I am currently building a cheap and dirty gigaohmeter.
The principle is pretty simple and basic: a precision CCS feeding the R.U.T. followed by a unity gain buffer.
No rocket science involved, but for 100G full scale with a 10V output, a current of 0.1nA is required as a stimulus.
Everything works as expected for lower ranges, but for 100G, the CCS fails.
Obviously, there are numbers of potential issues regarding leakage, etc., and I did my homework accordingly: I carefully selected an tested all the components involved: the BSP92 was chosen for its sub pA leakages between all electrodes, the TLC272 sample was selected for the same reason, etc.
I used guarding for the output, but it doesn't seem to be an issue: the main problem is the CCS. It should deliver 0.1nA, but it delivers nothing.
I measured the current through Rx and R2, they were identical and zero. When R2 is reduced to 250Meg, a current of 0.2nA flows through it and the Rx, and the indication is consistent.
With the total 1G of reference, no current flows, and the output of the TLC272 goes high, trying to reduce it even further.
At 1nA, its output is ~0.7V below the +12V rail. The 3V difference between the 15 and 12V rails allows the opamp to work comfortably.
There must be a leakage unaccounted for somewhere, but I am unable to locate it.
I have measured the actual gate current of the BSP92, an it's below 1pA, as is the -input current of the TLC272.
The reed relay has been screened too, and anyway in the 100G condition, all its electrodes are +referenced, meaning it could only increase the current, not diminish it.
The problem is clearly the CCS, but I don't see where it stems from.
Could be the board: it is an epoxy perfboard, but it doesn't seem to have issues: nothing measurable, and I milled it to isolate the critical nodes, and I isolated the -input of the opamp to connect it directly to the source of the BSP.
The principle is pretty simple and basic: a precision CCS feeding the R.U.T. followed by a unity gain buffer.
No rocket science involved, but for 100G full scale with a 10V output, a current of 0.1nA is required as a stimulus.
Everything works as expected for lower ranges, but for 100G, the CCS fails.
Obviously, there are numbers of potential issues regarding leakage, etc., and I did my homework accordingly: I carefully selected an tested all the components involved: the BSP92 was chosen for its sub pA leakages between all electrodes, the TLC272 sample was selected for the same reason, etc.
I used guarding for the output, but it doesn't seem to be an issue: the main problem is the CCS. It should deliver 0.1nA, but it delivers nothing.
I measured the current through Rx and R2, they were identical and zero. When R2 is reduced to 250Meg, a current of 0.2nA flows through it and the Rx, and the indication is consistent.
With the total 1G of reference, no current flows, and the output of the TLC272 goes high, trying to reduce it even further.
At 1nA, its output is ~0.7V below the +12V rail. The 3V difference between the 15 and 12V rails allows the opamp to work comfortably.
There must be a leakage unaccounted for somewhere, but I am unable to locate it.
I have measured the actual gate current of the BSP92, an it's below 1pA, as is the -input current of the TLC272.
The reed relay has been screened too, and anyway in the 100G condition, all its electrodes are +referenced, meaning it could only increase the current, not diminish it.
The problem is clearly the CCS, but I don't see where it stems from.
Could be the board: it is an epoxy perfboard, but it doesn't seem to have issues: nothing measurable, and I milled it to isolate the critical nodes, and I isolated the -input of the opamp to connect it directly to the source of the BSP.
Good trick with the separate supply for U1, so it can pull M1's gate as high as needed.
I have no idea what's wrong. Of course PCB leakage from the negative input to the output of U1 could cause it. Moist PCBs with flux residues can leak a lot, as can hot and wet PCBs, but I've built circuits working with 10 pA currents at 1.3 V supply voltages on perfboard and those worked fine - at room temperature with a dry board.
Apparently you lose some 85 pA...100 pA somewhere, as 100 mV/350 Mohm = 0.285714285714... nA instead of 0.2 nA.
In my experience, the TLC272 can very easily be damaged by static electricity. It then still works, but with much input current. Could that have anything to do with it or did you measure the U1's input current in circuit after determining that the current source doesn't work? The leakage can also depend on the common-mode voltage, depending on what type of ESD protection was used.
It could simplify things if you used the resistor under test as the feedback resistor of a transimpedance amplifier, although you then may need to put a low-leakage capacitor of a few pF in parallel for stability. You also can't guard it anymore then.
I have no idea what's wrong. Of course PCB leakage from the negative input to the output of U1 could cause it. Moist PCBs with flux residues can leak a lot, as can hot and wet PCBs, but I've built circuits working with 10 pA currents at 1.3 V supply voltages on perfboard and those worked fine - at room temperature with a dry board.
Apparently you lose some 85 pA...100 pA somewhere, as 100 mV/350 Mohm = 0.285714285714... nA instead of 0.2 nA.
In my experience, the TLC272 can very easily be damaged by static electricity. It then still works, but with much input current. Could that have anything to do with it or did you measure the U1's input current in circuit after determining that the current source doesn't work? The leakage can also depend on the common-mode voltage, depending on what type of ESD protection was used.
It could simplify things if you used the resistor under test as the feedback resistor of a transimpedance amplifier, although you then may need to put a low-leakage capacitor of a few pF in parallel for stability. You also can't guard it anymore then.
You are using force-current measure-voltage topology abbreviated FIMV. This topology is used for low resistances in precision test equipment. For high resistances, precision test equipment uses FVMI.
You are buffering the voltage across the 100G resistor with an op amp that has an input impedance of 1T ohm. You would want an op amp with much higher input impedance for this kind of buffer.
Make a simple picoammeter for use on the low side of Rx and have a well known voltage across Rx. There are open source schematics for a simple picoammeter on the web.
