DIY-SMU Project

[Kleinstein]

A SMU naturally needs a little more power and also needs to get rid of the heat, especially in sink mode. So like a linear lab supply it can't be build very small and just needs space for heat sinks and so on. So already the 1/2 rack width like the modern Keithley 24xx series is relatively small.

It is more that other instruments could get smaller: some of the simpler bench DMMs and DDS generators are quite some empty space inside and could get even smaller. The 1/2 size got a kind of standard and this helps when stacking instruments - nobody want's a mix of 1/3 , 1/3 and 2/5. It is already bad enough they often have different depth and the feet at the back instead of at a defined depth.

For precision instruments, one wants to avoid any unnecessary EMI - so bluetooth is kind of a no go, and also USB is not really the preferred interface as it is often more like an EMI source than a power source. In this respect I am still expecting some kind of lower end fiber based interface for lab instruments (not 10GBit LAN).  For a SMU the power over USB is too limited - it may be just OK for a bench DMM or maybe small DDS generator.

[wizard69]

A SMU naturally needs a little more power and also needs to get rid of the heat, especially in sink mode. So like a linear lab supply it can't be build very small and just needs space for heat sinks and so on. So already the 1/2 rack width like the modern Keithley 24xx series is relatively small.

It sounds like you where responding to my post.   In any event to clarify:

• The desire for a USB device is for something that can generate and read accurate low voltages.   The goal is to calibrate instrumentation in the common process voltage ranges.   Think field work here where a dongle would be ideal, size matters and big is not good.  We could easily skip current control or high current on the voltage reference.
• I realize that 1/2 rack is currently a standard but it also wastes a lot of space for many instruments.   I'm a bit of a fan of the old TM500 and 5000 series from Tektronix.   It might not make sense to bring back that concept but narrower instruments would actually work well on many benches.  We just need to rethink how to form the instruments.

In any event I don't see a need to make every meter or bench instrument 1/2 wide.   There are many power supplies on the market setting good examples as far as layout goes.

It is more that other instruments could get smaller: some of the simpler bench DMMs and DDS generators are quite some empty space inside and could get even smaller. The 1/2 size got a kind of standard and this helps when stacking instruments - nobody want's a mix of 1/3 , 1/3 and 2/5. It is already bad enough they often have different depth and the feet at the back instead of at a defined depth.

My experience is that unless you buy the instruments from the same vendor (not a good idea) stack ability is highly variable, as you note.   So I really don't see that as a big deal.   If we go narrow and higher it becomes easier to justify a shelf for each instrument row.

For precision instruments, one wants to avoid any unnecessary EMI - so bluetooth is kind of a no go, and also USB is not really the preferred interface as it is often more like an EMI source than a power source.

This may be true but on the other hand instrument makers have been over coming the USB issue for sometime.   There are few instruments these days without USB so the EMI problem can be over come.

In this respect I am still expecting some kind of lower end fiber based interface for lab instruments (not 10GBit LAN). 

Every time something new comes out with a fiber interface it seems to die rather quickly.   Thunderbolt was suppose to support fiber and for whatever reason it never became a thing.

For a SMU the power over USB is too limited - it may be just OK for a bench DMM or maybe small DDS generator.

For the USB based device I'm really thinking a more limited SMU type device.   Basically an easily programmed precise voltage source with the ability to measure low voltages.   The ability to source or sink current, less than 100mAs, would be nice but that does lead to an explosion in size.

In the end I have two different instruments in mind the more I think about it.   The other thing is that I'm leaning towards minimal user interfaces on the device itself and expecting the user to connect to a PC for complete control.

