things from a true, clean split rail, what should I care about an awful 5mV offset voltage if the output can just go a bit lower than 0V to shut it off.  Or said another way: if the + terminal of the split rail op amp is grounded, won't the op amp output ALSO be perfectly (within a few uV) of 0V too?  Even if the op amp has an Vos of like 5mv?

Anyway, here is a condensed schematic showing the important parts.  Assume the control signal can get down to 0V as I am actually using the AD8630 as the control signal with an Vos of 1uV. Also note that the actual full schematic is a lot more complicated...this is just a stripped down version for the purpose of this question.

[T3sl4co1l]

Yikes, a slow switching current sink!

Tim

[Bootalito]

Yikes, a slow switching current sink!

Tim

Slow switching because of the AD8630 doing the main control?  Or slow because the AD8630 exists at all? And why is A/us such a banner spec on electronic loads?  Like.. why is 6A/us good, but 1A/ms unuseable? 

[rstofer]

That AD8630 is around $7 at Digikey.  I would be more concerned about the packages than the price.  It seems to be available only in SOIC and TSSOP.  They aren't really hard to solder but they aren't as easy as through-hole DIPs.

Maybe the 'Zero Drift' parameter seemed important in the original design.

If you only need one precision op amp, the AD8628 is more appropriate, comes in a 5 pin SOP or 8 pin SOIC pkg and costs about $2.58

Parts cost simply isn't an issue when building something like this.

[T3sl4co1l]

Slow switching, because the NE5532 isn't terribly fast for ~mV offsets, and ~mV offsets because on average, one op-amp will hog all the current, but they'll likely switch between themselves erratically.

The insight is that the current sense resistor must be per transistor and op-amp, not ganged.

Tim

[Bootalito]

Slow switching, because the NE5532 isn't terribly fast for ~mV offsets, and ~mV offsets because on average, one op-amp will hog all the current, but they'll likely switch between themselves erratically.

The insight is that the current sense resistor must be per transistor and op-amp, not ganged.

Tim

From what I've seen of load teardowns there is only one main low value resistor(0.001R or 0.0005R )providing the overall circuit feedback for eachh opamp/mosfet pair, which are doing local control around a higher value ballast resistor (0.1R ish).  Have I mis-interpreted this type of control scheme, and that this won't work?

See color coordinated opamp/mosfet pairs for clarity:

[Bootalito]

Slow switching, because the NE5532 isn't terribly fast for ~mV offsets, and ~mV offsets because on average, one op-amp will hog all the current, but they'll likely switch between themselves erratically.

The insight is that the current sense resistor must be per transistor and op-amp, not ganged.

Tim

The BK precision 8601 looks like it uses individual NE5532 (6V/us) for each mosfet and achieves a 2.5A/us slew rate.  The rigol uses TL07 (13V/us) and achieves a 6A/us slew rate.  Can I ask what your metric is for fast given this application?

[T3sl4co1l]

The problem is literally feedback from the wrong side... check this:

Also added compensation caps or RCs, to show that.  And the slave opamps won't need 3V of input so a divider is a good idea to prevent the master opamp from commanding a ludicrous setpoint (unless they will, but that's fine too).

Tim

[Bootalito]

The problem is literally feedback from the wrong side... check this:

Also added compensation caps or RCs, to show that.  And the slave opamps won't need 3V of input so a divider is a good idea to prevent the master opamp from commanding a ludicrous setpoint (unless they will, but that's fine too).

Tim

Thanks. That was actually just a mistake. The full schematic LinkedIn the original post had it connected to the correct side of the shunt. Also the full schematic has op amp compensation. The schematic this post has been working off of is Justin trimmed-down version for the purpose of my original question. Which unfortunately still has yet to be answered

[T3sl4co1l]

Thanks. That was actually just a mistake. The full schematic LinkedIn the original post had it connected to the correct side of the shunt. Also the full schematic has op amp compensation. The schematic this post has been working off of is Justin trimmed-down version for the purpose of my original question. Which unfortunately still has yet to be answered

Okay, the link... would've been helpful to include a snippet of that rather than the other thing.

So, let's see here... yes, they do indeed have compensation, very nice!  But... what the fuck?  They literally have the opamps upside down!  Minus to plus!  Positive feedback!  The pin numbers aren't wrong, either!  It's really wired that way!  Who drew this!

It doesn't have a layout either, just a bunch of stuff scattered around... what is this!

Incomplete, I guess, but how many times do you really have to cook and re-cook a stupid LOAD circuit?  I know some companies are very "not invented here"-averse, but the kind of people that keep doing these (and homemade bench supplies, and..), man...

