New Enclosure

I'm working on the new 2U Half-rack (8.5" wide) case. Thanks to an ad in Nuts and Volts magazine, I learned about GoBilda hardware: an Erector Set for adults. In addition to motion control hardware for robots, they make the 1106 off-the-shelf rails that are nice for building boxes. These are 8mm square rails with 4mm holes on 8mm centers. The ends are threaded for M4 screws. To build a box, the flat front and rear panels can screw to the rails, the bottom and sides can bolt on with screws and internal nuts. The top cover needs to mounted with some type of blind screw threads. Current thinking is to drill and tap some of the 4mm holes to accept M5 screws.



The top, bottom and sides are simple rectangular sheet metal with a few mounting holes. Holes on the bottom are for mounting the various components. These will be 16AWG (0.060", 1.5mm) 6061 aluminum.  I generally cut rectangular sheet aluminum on my table saw using a soft-metal carbide blade.

The front and rear are either hand-made sheet metal, PC boards, or custom aluminum panels such as PCBWay. I plan to use a PCBWay aluminum panel for the front. It solves the problems of front panel finish color (using soldermask) and labeling (silk Screen). Color choices are somewhat limited for PC boards and PCBway, but so be it.  As an EE I like to do front panel layout using PCB design tools.

I have built similar chassis using 1/4" square aluminum myself, but the rails are a pain to make: lots of machining, drilling, tapping. My thanks to GoBilda for doing the hard part. 280mm (11.0") rails are $6 each, and they come in different lengths. Parts are on order.

For the front panel jacks, I plan to use 8mm (5/16") holes for 5-way binding posts. These allow more connection choices than Safety Banana jacks, but are not as safe. The 6 holes can be enlarged with a step-drill if you prefer safety jacks or other connectors. For connections I plan to use the 2x3 arrangement that Agilent uses. It is similar to what Keithley uses on their rear panels, except I replace their Guard sense (not super useful) with a Chassis Ground.  

The heat-sink and mounting for the Amp board need to be raised up a bit to clear the 8mm rail. 0.5" board spacers will do the trick. The heat-sink will be 0.250" x 2.5" 6061 bar stock, cut to length, drilled and tapped for the TO-220 transistor mounting hardware. This will be screwed to the chassis-side for mechanical support and thermal contact.

Open issues: Finalize front panel controls, lay out the Front panel / CPU board, paint finish for top, sides, bottom, rear. And cooling air holes.

CPU and Front Panel Board

Now that I have decisions about the front panel size and shape, control positions, and jack configuration, it is time to design and lay out the CPU board. It will have the Teensy Processor and the front panel encoder and buttons, and power, Main Board, and fan connections. It fits between the Nextion Display and the Instrument banana jacks. This board takes in +12V and has a linear regulator for +5V. It has a simple fan control from the processor. Since the processor can calculate the power being dissipated in the heat sink, it can figure out when to turn the fan on. The fan driver will run the 12V fan at low voltage (5V) and low speed when not dissipating much power and will turn it on to +12V when over a few watts are being dissipated.

I chose thru-hole, 6mm switches since they are multiple sourced, and many button styles and heights are available. I prefer the look of rectangular buttons, but have no easy way to make small rectangular holes in sheet metal. So for home-made panels, will use round buttons and holes. I can use the nicer rectangular buttons when I get machine-made front panels.

The CPU board front side has the controls, back side has the Teensy 3.2 processor, voltage regulator, and the various connectors. 
   
Here is the CPU board inside the new case. It fits nicely in CAD.


CPU Boards and parts are on order. Designing the hardware was the easy part. The part that scares me is to port the firmware from Arduino Leonardo to Teensy. I use some chip-specific timed interrupt routines for reading and de-bouncing the switches and encoder. Have never moved this code to Teensy, so I need to pay the price for writing it as non-portable code. Ah well.

• Nextion Display
  
• SPI code. Could bit-bang initially to get it working.
  
