DSO & AWG Based Curve Tracer

[RoGeorge]

Wonder if a better overall approach would be a pair of Howland Sources, each configurable as a bipolar Current or Voltage source and range scaling. One would be a lower voltage type but higher current, say to >|1a|, the other a high voltage but low current type <|50ma|. Add to this the transimpedance amp and you have a very flexible test function capable of covering a wide range of 2 and 3 terminal components over a wide range of currents and voltages!!

Very tempting! 

The difficulty would be the feedback opamp from the high voltage Howland pump, its positive input will see the full HV swing.

[moffy]

You basically have a Voltage and Current Source you can use as desired, so you could use the Howland Current Source to sweep if desired, and use either the AWG drive, or the Output for the DSO Ch1. We tried to create lots of flexibility for various uses.

JFETs can be traced by Sweeping the D/C and monitoring the S/E current and using the G/B Current Source switched to a Voltage Source with a a negative StairStep waveform for the Gate Voltage, same for DMOS. TRIACs and DIACs should also be available to be measured.

Thought it was interesting in this thread where we were able to show the negative incremental resistance of the reversed biased BE junction of a ordinary 2N3904 NPN in breakdown. Confirmed the incremental negative resistance by adding a simple shunt capacitor which created a relaxation oscillator!!

https://www.eevblog.com/forum/testgear/fooln-around-with-dso-awg/msg4422982/#msg4422982

Best,

Thanks for the link that was very informative, you adjust your test jig to suit the device, makes sense.

[mawyatt]

Wonder if a better overall approach would be a pair of Howland Sources, each configurable as a bipolar Current or Voltage source and range scaling. One would be a lower voltage type but higher current, say to >|1a|, the other a high voltage but low current type <|50ma|. Add to this the transimpedance amp and you have a very flexible test function capable of covering a wide range of 2 and 3 terminal components over a wide range of currents and voltages!!

Very tempting! 

The difficulty would be the feedback opamp from the high voltage Howland pump, its positive input will see the full HV swing.

Great thinking, that's exactly what I've been looking into

Best,

Edit: Attached is a concept that seems to fit the need. It's based upon scaling the Op Amp Vcc and Vee to follow the Output with an Offset set by the zeners.

[moffy]

Just curious, what % accuracy are you looking at for your current source/sink, Howland pump?

[mawyatt]

Overall accuracy should depend on Resistors and Op Amp. Since the DSO is used as a display, anything significantly more accurate than the DSO isn't required. So 3% should work but expect to do much better, but haven't looked much into this yet. The really nice thing about the modified Howland with the positive feedback buffer amp is the removal of the dependance on the sense resistor (ideally), which allows precise current scaling without need for feedback resistor adjustments.

With a good Op Amp(s) and Resistors think you can achieve comparable results to other precision current sources.

Here's an example with simulations with ideal Resistors, pair of OP-07s, Current Sense R 1K and 1K load. Load attached to swept voltage source and current measured. Shows a 1ma Output Impedance of 250Meg Ohms!!

Best,

[RoGeorge]

I had to sleep on it, so here's the new thoughts:
- Howland pump comes particularly useful when the load can not float and must be tied to ground.
- However, the load is not really grounded, but floated at GND level by the transimpedance amp (TIA).
- The voltage output from TIA indicates the current through the load.

So all is left to do is to use the voltage output from the TIA in a control loop together with the high-voltage amp (HVA), so to regulate the loop for constant-current (CC) instead of constant-voltage (CV). 

The advantage would be that the TIA is already referenced to GND, it doesn't have to swing at the full HV range like the feedback opamp from a Howland pump has to.  The TIA input was already protected against HV from the very first design, in case it wants to swing more.

[mawyatt]

You could use the feedback from the TIA to "force" the HV Output Amp to become a current source, might be tricky with all the delays and such and various unknown applied loads.

Just for fun included the HV Follower shown above with the Modified Howland Source as a HV Current Source.

We don't have a model handy for a HV Op Amp like the PA441 or OPA462 so used the OP-07 Buffered Driver Amp. Also, no HV transistor models so used the 2N3904 and 2N3906 since the Transistors and Op-Amp models used don't seem to include voltage breakdown effects.

Anyway, here's the schematic and pulse response.

Best,

[RoGeorge]

That could work, but won't be easy:
- ideally would need for the mirrors 2 pairs of matched HV transistors of at least 300mW
- the bootstrapped supply for the opamp has to swing a lot, about 275V, stray capacitances will see that swing
- will burn some power, 3mA from the mirrors plus, say 2mA opamp idle current, 1.5W of heat to evacuate
- because of the HV swing, will be slower and less stable than a TIA based CC (didn't check yet, only a guess)

A TIA near GND looks more appealing, and I think it would be faster because all its nodes will be close to GND.  There shouldn't be any delays in current, once entered in the DUT through the HV terminal, the current has to go out through the other terminal at the speed of light.

