DIY-SMU Project

[djerickson]

It should be OK to use a fixed 1:2 gain for the DAC and ADC part. This would reduce the full scale range at the regulator / diode OR part to something like  4.5 V instead of 5 V.  To still adjust the measurement and adjustment range, one would still need one point to adjust to gain, e.g. at R64,R65 So these would have to be a little (some 10%) higher. The current ranges (or shunts) would also change a little.

Skipping the extra gain / attenuation for the DACs / ADC, and running the regulator / cross over part with a 2.5 V level would compromise the regulation a little, but could simplify the circuit a little and a void one gain/attenuation stage.

Even than the measurement part is not made for highest precision, as there are quite some resistive dividers / gain stages in the signal path. So don't expect the SMU to get an accuracy level of a 7 digit DMM in the voltage ranges. The SMU circuit just needs more compromises to get the high grade current measurement and the large voltage range at the same time.

Also the performance at the very low currents is not the same an electrometer - there are quite some switches as parallel path to the 5 M resistor. So there is a limit to what resistor grade makes sense there.  For the lower value shunts good resistors make absolute sense. Here much is about having good power handling capability, as they may run a bit hot. Compared to a more normal DMM the votlage drop is quite large. So the S102 would be too low in power for the 50 Ohms. The 1 W power rating for the 100 Ohms resistors R40, R41 is still on the low side. I would not be ashamed to use 10 W rated parts (e.g. wire wound or with heat sink mounting) there.

For better performance towards higher currents (e.g. > 10 mA) one could consider gain, e.g. a INA  (maybe 2 gain settings) instead of what is now U21.1 + U21.2.

Hi Kleinstein,

Thanks,  good ideas. Yes, make the force DACs a more convenient +/-5V, then increase the gain in the measure circuits. But somewhere I'd need to

I'm getting <10nA low-current measure and force performance on the 1uA range at room temperature so I'm pretty happy with that.  Driving a 610B electrometer you can see the 33pA LSB steps. I'm pretty happy with that. Slowing the ADC for low currents should improve the measure noise.  Yup, the many switches across the 5M shunt R43 are the limiting factor. The CMOS switches and K2 have quite low leakage at room temp and 5V. If necessary, I could possibly select parts here. Ugh. 

For the 100mA shunts R40/41, 0.25W dissipation in a 1W resistor is decent margin on low-tempco resistors. Yes,  it could be better.

The - side of the current measure circuit U21 only has to measure ground thru a 100 ohm CMOS switch. Works pretty well with 100K Rs and without a 3-opamp instrumentation amp. With one I could lower the U21 resistors from 100K to 10K which wold help noise a bit.

Again. thanks, I appreciate the great feedback.

[RikV]

i'm ashamed to admid it: I can't find a way to get those python scripts running in Win8.
I installed Python 3.10.3 and I try to run a script through the command line (clicking on the filename very briefly flashes a DOS cmd window). (I also tried the same thing from the Python prompt >>>)
Immediately complains about a missing module 'matplotlib'. Installed (tried to...) it. got warning: wheel package is not available.(!!!???)
.py of course still won't run.
Isn't there a 'simple' way to get the whole thing installed in a user-transparant way? This whole process reminds me so much of my DOS-days. Why???
Could someone pls take me by the hand?

[Venturi962]

A few notes:

When running scripts, try to run them from a command prompt with 'python file.py' if the file errors, then the error will be available for debugging on the prompt.

Your error 'wheel package is not available' means you need to install wheel: 'pip install wheel'. This thread may also help https://stackoverflow.com/questions/64285251/problem-with-installation-of-matplotlib-in-python

Looking at the 'smuVoltCal1.py' file you'll also likely need to install numpy and pyvisa modules (and a VISA backend for pyvisa).  Use 'pip install numpy' and 'pip install pyvisa'  The script is setup to look for a HPAK 34401A on GPIB23, so this may need to be modified to your setup.

[RikV]

Dave, the pinout of U25 (Si8422) in the schematic AND the layout is wrong! In and out are swapped. Pins 2 & 3 and 7 & 6 should be swapped.

[SebastianH]

Thank you very much for sharing your great work, David!