• Controls: Buttons and encoder (needs 1ms timed interrupt)
  
• Fan control (calculate power, turn on Fan)

Here is the prototype enclosure with front, rear, one side and bottom. Gobilda.com makes the 8mm corner rails. The 6 sides are rectangular, flat 0.060" (1.5mm) sheet aluminum. I patiently await the new CPU board.

While waiting for the CPU boards, I spent a few hours porting the code to Teensy and getting it basically working. This required:

• Defining the I/O pin numbers for the new board
• Moving the 1mS interrupt from the AVR timer to a Teensy timer

Fortunately my 2019 Polysynth project already had a timed interrupt routine for Teensy, so it was easy to just copy the few lines of timer setup code. And it worked!  I had to fix one minor bug which the AVR compiler called a warning but the Teensy compiler called an error. It looked like 99.99% of the original code would just work. So instead of forking off a new build for Teensy, I just use #define AVRMEGA or #define TEENSY and where needed, #ifdef commands to make the code do both. So far so good.

When it compiled OK, I ran it on a bare Teensy 3.2, and it output the correct serial messages! I connected a Nextion display (just 4 pins) and Nextion worked. Porting to Teensy was much easier than I had expected. My thanks to Seitan for making his Nextion library portable.

All the major components are installed, except the new CPU board. So to get it working, the old front panel and CPU board are mounted at an awkward location.



2/8/21: The bare CPU boards arrived from PCBWay today. The 5V regulator, processor and Nextion display came right up, and the SPI pins had activity. I connected a 14 pin ribbon cable to the Main board and the encoder and buttons worked. It actually controlled the SMU, and it works, time for a beer! Tomorrow I drill holes in the front panel for the controls and mount the sucker. I found a few small bugs in my code that I was able to fix quickly. There is nothing like porting your code to a completely different architecture for finding bugs.

When you use a Teensy and provide it with local +5V power like I do, don't forget to cut the power jumper under the USB connector. Otherwise Your +5V will try to power the USB power and draw a lot of current.

Here is the new case with the new CPU, all cabled up and working.
 




Above is the hand-made prototype front panel. Next version will be machine-built and have proper labeling. Something like this (below). Haven't decided the colors, but am limited by the normal soldermask and silk-screen colors: White, Black Blue, Red, Yellow, Green. Silk screen is black or white. I plan to use larger rectangular buttons so the buttons cutouts will be changed. And I'll change the jack labeling from +/- to HI/LO. And add a cool logo. Haven't yet decided what to do with the bottom 2 buttons. Maybe Remote sense and Mode.

Or...

How about blue with the cool square buttons.

New Stuff: 1.5V range and Remote Sense

Problem: Voltage clamp in the presence of an external voltage

Problem: Step-response Overshoot!

Another engineer, Jaromir, also built an SMU, based on the K236 with inputs from my design. He noticed that with voltage steps, there was about 10% overshoot, followed by a short settling time. I Let him know that my unit exhibited a nice well-damped step response: exponential, with no overshoot. Here is a 200V step response. No overshoot, but perhaps a bit slow, about 400uS to settle.


I had 2 systems, and had not tested the step response on Board #2. When I tested it, I saw similar response to what Jaromir reported. Hmmm. I compared the two boards carefully, even removing some components to measure them accurately. There were a handful of differences between the two boards. Board #1 had been used to debug many initial problems, and some of these changes hadn't been put back.

• Force DAC filter capacitors. No difference.
  
• Integrator P term resistor R27: #1 had 1K, #2 had 2K. Changed, no difference.
  
• Checked most of the compensation caps: no difference.
  
• Checked value of Y capacitor C5, #1 had none. No difference.
  
• There was no C34 on #1, no difference.
  
• #1 had no output relay installed. I installed it and no difference.
  
• I installed #2 in the enclosure, so the power CPU and mostly, the Amplifier Board were different. It still had overshoot. 

LVGL?