The only case where the already existing TIA can not be used would be when testing a 3 terminals DUT with both sources in the CC mode.  To overcome this, each CC source will have to have its own TIA, so 2 TIA, one for each CC source to sense their corresponding current individually, each maintaining their own virtual potential.

Seems like it should work, but to be sure I'll have to draw the 2 TIAs.


Another thing, I've just checked out of curiosity how sensitive is the Howland pump against resistor's tolerance.  I know it's supposed to be very sensitive, but never looked into the actual numerical values.  Did a simulation today, and even for as little as 1% resistors tolerance, the worst case is close to unusable.



The input stair voltage is 100mV per step.  The worst mismatching case with 1% resistors and 20Vpp swing on the load looks very, very bad.  The current error in the DUT (induced by the DUT's own voltage swing) is as big as it would be when varying the input control voltage with 4 steps (400mV in this numeric example).  Same results with an ideal opamp instead of LM358.

I'm thinking of replacing the Howland CC with something else.

[mawyatt]

The mirror just needs to consume slightly more than the op amp base current and the output current.

The R1/R2 and R3/R4 ratio needs to be matched not the absolute value, only Rs determines the scale error with the mentioned matching. To keep op amp offset error low use as high a Rs as practical which requires a high input voltage.

0.1% resistors are available and not expensive and easily matched to much better if desired.

Agree this is still a somewhat messy solution with all the HV biasing and such.

Best,

[moffy]

Borrowing ideas from RoGeorge, mawyatt and of course J.Williams, I thought I'd contribute my WAY OVERDESIGNED and complex HV precision transimpedance amplifier. It doesn't rely on any Vbe matching for current mirrors, that would be too simple  , it needs 2 pairs of well matched resistors R4/R5 and R7/R8 and limits the error currents by using darlingtons. Only the difference between Q5/Q6 and Q7/Q8 collectors appear at the output, so load currents like the supply current for U1 are eliminated, not original but worth stating.
The complexity comes from the mirrors which need their own opamp and supply, but in the sim with a 4vp-p input it managed 80db linearity or 0.01% at 1kHz. The component choices are just for the sim, not suggested for real life. The current mirror also can provide current amplification as shown easing load and power requirements on U1 and the HV supply.

P.S. Added .asc file as per RoGeorge's recomendation.
P.P.S. It is of course a Transadmittance not Transimpedance amplifier as RoGeorge has pointed out.

[RoGeorge]

Looks good.  I like those two opamps controlling the current-mirrors! 

WAY OVERDESIGNED

That always happens to me.  I start with something very simple, say, adding a series resistor and call it a current-source, then replace it with an active current source, then make it a voltage-controlled current source, then add a DAC to it, then add an ARM microcontroller to control the DAC, then add Linux and write software to control it, then design a GUI for it, then optimise the boot time, then estimating how long it will take to complete all these... 

Then, after agonizing for a full week about all that, just go back to the initial schematic and add a series resistor to it. 


P.S.  When adding LTspice printscreens it's a good practice to also attach the .asc file, so others can simulate it, too.

[moffy]

WAY OVERDESIGNED

Looks good! 


That always happens to me.  Start with a simple circuit, like e.g. just add a series resistor and call it a current-source, then replace it with an active current source, then make it a voltage-controlled current source, then add an ADC to it, then add an ARM microcontroller to control the ADC, then add Linux and write software to control it, then design a GUI for it, then optimise the boot time, then estimating how long it will take to complete all these... 

Then, after agonizing for a full week about all that, just go back to the initial schematic and add a series resistor to it. 


P.S.  When adding LTspice printscreens it's a good practice to attach the .asc file, too.

Added the .asc file as you suggested. Thanks.

P.S. Simulated output impedance is 143M ohms. For a 10V step on the output driving 9.998ma the output current changed by 70na.

[RoGeorge]

Added the .asc file as you suggested. Thanks.

P.S. Simulated output impedance is 143M ohms. For a 10V step on the output driving 9.998ma the output current changed by 70na.

Wow, gave it a run and it works rock solid, congrats! 

That's SMU grade performance.  Gonna use the same idea as a standalone AWG buffer, thank you.  A simple switch can turn the same circuit from a transadmitance amplifier, like it is now (Iout/Vin), into a current-feedback voltage amplifier (Vout/Vin).  Perfect!

[moffy]

Added the .asc file as you suggested. Thanks.

P.S. Simulated output impedance is 143M ohms. For a 10V step on the output driving 9.998ma the output current changed by 70na.

Wow, gave it a run and it works rock solid, congrats! 

That's SMU grade performance.  Gonna use the same idea as a standalone AWG buffer, thank you.  A simple switch can turn the same circuit from a transadmitance amplifier, like it is now (Iout/Vin), into a current-feedback voltage amplifier (Vout/Vin).  Perfect!