Although I'm not ready to build as complex of a device as a SMU, I'd really like to build one in the future. For now I'm very interested in the V/I crossover circuit design of this SMU.
I simulated such a crossover circuit (as an addition to my DC load circuit) and it generally seems to work fine - but with some limitations, especially for my application: It slows down the transition between two different current setpoints (force) quite significantly even if the voltage is between CL- and CL+, because the clamping circuits always work against the output of the "force amplifier" as soon as there is an error voltage (the closer the voltage to the clamping voltages, the more the output is affected).
Am I missing something here?
In its power supplies and DC loads HP/Agilent often uses separate integrators and then ORing those afterwards, so I guess this problem wouldn't occur there, although then windup might be an issue.
I'd be interested in your experience with this circuit in the SMU, especially regarding changes of the setpoint.

[RikV]

Dave, have you ever considered using an LT1991 or LT1997-1? They claim 0.04% or 0.01% and better than 1ppm/K. They are in stock at Mouser and at less than 9€ they are cheaper than a combination NOMCA-OPAMP (NOMCA is NOT in stock and will take forever to be supplied). I believe your "drift sheet" would much improve by them.

[Kleinstein]

The ready made difference amplifiers are definitely an option at a few places.
Suitable resistor networks (4 x 10 K or 8 x 10K)  are also available at mouser.
Here NOMCT (NiCr resistors) is lower noise than NOMCA (TaN resistors). The scaling part for the ADC and DAC does no need high accuracy, just good stability.
Other types are also available.

[ducreux92]

Hi, i fix your attention about this point : the inputs of ad7190 are normaly 0 to 5V but
they are no protection and in some cases the amplitude can vary of -5V to 10 V !!! (see schematic).

 by by ad7190 ...

Likewise U4 is powered with 0 +5V but inputs accepts -5V to 5V  (up to -15V to +15V without software protection )

[Kleinstein]

U4 (TLC2272 or similar rail to rail output precision OP-amp) provides the protection for the AD7190: it can hardly output more than its 5 v supply.  The input to U4 only comes though the resistor and this protects the OP, probably to better than +-50 V.

[Roehrenonkel]

Hi. The slow control response is probably because with no ADC, the ADC read software is timing out, making the main loop s-l-o-w. Comment out the ADC and things should get nice and fast. The latest firmware has all the switches working except PREV and NEXT.  Maybe you are using old firmware?

I have not finished the Main Board EEPROM HW or SW. It was intended that each main board would carry it's own calibration. Having EEPROM on the CPU is good enough though.

NPLC is not implemented yet, its coming....

I find that most 12V fans will run slow at 5V. Try another. The one I specified works well.

Mouser has Vishay MPM 10K/10K parts in stock. I just ordered some, and also some 0805 5ppm resistors.

Best of luck! If you need a DIY-SMU front panel to hold your CPU, Nextion, and controls, let me know.

Dave

Hi Dave,
 
yes, i used the first/oldest software. Without the ADC-subroutine it runs much smoother.

Another suggestion:  command to switch between Local- and Remote-operation
(Gpib "LLO" Local-Lock-Out and "GTL" Get-To-Local afaik).

Thanks for the offer on Front-panels. I am not sure if i even need the display and use different switches (still have two hands-full of them).
The supply-situation is a mess. And no (HG-) wetted relays anymore. What has the world come to?
Is there a (timing-) problem when using mechanical relays instead of the ADG400-switches?

Have you done some tests on your four units regarding resolution and repeatability?
 
Many thanks for your work,
best regards
 

[Kleinstein]

For the current range switching mechanical realys (e.g. like the EC2 type aready used for the external sense, or similar Fujitsu FTR-C1) should work as well.
Relays would be more a thing to replace the photomos part, as these can have some leakage that can effect the very low currents.
CMOS switches are still avialable and there are many even pin compatible parts with very little difference in the parameters. With a little change in the SW the other polatiry may be an option, and with a slight change in the circuit also a 4:1 or 8:1 MUX chip like DG409 or similar.

Especially with relays one can simplifiy things also a little and get away with less switches. The idea would be to have the shunts in series and mainly switch the current path and have the sense path through some of the unused shunts. So only 1 switch per shunt. Thus is alt least OK for the shunt smaller than some 10 K. With the larger shunts the series resistance would add some noise.