SCPI Control

For SCPI I found 3 libraries: 2 are Arduino Libraries: https://github.com/LachlanGunn/oic and https://github.com/Vrekrer/Vrekrer_scpi_parser. Vini from the OSMU project has been helping me.  He pointed me to the full-featured  SCPI library:  https://www.jaybee.cz/scpi-parser. One of OIC's examples implements a simple Source-Measure unit using Arduino ADC and PWM pins. This is a perfect example. I tested the parser code by sending various number formats for the voltage settings: Integer, fixed point, and scientific notation. They all work well, which is wonderful. Now I'm changing the example to implement ranging, current measure, etc.  Nice job, LachlanGunn! I'll try out the Jaybee parser as well.

Vini is helping to get SCPI working on the SMU. But meanwhile, I want to get experience with SCPI, and also want to implement it on older and future projects. I began with the OIC library. I got it working on a Teensy, and was happy with how quickly it came up. My 18b DAC project was crying for a proper way to control it via USB. And it is a simple enough project: *IDN?, *RST, set and get the DAC to a voltage, done!.  See the 18b DAC project for details. I tried to install all 3 libraries. OIC and Vrekrer worked well. Jaybee is much larger, with 8 .c files and 10 .h files. I could not get it to compile on Teensy. I settled on the Vrekrer library. It is simple to use and works well.

My next step was to use the 18b DAC as a test platform for additional SCPI commands. This worked out well without the complexity of all the SMU code.

Adding this library to SMU on Teensy went fairly smoothly. I found that I needed to control the SMU data structure variables, set the hardware, and update the display. This forced me to refactor several additional control functions such as setOnOff(). It's all good. Here's the first pass SMU SCPI command list, containing the major settings.

Measurements and error plotting: ADC errors

As rewarding as it is to type in SCPI commands and see them execute,  it is far more exciting to write a program to control instruments via SCPI. For a control program, I am currently using Python Spyder and pyvisa. For the first time I can automatically sequence through a voltage or current range and compare the output of the SMU to my 6.5 digit DMM.

This first Python plot compares SMU Measure to the DMM over a -15V to +15V voltage range set by the SMU and shows the error. Note the mirror-image bumps around -9V and +9V. The rise at -15V is another error. Is it the voltage measure circuit, the ADC, maybe my DMM? I suspected the ADC input op-amp U4, a TLC2272A. The large rise on the left is the op-amp output near VCC, something else to look out for.

I changed U4 to an MCP6072, another mid-precise 5V RRIO part I have used in the past. The ~6mV offset error is because it is not re-calibrated. Completely different error shape! Similar dip at -9V though.

This argues that the gross error shape is the non-linearity of amplifier U4, and not the ADC or the DMM. The results are very interesting. I purposely use an inverting amplifier for U4 in a current-summing configuration to minimize any non-linearity caused by input common-mode variation, a problem with many RRIO amplifiers. The amplifier + input is GND, and the output is biased to mid-scale by a very precise reference.  This amplifier is operating at a gain of 0.55: +/- 5.5V input range, 0 to 5V output. It looks as though the op-amp input error varies as a function of the output voltage. Something that no op-amp specifies! Over a 30V range (+/- 15V) a 5mV error is +/- 0.008%: significant but not terrible. Here is the schematic for the ADC input buffer, U4. U4 is powered from +5V and GND. I will try some other op-amps.


By the way, here is the Force error, effectively the DAC and system INL, 500 samples, over a -15V to +15V range This is about 1mV p-p or +/- 0.5mV. The force DAC is bipolar 16b, the ADC is about 23b. So the 5mV ADC error is concerning. The ADC error should be should be < 0.1x of the DAC, not 5x worse than the DAC. I plan to use the Measure ADC to calibrate the Force DAC. Not until I fix the ADC errors.

Here is the new OPA3240. This is a 5V RRIO part with better input and output specs near the rails. I re-calibrated the 15V Measure range to correct for offset and gain (slope).  This is better, note the scale: about 1.2mV p-p. Still the mysterious bumps at +/- 9V. Roughly 25% and 75% of the ADC input range of 0 to 5.0V. Hmmmm.

I tried turning the AD7190 ADC input buffer ON. It has been OFF until now. With the buffer ON, the ADC loses about 0.25V of input range near each rail. No difference was measured since my circuit does not go that close to the rails. So it's not a ADC buffer-related issue. Next I'll try some chopper amps, OPA2333.  OPA2333s arrived, I soldered one in. No change. The two peaks are still there, and about the same amplitude. That's the 4th opamp. It's not the opamp.