Hope it will be of use. Just be careful of C1 it might degrade the AC performance acting as a source/sink for currents rather than the darlingtons. Might want to reduce it in size, not sure.

[mawyatt]

Here's the concept described by RoGeorge (we believe, please correct if incorrect). Using the TIA as a current sense and negative feedback to a HV Amplifier to make the HV Amp "look" like a HV current source. Note this technique does not require a HV op amp for current sensing because the input voltage is forced to virtual ground by the TIA negative feedback and thus the voltage levels are just within the range of Iload*Rsense. Also the resistors are not critical, and Rsense with the ratio of R4/R3 determine the TIA gain which because of the overall negative feedback by the HV op amp is ~ G(Iload/Vin), where G is Rsense(R4/R3). We are just using a buffered OP07 model which has no voltage supply limit for the HV amp, normally this would be proper HV amp, a OPA462 or PA441 for example. Seems to work well as a current source, but no detailed effort has been applied just yet, so an opportunity for some folks to jump in and provide a more detailed analysis!!

BTW Rsense and Rload were swept from 100 to 100K and 100 to 10K respectively for the plot. 

Best,

[moffy]

Here's the concept described by RoGeorge (we believe, please correct if incorrect). Using the TIA as a current sense and negative feedback to a HV Amplifier to make the HV Amp "look" like a HV current source. Note this technique does not require a HV op amp for current sensing because the input voltage is forced to virtual ground by the TIA negative feedback and thus the voltage levels are just within the range of Iload*Rsense. Also the resistors are not critical, and Rsense with the ratio of R4/R3 determine the TIA gain which because of the overall negative feedback by the HV op amp is ~ G(Iload/Vin), where G is Rsense(R4/R3). We are just using a buffered OP07 model which has no voltage supply limit for the HV amp, normally this would be proper HV amp, a OPA462 or PA441 for example. Seems to work well as a current source, but no detailed effort has been applied just yet, so an opportunity for some folks to jump in and provide a more detailed analysis!!

BTW Rsense and Rload were swept from 100 to 100K and 100 to 10K respectively for the plot. 

Best,

It is a very nice concept, but might under certain load conditions run into stability issues because of its closed loop nature. I am sure there are things that could be done for stability, the major difference with my concept is that it is open loop but more complex as a result(to maintain precision), but stability with load should be a given. From the discussion I assume you would be analysing loads other than resistance? Some that might have even negative resistance.

[moffy]

I was unhappy with my previous design in one particular aspect: The input amplifier U1 couldn't be decoupled at the supply pins properly without compromising the current flows and matching for the transadmittance amplifier. Odds are that a fast op amp could go unstable so I am proposing a slight change in the input topolgy that overcomes this limitation. Comparing the schematics should explain it all, if not I could explain any detail needed.

[RoGeorge]

The input amplifier U1 couldn't be decoupled at the supply pins

C1 doesn't have to be there at all.  It shouldn't oscillate without C1, and anyways, those are not power pins any more.  Those pins are now part of the active circuit, and is essential that no other circuit branching can steer away any current from those former power pins.  All the current has to go to the R4/R5, in the external mirrors.

The new power pins are the terminals of the R4/R5/1k at the high voltage.  Decoupling "near pins" has to be done now between the HV pins of those resistor and GND, at the node between R4-V4 and GND, and the other one between R5-V5 node and GND.  When a symmetric supply is used, two decoupling are needed, one for each supply.

Another thing, on my LTspice simulation hangs at 100% without going to the next steps, so I had to add in the schematic an ".options srcsteps=0".  The minus of V4 would be better to be tied at GND, same the plus of V5.  In simulation it doesn't matter, but the physical circuit will be better with V4 and V5 tied to the ground, and not in series with V1 and V2.

Using the TIA as a current sense and negative feedback to a HV Amplifier to make the HV Amp "look" like a HV current source.
...

Yes, indeed, thank you!

I did try to do that yesterday and failed, it was oscillating all the ways.    My try was with 2 opamps, and instead of U3 I've tried to sum the TIA output voltage directly to the U1.  Should have worked, but it didn't.    I have a serious lack of skills when it comes to loops stability.

Then I have to leave it, and today spent some time with the moffy's version of buffer, trying to modify it from a Voltage-Controlled Current-Source (VCCS) into a Voltage-Controlled Voltage-Source.  I was hoping to do that by using current-feedback (CF), so to keep it fast (as an AWG buffer)

With CF the results were not very good.  It was stable, but the Vout didn't follow Vin very well with low impedance loads.  With 100+ ohms it was OK.  For some reason I couldn't reduce enough the output impedance.  I have to understand CF better before fiddling again with it.