Another alternative would be JFET switching - still smaller and cheaper than a relay.

[djerickson]

Dave, the pinout of U25 (Si8422) in the schematic AND the layout is wrong! In and out are swapped. Pins 2 & 3 and 7 & 6 should be swapped.

Yes you are correct.  I discovered this recently. U25 should be Si8423 or Si8424, not Si8422. Sorry for the confusion. If you have Si8422's the rework is fairly simple. 
BTW I have not yet implemented this circuit or the Firmware to display clamp conditions. I think ideally there should LED(s) on the front panel to display ON/OFF and Clamp. Without the front panel LEDs, I plan to display Clamp on the LCD.
Dave

[djerickson]

Hi, i fix your attention about this point : the inputs of ad7190 are normaly 0 to 5V but
they are no protection and in some cases the amplitude can vary of -5V to 10 V !!! (see schematic).

 by by ad7190 ...

Likewise U4 is powered with 0 +5V but inputs accepts -5V to 5V  (up to -15V to +15V without software protection )

The AD7190 ADC input voltage range is protected by the output clipping range of the 0.0V to 5.0V RRIO opamp feeding the ADC. Should be fine, no?
Thanks,
Dave

[djerickson]

For the current range switching mechanical realys (e.g. like the EC2 type aready used for the external sense, or similar Fujitsu FTR-C1) should work as well.
Relays would be more a thing to replace the photomos part, as these can have some leakage that can effect the very low currents.
CMOS switches are still avialable and there are many even pin compatible parts with very little difference in the parameters. With a little change in the SW the other polatiry may be an option, and with a slight change in the circuit also a 4:1 or 8:1 MUX chip like DG409 or similar.

Especially with relays one can simplifiy things also a little and get away with less switches. The idea would be to have the shunts in series and mainly switch the current path and have the sense path through some of the unused shunts. So only 1 switch per shunt. Thus is alt least OK for the shunt smaller than some 10 K. With the larger shunts the series resistance would add some noise.

Another alternative would be JFET switching - still smaller and cheaper than a relay.

Thanks, Kleinstein.  When I was doing the design, I saw that the K236 has a fairly complex method to ramp the high current range JFET switches ON and OFF. I believe this is to compensate for the slower relay control used on some current ranges. Can't have break-before-make. I was trying to avoid this complexity.
On the K2400, I don't see any low current relays. They figured out how to do 2A down to 1uA ranges without relays. I'd like to know how they build 2A range switches with sub-nA leakage. That's part of their magic.
 

[Kleinstein]

For high current electronic switching the parts of choice are 2 MOSFETs in series to get around the parasitic diode. For low leakage one could force the center pin to essentially the same votlage as the critical side. So the main part of the leakage would be to the center and the critical MOSFET would see only some 100 µV of so.  So one could get low leakage switching, but it is quite some effort.

Make before break could be achieved by the logic, though this means 1 relay per range and hard to save on the relays by using more contacts.

[djerickson]

Hi Dave,
 
yes, i used the first/oldest software. Without the ADC-subroutine it runs much smoother.

Another suggestion:  command to switch between Local- and Remote-operation
(Gpib "LLO" Local-Lock-Out and "GTL" Get-To-Local afaik).

Thanks for the offer on Front-panels. I am not sure if i even need the display and use different switches (still have two hands-full of them).
The supply-situation is a mess. And no (HG-) wetted relays anymore. What has the world come to?
Is there a (timing-) problem when using mechanical relays instead of the ADG400-switches?

Have you done some tests on your four units regarding resolution and repeatability?
 