So it must be the ADC. I read the entire AD7190 data sheet and found a few small issues, plus a hint. The INL spec is 5 ppm Max, 1ppm Typ. And the INL graph shows a suspicious  similarity to my error plot, but at 1ppm, not 30ppm. The other 2 small issues:

• They recommend input RC filters for common and normal mode to remove noise that is near the internal sampling rate. I doubt this is it. Also this is a note only, no schematic!
• Also improved bypassing: short leads and separate Dvdd and Avdd.
  

I added a 0.1uF cap from power pins 19-18 and 20-21. Slight change in gain, but the +/- 0.5mV errors persist.

Added a 0.1 cap to the Ref+ pin. The data sheet warns that the reference input is dynamic, meaning it draws AC current. This shifted the peaks from +/- 9V to about +/- 12V. Same amplitude though. This could be a clue!

• Chop mode. This reduces small offset values and offset drift bus slows conversion time 3-4x. No difference
  
• Filter select: This trades off settling time vs conversion time. No difference
  
• Buffer: Worse with buffer off, depends on op-amp driving.
  
• Convert time: Tried longer and shorter settings, no difference.
  

I looked at the designs of two eval boards for the similar AD7193. There is no AD7190 eval board, but AD7193 is the same ADC core and specs and a 8:1 mux. Nothing in the layout jumped out at me. However both boards have a 10uF tantalum on the reference. I have a 0.1 uF only, per the recommendation of the ADR431. When I paralleled a 10uF ceramic cap, the evil error curve that I have been waking up in the night thinking about for 2 weeks WENT AWAY! The AD7190 does not call for specific bypassing on the reference input. Finally, noise and error in the +/- 100uV p-p range out of 30V p-p. that's 4ppm. Love it.

There is (as always) a side effect to this change. The ADR431 reference doesn't like large capacitive loads, which cause high-frequency peaking (see data sheet) unless additional compensation components are added. I will switch to the ADR421 which doesn't have this issue and also is lower noise: 1.7uV vs 3uV.

I implemented the ADC and reference reworks to unit #2. It still has an ADC non-linearity that is slightly different than than unit 1. There is also a gain error that will reduce after calibration. But there is still a symmetrical error curve of about +/- 0.4mV on the 15V range. Since this is holding up the calibration process (below), I will defer the final linearity correction to the next PCB revision, and move on. Calibrating over the range of +/- 11.0V eliminates most of the non-linearity errors. Also the 34401A DMM voltage ranges are only up to +/- 12V. Scandal! 34401A really isn't 6.5 digits, which generally means 19.999V, it's more like 6.2 digits (1,2000,000 counts). BTW, the HP 3478A sold as a 5.5 digit meter is actually 300,000 counts so for measuring 12V to 30V, it has the same resolution as the 34401A.  By using +/- 11V, the 34401A  does not auto-range, which saves both measurement time and relay operations.

SCPI Instrument Program

For a SCPI instrument connection program, I use Keysight's Connection Expert. It requires you to register and install 1-2 other programs. Here it is controlling DIY-SMU, my 18b DAC and two HP GPIB instruments.

Very Low Source Currents

To measure the lowest 1uA full-scale force current range, I have been using a 1Meg precision resistor, or the 1M/10Meg input of my scope and DMM. I have access to a Keithley 610C Electrometer which measures extremely low currents. The 610C lowest current range is 1e-11 x 0.001 or 10fA full scale! Not sure what I can use to test this. It certainly needs a shielded and guarded enclosure. Measuring SMU, it was rewarding to see this meter stepping through 1LSB changes of the Force DAC (~30pA) on the 1uA range.
  
  

New Front panel and Case

Here is the new front panel mock-up. The panels were ordered from PCBWay, 5 for $50. I moved the mounting rails in 1mm to better accommodate the cover and bottom sheet metal. A one-piece, 3-sided cover is in process as is a PCBWay rear panel. Each factory panel saves me hours of hand sheet metal work.  Let me know what you think of the color. The cover will probably be light grey or greige.