The moffy's schematic resembles very well the topology of a CF amplifier (I know them as Norton opamps, not sure if this name is widespread), like shown in Fig.5b here:  https://archive.org/details/edn-1989_01_05/page/164/ except moffy's doesn't have the second voltage follower at the output.

For a VCCS the output voltage follower is not needed, but to make it VCVS I had to add a voltage follower.  It's enough to buffer the voltage for the feedback network only, so a low power opamp should be enough (the very same opamp that measures/display the Vout on the R_load).  However, to make the VCVS immune to load variations I had to drop the CF idea. 



The schematic is drawn with the switch in CV mode.  Notice that for the CC mode, there is no global feedback from load to input, and also no current measurement in the load! 

Hope I'm not wrong with this one, but I think there is no need to have a global feedback in for CC mode, because of the Kirchhoff law in the node "out", and in the node-opamp "U1".

The two current mirrors have their own local feedback loop.  The current-mirrors driven by the hanging rail-to-rail opamps in the moffy's initial schematic (instead of a classic transistors pair mirror) also solve the thermal runaway, and make it easy to set the idle current in the power transistors.

Another advantage would be that nothing swings in high voltage, except the "out" node.  And all is current controlled, which means induced noise and Miller or the stray capacitance's bad influence are kept to a minimum.

Or at least that's how I believe it is (please correct me if wrong).

If it will work in practice as well as in simulation, then two buffers like those can make a dual channel SMU like you said, one with HV and lower max currents, the other with lower voltage and high I, and bot can be tied to ground to know Ib and Ic .

OTOH, would be a pity to not use a ready made HV opamp like that APEX. 

[moffy]

The input amplifier U1 couldn't be decoupled at the supply pins

C1 doesn't have to be there at all.  It shouldn't oscillate without C1, and anyways, those are not power pins any more.  Those pins are now part of the active circuit, and is essential that no other circuit branching can steer away any current from those former power pins.  All the current has to go to the R4/R5, in the external mirrors.

The new power pins are the terminals of the R4/R5/1k at the high voltage.  Decoupling "near pins" has to be done now between the HV pins of those resistor and GND, at the node between R4-V4 and GND, and the other one between R5-V5 node and GND.  When a symmetric supply is used, two decoupling are needed, one for each supply.

Another thing, on my LTspice simulation hangs at 100% without going to the next steps, so I had to add in the schematic an ".options srcsteps=0".  The minus of V4 would be better to be tied at GND, same the plus of V5.  In simulation it doesn't matter, but the physical circuit will be better with V4 and V5 tied to the ground, and not in series with V1 and V2.

Unfortunately they are the power pins for the opamp and I was even able in LTSpice to get opamp power supply oscillations depending on the opamp I used. The LF412 when it first came out was notorious for power supply oscillations if not properly decoupled, and many mid to high BW opamps these days have the same issue. There is another issue with using the darlingtons to provide power to the opamp. The emitters are modulated by the varying current which then can appear across the load depending upon the PSRR of the opamp and the frequency of modulation, this introduces distortion into the output. I was able to reduce the distortion level by 20db by moving the opamp supply out of the current path at 10kHz, from -60db to -80db.

P.S. Meant the LF351 instead of the LF412, memory.

[rhb]

@mawyatt

I wish you could appreciate the trouble you’ve caused me.  I’m in the process of installing ABEC 7 bearings in a small lathe in place of the ABEC 1 taper roller bearings which have ~0.0015” TIR.  The tolerances for the ABEC 7 bearings are +0.0/-0.0002” for the spindle and +0.0/-0.00035”  the housing bore.  Not easy for anyone. Very intimidating if you’ve never done that.  So I’ve been rather focused on that

So after a long hiatus from EEVblog I wander by and am  confronted by your work.  OMG!  Color me extremely impressed.  In one sense you’ve pointed out the obvious, but until someone does, it is not obvious.  Well done! But now I have another project.

Have Fun!
Reg

[mawyatt]

Hi Reg,

Thanks!!

Sorry about that!! Never tried anything like your bearing installation, but seems very difficult indeed!!

And yes, many things are "hiding in plain sight" so to speak, and are obvious after the discovery!!

This whole episode of the Curve Tracer endeavor has led us to acquire an old Tek 577 and refurb it back to life and good working order.

https://www.eevblog.com/forum/repair/old-tek-577-return-to-life-story/msg4479919/#msg4479919


The issue of software development (we're not very good at this) & some reoccurring family issues has kept us from going forward with the full custom approach, so this is on "hold" at the moment, and the refurbed 577 has filled the immediate need for the Curve Tracer.

Here's another related post using a DSO for Impedance Plots you might find interesting.

https://www.eevblog.com/forum/testgear/capacitive-impedance-plots-with-sds2104x-plus-bode-function/msg4335745/#msg4335745

Best,