Many thanks for your work,
best regards

Thanks, Roehrenonkel.
As far as Local / Remote, I am hoping to avoid the mode switch and the controls and displays needed for it. I am trying to simplify things, but maybe I'm missing something.
My plan is to allow instrument settings to be done any time by either the operator or SCPI. Let the operator beware.  The only place that a real conflict exists is when doing ADC measurements. These take some time time and need to be synchronized, particularly to SCPI requests.
I was thinking of something like 'Scope Auto-Trigger'. The ADC re-triggers automatically, making measurements continuously. If a SCPI Read occurs, it takes priority, then the instrument waits (like ~1 Sec) for another SCPI Read. If no read occurs before that time, the instrument reverts to continuous reads. So SCPI has priority. This is so an external host can perform a fast sequence of Force/Measure operations. This is where the instrument needs to operate at its fastest speed.
Currently the firmware measures both current and voltage every pass, which slows things down 2x. After a SCPI Read, only V or I would be read to save time. I'll need to blank the other one on the display.
I'd be interested to know what you or others think of this approach. 
Thanks,
Dave

[Bansci]

Hi Dave,
I didn't take up your offer of a unit to program as I'm in the UK and this is probably too much hassle for you, but I thoroughly intend to build my own unit and I'm currently winding transformers.

About the programming, with a micro I would use interrupts to continuously read ADCs into a circular buffer, regardless if the system is triggered. Not having to change the read-store code depending on mode/request is a great simplification as the system is always triggered. You also gain precise regular timings this way too as the interrupts are simple.
 
The control logic becomes much simpler too. Say a single SCPI read is called, you can instantly return the last reading taken. If you want a batch of readings, you can return the past N readings OR, if you want to trigger N readings, you can reset the place of the write head into the circular buffer so that it is N readings from the end and have a special event for when it "wraps" around the buffer.

Averaging is easy with a circular buffer as its the same as "last-N" code, UI updates can be done in the main loop separate from the reading and run outside the interrupts so that its done when possible.

For my own version of the SMU, I've written a Keithley 2450/DMM6500 style interface in python/html to run on a raspberry pi touchscreen (https://github.com/toastedcrumpets/keitha). I haven't made a demo/video of it yet, but I will soon. I am amazed at your own module based interface, this seems like a lot of work!
All the best,
Marcus

[Kleinstein]

For the ADC it can make sense to run the ADC with a slightly higher data rate, like 50 or 60 SPS and thus just fast enough to get mains hum suppression and than do some averaging by hand instead of a slow conversion mode from the ADC. The SINC3 or SINC4 filter looses quite some time (3 or 4 readings) for settling and faster readings can reduce that settling. So chances are the normal mode would be some 4 samples waiting for settling and some 3-6 samples averaging and than switching to votlage / current.

I would consider the mode of reading the current and voltage the normal case. Contineous conversions and reading the last available reading makes sense and for the normal display I don't think one would need it faster.
Under SCPI control it would make sense to also have a mode for reading only voltage or current. This should be more than just 2 x faster, more like 3 x due to the time lost when switching and possibly a faster reading rate (e.g. 50/60 SPS without extra averaging or even faster). The speed of the voltage / current regulation can be somewhat limiting, but it can still be interesting to also see the transients as a kind of self test. Some synchronization to the DAC steps would be nice, but the speed is limited, both from the ADC and the regulator.

[djerickson]

Right! I forgot that the ACD is more than 2x faster when reading one channel vs. 2 due to settling issues. I'll still need to optimize the ADC for fast settling time if I want to: 1) Set DAC, 2) Wait for instrument settling, 3) Measure ADC, 4) Loop. This seems like a general case for a SMU doing VI curves and curve tracer. A real curve tracer does a 10 curve plot in significantly < 1 Sec. Can't go that fast, but don't want to be toooo slooooow.

[Kleinstein]

The ADC has many speed settings and if needed can read quite fast. It is just the quest on how fast to go. The critcal steps is 50/60 SPS to get mains hum suppression. Faster conversions are possible but also more susceptible to hum.
With hum suppression one would need at least some 100 ms per reading (80 ms settling+1 reading of 20 ms). A reasonable curve needs some 50 points and this some 10 seconds.  Anything faster may have to compromise on hum suppresion. AFAIR the AD7190 does not support a simple SINC filter and thus simple integration mode with no extra delay.

The SMU part itself should be fast enough to get settling on a few ms at most. So a fast reading is possible at least for the less sensitive ranges.
For the SCPI readings the speed would be a parameter to change.