In short: answering the original question is much lower value to you than addressing the mess, that is whichever of these schematics -- the thing has to work before you even need to think about trimming out millivolts and microamperes.  And tempcos and drift and precision.  And anyway, it's a LOAD, what does precision matter?  An SMU sure, but that has inputs and outputs, and wide dynamic range, and programmable control so it can do quite a lot of things, automated.  That's a lot more work than rubbing two opamps together.

So that's fundamentally why you hadn't gotten an answer yet.

But anyway, to answer your question properly: no, you don't need negative supplies to deal with Vos, because Vos is an input-referred spec, not an output spec.  You have to look at V_OL and V_OH for output voltage range.  Typically a RRO amp can swing within 10s of mV of the rail, which is good enough for most purposes.  If you need a very accurate zero, you can just as well bias up the shunt feedback signal instead (which has to be done per transistor-amp, because the master amp can't force them to zero).

Remember that an opamp senses an input voltage difference.  Adding an offset to one input is the same as subtracting it from the other.  This can save you a lot of tiddling with -1.5V supplies!

On that note, I'd be happy enough with TL2372s instead of NE5532s, and eliminate the -15V rail altogether.  No need to burn so much supply current, on an ancient and mediocre audio opamp, and certainly no need of a +/-13V swing at the gates.  A +9/0V swing is more than enough, and helps limit transient short-circuit current draw.

HTH,
Tim

[Bootalito]

@T3sl4co1l
Thanks very much for your thoughtful reply .  I've gone through many design choices for this project so its not surprising that I have my op amps upside down, lol.  My intention is to create a load that is one or two steps up for a standard DIY load complete with full characterization, control via serial commands (including telnet and SSH), a fully interactive webpage, users manual, etc.  I'm using this first project as a proving ground for component refinement and also as a way for me to learn about all of the aspects required to create a product that I hope to sell some day.  My intention is to create a 24-bit version so I'm paying very careful attention to tempco, drift, low noise, separating analog and digital ground, etc.  This is why I've opted for a linear design (with the intention of creating a precision instrument).  I've completed a decent 5V precision reference (https://www.eevblog.com/forum/metrology/5v-precision-voltage-reference/) so I've already cut my teeth on the rabbit hole that is metrology and am well aware of some of the nuances required to create a precision circuit.  You should see my deisgn spreadsheet, lol.  I'm building in a LOT of margin especially for power dissipation and tempco. Now I'm learning about power!  Next is to combine the two

Although I had a thought...I'm selected two different ballast resistors for testing. El cheapo 5W 0.1R cement resistor with a measured tempco of 2500ppm (measured myself)
 (AliExpress.com listing), and a constantan resistor typically used as a shunt AliExpress.com listing.  This one is still in the main so I have yet to characterize its actual temp co.
However...I don't think I care about tempco for the ballast resistors, as the main op amp being controlled by the main current shunt will simply tell all of the mosfets to to increase Vgs as the ballast resistors resistance rises.

Anyway I'm rambling

I do have a question though if you've made it this far.  How do I handle not blowing up my circuit when the user reverses polarity?  I've considered putting a high voltage diode such as the RURG5060 after the current shunt.  I've also considered just adding a full bridge rectifier on the main load to always ensure correct polarity.  But either option prevents my from testing any load less than 1.5V (3V+ for the bridger rectifier!).  I also considered adding a low side reversed n-fet (not sure how I'd turn on the floating fet, more research is needed).  As well as a high side p-fet powered from some charge pump (adds circuit complexity).  Are there any other options I haven't considered?  What do you think would be the best choice?

Thanks for your suggestion about the TL2372's I'll get a handful of them to test as well

[David Hess]

If I have an op amp that has a split rail do I really care about offset voltage in order to get the output voltage to 0.0?

Offset voltage is offset voltage whether the dual supplies or a single supply is used.  Note however that the offset voltage specification is given at a specific supply voltage and common mode voltage and when different supply voltage are used, it will differ depending on power supply rejection and common mode rejection.

In single supply applications where the output cannot actually reach ground or Vcc, the input offset voltage may create a dead zone.  If the output bootstraps the input like with some reference circuits, this may result in the circuit latching off if the input offset voltage is in the wrong direction.  To avoid this, an external offset is added, sometimes through the offset null pins if available, which guarantees that the effective input offset voltage is always the correct polarity.

The Tektronix PS503 power supply deliberately does the same thing with 741/301A type operational amplifiers to guaranty that the output can reach zero volts no matter what the offset voltage is.  They actually go slightly negative.