Forcing a very tiny current into a very large resistance is subject to noise injection and to charge pickup from proximity to a charged object like a moving hand nearby. I have seen this behavior in semiconductor wafer testing. The cure for eliminating the noise injection was to switch from FIMV to FVMI.
You are buffering the voltage across the 100G resistor with an op amp that has an input impedance of 1T ohm. You would want an op amp with much higher input impedance for this kind of buffer.
Make a simple picoammeter for use on the low side of Rx and have a well known voltage across Rx. There are open source schematics for a simple picoammeter on the web.
Forcing a very tiny current into a very large resistance is subject to noise injection and to charge pickup from proximity to a charged object like a moving hand nearby. I have seen this behavior in semiconductor wafer testing. The cure for eliminating the noise injection was to switch from FIMV to FVMI.
You will also need guarding and shielding. The settling time for this measurement could be very long as in many minutes or more than an hour depending on the effective capacitance of your fixturing. You will have a very tiny current charging some unknown capacitance. You can watch the RC time constant with a display of the picoammeter current being measured.
I used a high resistance meter, and the probe wiring is very tricky. The instruments I used applied up to 1kV to the Unit under test and measured the resulting current and assumed the source voltage would be somewhat shorted by leakage. The current sense has to be shielded so that any leakage goes to ground and not the current sensor. I found that everyday things like AC power connectors had significant leakage.
I noticed that, and the TLC272's I tested were in pristine condition
I measured it in circuit
The oscilloscope shows nothing, just a flat line with a hint of 50Hz (at both opamp outputs)
I have used this method to be able to use a cheap panel voltmeter as indicator.
The impedance in follower mode is greater than 1T, and anyway the problem is not there, it is in the CCS.
The output of the follower serves as a guard, and the whole circuit is completely enclosed in a metallic case
Maybe I don't get it but it looks like the bsp runs entirely on leakage current.How do you even open it with 100mV? Are you using tlc272 offset to open it?
I am operationally quite familiar with making and taking gigaOhm measurements, and I suggest that your topology is flawed.
Usually the stimulus is regulated voltage, and the leakage current is the measured quantity. Also, in general these tests are performed at a minimum of 50V and up to 10000V, depending on the characteristics of the DUT.
What is the purpose of the FET? Why not just use the op amps?
When you want to look at a sensitive measurement with an oscilloscope, you connect the scope across the guard output. Commercial high resistance gear has a guard circuit that can drive milliamps.
Here is an example of a picoammeter circuit. Buffer R2's voltage and look at that on your scope.
When you want to look at a sensitive measurement with an oscilloscope, you connect the scope across the guard output. Commercial high resistance gear has a guard circuit that can drive milliamps.
Here is an example of a picoammeter circuit. Buffer R2's voltage and look at that on your scope.
What is the purpose of the 1 Mohm and 10 Mohm resistors? Frequency compensation and overload/ESD protection?
If you haven't seen it already, Keithley's Low Level Measurements Handbook will likely be of value. You're basically building a source meter.
In other news: Keithley is apparently now part of Tektronix.
Tom
In other news: Keithley is apparently now part of Tektronix.
Tom
I am aware of that: I have built a teraohmmeter 10 or 20 years ago, and it does work, but it is demanding regarding guarding, shielding, etc.
I wanted to build a lighter, easy to use instrument, hence the "cheap and dirty" denomination.
I did my homework, and tested every critical part of the circuit, which is why I landed on the BSP92 as current controller (other types like BSS92 had a sizeable leakage).
As I said earlier, I opted for this "non-optimal" topology for simplicity reasons (direct voltage to resistance relationship).
I have somewhat progressed: I have retested everything in-circuit, and found nothing untoward.
Kirchoff laws have to be respected though, and since the components aren't the culprits, the parasitic current(s) must originate from somewhere else.
I had already milled the space between the critical nodes, but only on the solder side: that's where problems are the most likely, because of contamination, rosin residues, etc.
It didn't change anything.
I then resorted to scraping the component side: I couldn't use milling, as it would have damaged the components. It looked promising: the CCS current seemed to increase. Then disaster struck: the circuit touched the raw supply, resulting in a fried 78L12.
I mended the devastation, and I went further: I milled through the perfboard, but from the bottom side, creating isolated islands around the critical nodes .
It did improve matters: with the 250Meg resistor, the current was increased by 50pA. Not enough, but a step in the right direction.
This makes me think that the problem is caused by contamination trapped between the FR4 base and the top solder mask (which is purely decorative, since the perfboard is single-sided).
My only gamble was the insulation of FR4, but it seems to be the only point of failure. Now I have to find ways of rescuing the project without a complete overhaul: I have milled all that could be milled, and I have to be creative
I suggest, using your topology, sourcing 1mA into 1k ohm to create 1V. When that works, move on to 100uA into 10 k ohms and so on. At some combination of low current and high resistance, your topology will break down.
The picoammeter topology is so simple why spend the time on FIMV?
At the moment, I am working on creating a very repeatable and stable 1G ohm standard. It is not trivial. I want to have 10G, 100G and 1T standards as well eventually. I have a 10v standard + / - 1ppm and a good picoammeter which helps.
The picoammeter topology is so simple why spend the time on FIMV?
At the moment, I am working on creating a very repeatable and stable 1G ohm standard. It is not trivial. I want to have 10G, 100G and 1T standards as well eventually. I have a 10v standard + / - 1ppm and a good picoammeter which helps.
I had that thought as well. Would you be able to spin the board on Rogers or similar material?
Tom
Ordinary PCB material is used for sub pA instrumentation. Guarding, removal of material and good circuit design is all that required. That said, manufacturing yield of sensitive instruments is not 100%