[Roehrenonkel]

Thanks, Roehrenonkel.
As far as Local / Remote, I am hoping to avoid the mode switch and the controls and displays needed for it. I am trying to simplify things, but maybe I'm missing something.
My plan is to allow instrument settings to be done any time by either the operator or SCPI. Let the operator beware.  The only place that a real conflict exists is when doing ADC measurements. These take some time time and need to be synchronized, particularly to SCPI requests.
I was thinking of something like 'Scope Auto-Trigger'. The ADC re-triggers automatically, making measurements continuously. If a SCPI Read occurs, it takes priority, then the instrument waits (like ~1 Sec) for another SCPI Read. If no read occurs before that time, the instrument reverts to continuous reads. So SCPI has priority. This is so an external host can perform a fast sequence of Force/Measure operations. This is where the instrument needs to operate at its fastest speed.
Currently the firmware measures both current and voltage every pass, which slows things down 2x. After a SCPI Read, only V or I would be read to save time. I'll need to blank the other one on the display.
I'd be interested to know what you or others think of this approach. 
Thanks,
Dave

Hi Dave,

On a Range-change upwards the value will be multiplied by 10.  Not so safe.
Is it because of my dry-run without ADC-/DAC-chips?
"Let the operator beware." ..even with +-150 Volts?
Operators get tired and make mistakes, automated tests can run errorfree for ages.

Local/Remote:
Imho in Remote-mode no user-inputs except the "Local/Remote"-Key should be read.
This is to prefent intended or unintended user-inputs messing up messurements.
I see that it is not too easy to implement in the current versions (hw+sw).
An indicator and switch plus the software is needed here.
For "Output On" a LED (at Interface-potential) also would be nice.

Maybe consider two modes (Fast/Precise).
As long as no clamp-condition is set the force-value
doesn't have to be read, it equals the Set-value, correct?
In precise-mode all monitors would be read with the highest resolution.

A HW-Lockout can easyly be made by a switch/link in the supply-line to the
Output- and Sense-relays and is usable for Emergency-stop, no fixture connected,
safety-hood (like on the Tektronix 370/576 curve-tracer) not closed.

V-ref+: I'll buffer the positive V-ref with an OP for the dynamic loads (ADC/DAC).

Finally i've found relays that meet the microvolt- and picoamp-requirements.
Minimum switching-voltage and -current are rarerly given in data-sheets.
But i hope these mercury-wetted reed-relays made in 1985 will do the job for me.
Have to put them on an external pcb for the upright/vertical orientation.
One more board, but gives more flexibility too.

[djerickson]

For pA performance,  relays probably need to be shielded (guarded) for both noise (coupling from the coil to the contacts) and leakage reduction. Shielded relays are expensive and hard to get. The K236 uses a bunch of these.
This is another reason that I decided to keep DIY-SMU's current noise floor at about 1nA.
Thanks,
Dave

[julian1]

I was considering the possibility of using two relays in parallel, for output switching and current range steering/switching in an application with high current range. (wetted) reed relays maintain good contacts down to very low currents, while larger relays can handle higher currents without the contacts sticking. But leakage currents will still be present across the large relay, so not sure it's a solution.

[Roehrenonkel]

Hi all, hi Dave,
 
i've been reading through the 40 pages of the AD7190 datasheet and learned a few things:
The internal clock has +-4% tolerance and probably some jitter.
An external clock of 4,9152 MHz (crystal or oszillator) can improve the acuracy
for 50 and 60Hz-rejection and provide a more precise/stable data-output.
At a data-output-rate of 10Hz an external clock gives 20dB (!) better
normal-mode-rejection for 50 and 60Hz.

[djerickson]

Hi all, hi Dave,
 
i've been reading through the 40 pages of the AD7190 datasheet and learned a few things:
The internal clock has +-4% tolerance and probably some jitter. An external clock of 4,9152 MHz (crystal or oscillator) can improve the accuracy
for 50 and 60Hz-rejection and provide a more precise/stable data-output. At a data-output-rate of 10Hz an external clock gives 20dB (!) better
normal-mode-rejection for 50 and 60Hz.

Thanks, Roehrenonkel, you are brave With internal clock the AD7190 50/60 Hz rejection is pretty good at 60-70dB. I did some quick tests a while back to operate the ADC at non-power line frequencies and the noise was still good.  Generally an SMU is not used in a heavy industrial environment or with long cables to low-voltage sensors, no? More of a quiet lab instrument.  Do I underestimate where SMUs are used?
Thanks
Dave