[djerickson]

Hi All,

Announcing recent upgrades to the DIY-SMU project:
Project web page: http://www.djerickson.com/diy_smu  The latest stuff is on the bottom of http://www.djerickson.com/diy_smu/smu-bringup.html
Video about the new changes and chassis: 
Study of DC-DC and AC-DC common-mode noise at http://www.djerickson.com/dc-dc/
Gobilda Hardware: https://www.gobilda.com/ Don't miss their https://www.gobilda.com/product-insights/

Changes are:
1) New half-rack 2U enclosure. I came across Gobilda hardware and used 4 of their 8mm square rails to build a frame. The prototype front panel is larger than the original box, so I was able to design...
2) New Teensy 3.2 CPU and front panel board. No more hacked Arduino Leonardo board and front panel controls.
3) Ported the Leonardo code to Teensy. Pretty easy job. Headroom is good......
3) Solved the 300KHz DC-DC noise issue by changing from Meanwell to Recom DC-DC. If you're interested, See my DC-DC and AC-DC common-mode noise study.

Thanks to all for your excellent feedback.

Dave

[nikifena]

Dear Dave.
I think you have to put a resistor between the U8.2 opamp output and the capacitor in order to prevent oscillations as shown

Also, what about adding a single 100nF next to each opamp between the 15V rails instead of two from positive to ground and the negative to ground. If you want to isolate the supply, you can add two 10ohm resistors in series of the 15V and -15V and each opamp.

[wizard69]

Looks like you are on the fast track here.  By the way the boards appear to be smaller than expected.

As for Teensy is there a reason for going that way over simply soldering in a raw chip?

Hi All,

Announcing recent upgrades to the DIY-SMU project:
Project web page:
...
Study of DC-DC and AC-DC common-mode noise at ...

Changes are:
1) New half-rack 2U enclosure. I came across Gobilda hardware and used 4 of their 8mm square rails to build a frame. The prototype front panel is larger than the original box, so I was able to design...
2) New Teensy 3.2 CPU and front panel board. No more hacked Arduino Leonardo board and front panel controls.
3) Ported the Leonardo code to Teensy. Pretty easy job. Headroom is good......
3) Solved the 300KHz DC-DC noise issue by changing from Meanwell to Recom DC-DC. If you're interested, See my DC-DC and AC-DC common-mode noise study.

Thanks to all for your excellent feedback.

Dave

[nikifena]

^
I think there is a lot of free space on the mainboard. Also, it's good to put the ribbon cable on the opposite side - next to the controller.

[Kleinstein]

Dear Dave.
I think you have to put a resistor between the U8.2 opamp output and the capacitor in order to prevent oscillations as shown

The 10 Ohms resistor as shown does more harm than good. The resistor only makes sense in combination with an additional capacitor as direct feedback.

The OPA2140 is a nice OP, but not really the right choice as a reference buffer / inverter. In this circuit a BJT based OP, e.g. like OPA2202 (new and relatively cheap) or AD8676 should give better performance - one still needs to check the common mode voltage.

[djerickson]

Thanks again for the comments and help.

From Wizard69:
"As for Teensy is there a reason for going that way over simply soldering in a raw chip?"

Sure. My motivations are: Design expediency, ease of hand-assembly, Arduino tool and library support, easy Upgrade to the (amazing!) Teensy 4, previous experience, and love for all things Teensy. $20 is one of the high-ticket items on the BOM. Raspberry Pi Pico? Hmmmm.

From Kleinstein via Nikifena:
Dear Dave.
I think you have to put a resistor between the U8.2 opamp output and the capacitor in order to prevent oscillations as shown.

I agree with Kleinstein. A 10ohm resistor will make the circuit less stable, without a feedback cap from an opamp OUT to - IN. This circuit is a bit of a hack, I freely admit. Turns out putting a larger cap with higher ESR (cheap electrolytic or Tantalum) works well on the output of most op-amps, even with an emitter-follower buffer. You get a free ~1-2 Ohm resistor with every cheap cap :-)  Try simulating this circuit stability vs ESR and you'll see that some ESR is your friend, and low ESR causes instability. "But", you say, "There is no guarantee of minimum ESR, Dave". True, but 1) Without some ESR, most 3 terminal regulators aren't stable and 2)  If cap manufacturers could build the same cap with very low ESR, they would sell them for more $$ as low ESR caps, no?
Admittedly, if I were building hundreds of these (by machine) I might add a 1ohm R on series with the cap or the 10 ohm R + feedback C.