Here is the new rear panel CAD model. Parts are on order.



The new rear panels arrived from PCBWay. They look and work great, just like the CAD model. What a joy it is to just screw in the components: no measuring, marking, sawing, drilling, hole-sawing, filing, de-burring, sanding, priming  and painting. No blank panels or crummy stick-on labels.  Next task is to convert the lower unit to the new chassis and panels.

Chassis Wiring

It occurred to me that I have all the chassis interconnect details in my head, and that anyone building a DIY-SMU would benefit form a wiring schematic, showing the chassis components, AC and DC wiring, and cables.Here is the schematic.

Calibration Update

I have been working on a new calibration method. Now that SCPI is working, it makes sense to calibrate using an external program and external instruments. I mentioned before the method I have been using to calibrate manually. I set the force to 0V, and get a Force offset cal correction, entering the value into the code and re-compiling, then repeating for Force gain. Then repeating the process for both Force and Measure on 3 voltage and 6 current ranges. Not pretty or fun, not to mention error-prone. While testing the Force and Measure linearity with Python, I found that taking a series of measurements and then doing a least-squares linear fit on the error data works very well. I use the Python np.polyfit() function to provide offset and gain cal factors simultaneously.  The Python program generates cal factors for each voltage range. I needed a way to pass the cal factors to the SMU via SCPI. I chose CALibrate:VOLTage index, value to send one single-precision, floating point cal factor at a time. The index will use 0,1,2,3 for FVGain, FVOffs, MVGain, MVOffs for the 1.5V range, then proceed through the other voltage and then the current ranges. Values will be sent to RAM first, then copied to EEPROM.  Another SCPI command disables or enables calibration: CAL:STATe ON/OFF. I also need a way to identify a blank EEPROM and not load cal from it.

Next is Calibration of the current ranges, and commands to copy the data to EEPROM. For current cal, the cal program prompts the operator to manually change the load resistor for each range. I built a handful of precision resistors on dual banana plugs for this.  I also found and fixed a few bugs with Force current setting over SCPI.

Temperature Drift Measurement

I tested one instrument for temperature drift of Voltage Force and Measure over a 10°C range, from 28°C to 38°C. The overall Force and Measure drift is primarily caused by about 10 resistors, each contributing between 1.0 and 0.5 of their drift to instrument gain drift. These are mostly 0805 0.1% 25ppm/°C resistors, except for the four 0.01% 5ppm ones used in the voltage-difference circuit.  The Reference, DAC and ADC are all in the 1-3 ppm/°C range. Here is a Voltage Gain Drift analysis spreadsheet showing the temp drift contribution of each component in each of the Voltage Force and Measure signal paths. RMS assumes Gaussian distribution for the drifts, and uses the square root of the sum of the squares. I calculate about +/- 60ppm and measured +20ppm/°C for both Force and Measure drifts.

This spreadsheet is useful to show the largest contributors to drift. To improve drift, 0805, 10K, 10ppm resistors are available for about $.50. Other value 10ppm resistors are about $1.00. 5ppm 10K's are about $1.30 and other 5ppm values are about $2. The crossover uses two  connected 10K's which could be a SOT23 network to reduce its drift to ~1ppm.

Here is my DIY bench-top temperature setup. The top box is a small oven I built to test boards and components. The chamber bottom is open to allow it to be placed over an experiment. The bottom cardboard box is an adapter to allow the too-large DIY-SMU to fit.

10/20/21 Updates:

Calibration Code

I completed the calibration Python code and am happy with it. The voltage calibration calibrates all 3 voltage ranges for force and measure in one step. The current calibration requires changing the load resistors for each current range. I created a simple menu to select the current range and prompt the user to install the correct resistor. 

• 1uA:   1 Meg-ohm, 1V
• 10uA: 1 Meg-ohm, 10V
• 100uA: 100K, 10V
• 1mA:   10K, 10V
• 10mA: 1K, 10V
• 100mA: 10 ohms, 1V