I use OPA2140, a fast JFET for a DC reference buffer since I already have a handful of them on the board. It has good enough drift and noise.

From Nikifena:
"Also, it's good to put the ribbon cable on the opposite side, -  next to the controller."
True. The signal flow is SPI->Isolator->DACs->Crossover->Amplifier->Irange->Vmeasure+Imeasure->Output. To put the output near the front, the SPI ended up in back. A long ribbon cable is a minor compromise, particularly since it is isolated.

BTW how do I include quotes from others in this blog thingie? How about in-line images? I'm new at this.

Thanks,
Dave
 

[Kleinstein]

One can get a quote from an older answer, just click on the "inser quote part" to the top right of the answer. If needed delete the part not needed.

 Inline pictures get be inserted similar to other attachments, just chose the right type.

They start to appreciate ESR: there are extra MLCC caps with added well defined ESR available.

[jbb]

Thanks for the website and video updates. I know it’s a lot of work and I appreciate it.

Here’s a question about the output amp: do the cascade MOSFETs do much to the amplifier dynamic performance? I was thinking that if I built such an instrument I might not need (or indeed want due to safety concerns) a 150V range. Would it be practical to use a lower DC rail (eg +-30V) and only use the TIP41/TIP42 BJTs?

[Kleinstein]

The cascaded MOSFETs should only be reponsible for the high voltage part. When used at low voltage they should not contribute.
At lower voltage (like +-40V) a pure BJT based design should work too.  I would still consider faster transistors than TIP41/42, more liky D44H / D45H or BJTs made for audio. These may have a reasonable SOA and speed.
One could still use Juts 1 MOSFET each for a lower voltage version, so essentially without the cascade.

[djerickson]

To see how Keithley does both +/- 200V 100ma and +/- 20V 1A, check out the 2400 service manual.  It has a simplified schematic for the amplifier design. It has one bipolar cascoded with 2 FETs for the low voltage high current, then another bipolar plus 2 more FETs for the 200V. I suspect the 2 bottom FETs are there to share power, otherwise the single bipolar needs to dissipate all the power. When used as a 1A load at 20V: About 20V + 20V + 15V headroom x 1A ~= 55Watts.

With 200V heat-sink isolation, you need a good electrical insulating thermal pad on your transistor, which adds ~1C/W thermal resistance. At 55W, that's 55C rise just for the stupid insulator, never mind the package and heat sink.

I have put dangerous voltage and current on a 5 lb heat sink with 4 FETs and 4 power resistors just to avoid those damn 1C/W thermal pads.

Dave
 

[vmiguel]

Hello Dave.

I have been thinking about your DC-DC converter problem I came with the following transformer design for the transformer:

It's a PCB coil design using a tiny U ferrite core with good isolation between the windings.

The top and bottom layers are used as individual shields in both coils and the board cutout and distance provides little coupling capacitance.

I did not make any calculations in the windings, considering 5V input -+15v output as an example.

This way would be very easy for anyone to assemble the converter.

Regards,
Vini

[arlo_g]

Re heat sink insulation:  I haven't seen them offered at the big distributors yet, but Aluminum Nitride insulators for TO-220 and TO-247 devices are available fairly inexpensively from Aliexpress (with some risk that not all vendors sell quality product). 

Ceramic AlN has very good thermal conductivity and dielectric strength: somewhere from 100 W/mK to 200 W/mK or almost as good as BeO with none of the toxicity.  A 1/16" = 1.6mm thick TO-220 insulator should have at least 6x better thermal resistance than 1C/W of filled silicone thermal pad.  Of course, the AlN insulator would need messy thermal grease at both interfaces. 

[djerickson]

Re heat sink insulation:  I haven't seen them offered at the big distributors yet, but Aluminum Nitride insulators for TO-220 and TO-247 devices are available fairly inexpensively from Aliexpress (with some risk that not all vendors sell quality product). 
Ceramic AlN has very good thermal conductivity and dielectric strength: somewhere from 100 W/mK to 200 W/mK or almost as good as BeO with none of the toxicity.  A 1/16" = 1.6mm thick TO-220 insulator should have at least 6x better thermal resistance than 1C/W of filled silicone thermal pad.  Of course, the AlN insulator would need messy thermal grease at both interfaces. 

Oops, my mistake, I confused thermal conductance and resistance, Doh! For a TO-220 insulating pad, thermal resistance is about .3-.5 C/W, not 1.0 as I stated. Not a problem when burning 10-20 W,  but something to watch at high power.   Some non-insulating ones are lower but I need insulation to a few hundred volts.
I see TO-220 and TO-247 Al/N pads on Ali Express. Specs are nice, I will buy some.  Having those niceties: you know, like a manufacturer's web site and a data sheet would be lovely.  Thanks for the lead.

Dave

[jbb]

I generally assume 1C/W for a TO-220 insulator too. If I ever do a high volume design I’ll clearly have to do some careful selections and qualification.

The flip side of isolation is capacitance, and that’s where ceramics could be really nice; thicker part = less stray capacitance. I guess a little grease will be needed between the device and the pad and the heat sink. (One good argument in favour of a Sil-Pad or similar.)

I do wonder what shapes are really available. For example, I would prefer to not have a hole in the insulator (which has a very short creepage path) and use a spring clip. Maybe one could source modestly sized rectangles and mount a couple of TO-220s.

The joyous endpoint of this kind of thing is the integrated power module, where one side of the ceramic sheet has circuit traces (ie it’s a single-layer PCB) and power semiconductors (soldered/sintered on, some connections made by wire bond) and the far side bolts straight to the heat sink. Prototyping costs might be high :-)

[djerickson]

Here's the AliExpress page I ordered from. Different sizes, 0.6 or 1mm thickness, hole or no-hole. They specify 350 W/mK, but they are thicker than plastic and other materials. I bought the thicker ones hoping they are more rugged.
https://www.aliexpress.com/item/1005001350971494.html

Dave

[jbb]

Hi Dave

Saw on your website that FIMV mode could cause trouble with voltage clamps.

I think I saw something relevant in a Keithley SMU manual, but didn’t understand it at the time.  It was a couple of MOSFETs that switched the resistor divider ratio of the inner current clamp BJTs of the power amplifier.  You could have a ~120mA amplifier limit for the force 100mA range, and a lower limit for the force 10mA range and below.

Another option would be looking at clamping the maximum voltage drop across the current shunt assembly. If you set up +-7V clamping that would limit the dissipation of the 499R shunt resistor to around 100mW, which might be acceptable.  As usual, such a clamp would need nice low leakage when not conducting and be able to shunt the worst case amplifier output current when required.

[djerickson]

Hi Dave
Saw on your website that FIMV mode could cause trouble with voltage clamps.
I think I saw something relevant in a Keithley SMU manual, but didn’t understand it at the time.  It was a couple of MOSFETs that switched the resistor divider ratio of the inner current clamp BJTs of the power amplifier.  You could have a ~120mA amplifier limit for the force 100mA range, and a lower limit for the force 10mA range and below.
Another option would be looking at clamping the maximum voltage drop across the current shunt assembly. If you set up +-7V clamping that would limit the dissipation of the 499R shunt resistor to around 100mW, which might be acceptable.  As usual, such a clamp would need nice low leakage when not conducting and be able to shunt the worst case amplifier output current when required.

Thanks much for your inputs, jbb. 7V would be safe, 15V if I beef up the 499R, I'm a little reluctant to add more diodes and leakage, so am leaning towards the 236 solution: lowering the current limit for all but the 100mA range.  Without lowering the current limit, you would get >100mA fault condition on the 10mA range.  Could fry a sensitive DUT.
I considered 2 other solutions. 1) Detect and fix it in software by increasing the clamp V. Is that a hack, or what?  2)  Hardware to detect this fault and drive the integrator to override the clamp. That could be a stability headache.
Dave

[jbb]

Yeah, solutions 1 and 2 above sound ... challenging. Adding more control loops to the system will get more and more annoying, and I expect you’re already having some battles choosing acceptable control gains for all the voltage and current range options.

Well what a surprise :-), Keithley have a really good solution in hand. It’s not perfect, though: the main control integrator will wind up when the CV amp (whether high or low) is controlling the loop and the output amp is limiting the current.

[Kleinstein]

Well what a surprise :-), Keithley have a really good solution in hand. It’s not perfect, though: the main control integrator will wind up when the CV amp (whether high or low) is controlling the loop and the output amp is limiting the current.

The current limit from the output amp should only be a short time or last resort limit. Normally the fine adjusted current limits (upper and lower limit separate) should set the limit at the integrator input.

[jbb]

The current limit from the output amp should only be a short time or last resort limit.

Hmm. Perhaps one could detect the current over range with some comparators and trip the output off? This is, after all, a fault condition.

If pursuing this approach, I would seriously consider implementing the trip logic in the analog control  board rather than trying to catch it with the microcontroller.

Normally the fine adjusted current limits (upper and lower limit separate) should set the limit at the integrator input.

Yes, but I think Dave’s problem is with the FIMV mode, where the current is uncontrolled once you hit the voltage clamps.

Maybe one could add a set of ‘outer current clamps’ to the crossover circuit which can override the voltage clamps? These could be hardwired to, say, 130% of range setting. That would scale the clamping with the range and avoid integrator windup.

[djerickson]

Thanks for the current limit feedback. For now I'm going with the Keithley solution: MOSFETs to switch the current limit for the 100mA range: 15mA on the 10mA and lower ranges, 150mA on the 100mA range. Yes, the integrator will wind-up, but this is a serious (but possibly common) fault condition.  I did a simulation of the current limit circuit, and it seems to work well.

I plan to switch the display from Nextion to LVGL. I found what looks like a nice Dev board: Teensy 4.0 + 3.5 TFT with Cap. Touch + LVGL examples.
http://skpang.co.uk/catalog/teensy-40-classic-can-can-fd-board-with-480x320-35-touch-lcd-p-1611.html

Anyone know of other Teensy + LVGL examples out there?

Thanks!
Dave

[jbb]

Doing the emergency current limit in the power amp seems sensible. You want that kind of thing to be as simple as possible. Can you configure it so that it defaults to 15mA limit if the relevant control wire falls off?

You mentioned how to source this switch signal. I would favour using an inverter on the active low 100mA range; that way some confusion in the software can’t leave you in the dangerous configuration.

On integrator windup: I suspect the best approach is to trip/disconnect the SMU in this case; after all, something has gone quite wrong. I assume that a 15mA limit (plus voltage clamping to floating +-15V rails) is enough to protect the delicate parts in the current sense array?  If so, I guess you could detect current over range from the sense ADC. Maybe the trip could be “set voltage clamps to +-160V, set current ref to 0”?

In fact, that makes me think about the general challenge of enabling the SMU when wired to an energised device under test. Your PS-Load project had a nice take on this (a switch to disconnect the output MOSFETs from the preamp and a servo loop to preset the integrator). Is there anything like that in your current system?

Sorry I don’t have much on LCDs.

[riyadh144]

My god, that is some great work, I am commenting here to stay uptodate I guess I know what is my next project.

You need to sell that man.