High-value resistors and leakage

Electrostatic voltmeter! There are some beauties on ebay now and then.

I have a Sensitive catalog around here somewhere...

ftp://jjlarkin.lmi.net/Sensitive.JPG

John

Phil Allison is right - 10M isn't spectacularly high.

Surface mount doesn't really give you enough space between mounting pads to stand off several hundred volts, and an SMD resistor sits close enough to the board to trap droplets of dirty water or other contaminated solvent which can introduce inconvenient leakage currents.

Buy resistors rated for the job - preferably through-board mounting and use "fish beads" (little ceramic beads threaded onto the leads) to stand them off from the board. For really high resistances, you can buy tinned pins mounted into a cylindrical teflon (PTFE) collar, which you then plug into an appropriately-sized hole drilled through the board.

I've done divider networks for photomultiplier tube dynode chains that way - it drives production nuts, but the teflon collar stands off a kilovolt without any trouble.

John

More like the DC bias would completely interfere with your measurement.

Think, though: if you had a friend in the semiconductor industry you could make a 500V varicap, and wiggle the ground end to check the capacitance. Then deduce the top-end voltage from the capacitance.

Or use a 1N400x.

When you're done you may have spent ten times as much work for a circuit that uses ten times as much power to get a measurement with ten times more error than you could get with a resistive divider!

You could use s string of many (say, 20) equal resistors and tap your measuring voltage across the "bottom" one. They'd all behave the same (capacity, tempco, ...)

robert

The DC bias is the thing you'd be measuring!

Think? That is what I had in mind.

I never said it made sense.

You could build a varicap-tuned LC oscillator and measure its frequency. Any diode is a varicap.

The problem with varicaps is that they have terrible TCs, and the TC varies with the bias voltage. Messy.

John

I thought you meant a variation on what Jim did, which is to vary the capacitance and look at the resulting AC current (or voltage). You can do that with a motor and a capacitor, but not with a varicap.

^^^^^^^^^^

We'd need to define that word :-)

As Phil said, 10M ain't spectacularly high. But 0603 is no good for several hundred volts. Vishay has reasonably priced long form factor resistors. Or use several 1206.

Ok then, here is an alternative:

Voltage t.b. measured | \ / \ / | o | o +-|| ->|| +-||- Pulse to measure | o | o------------------o to ADC | o | \ / \ / | GND

(created by AACircuit v1.28.6 beta 04/19/05

FET gets pulsed on only for a brief period so the ADC can measure the voltage. Then it gets turned off until the next measurement. This will reduce the current consumption of the divider greatly. The lower the duty cycle the less energy it'll steal. The FET needs to be rated at the voltage level of the rail you want to measure plus the usual margin. Don't use a big old fat FET because then you'll have to wait for its capacitances to settle out.

The FET method was my #1 choice, and what we settled on -- he didn't want to mess with special resistors or worry about circuit construction issues (they just recently had a bunch of circuit boards die in board test because of a process change by their assembly house, so he's extra- sensitive to that this week).

But I wanted to have another solution in my pocket -- you always want to have choices.

Hmm..never thought to try SMT "divider" schemes; have used a 1Gohm leaded resistor with no problems. Try a "U" shaped driven shield around it as well as a driven plate on the back side. That should give isolation to the point of variations are solely due to the resistor.

I have some samples of some 1T 0805 resistors.

John

Just the thing for shunting 1G resistors to bring them down a tenth of a percent or so.

"Naomi Price"

The original holder of the " 'Flatso" trade mark is long out of business.

Maybe you can provide a link to some in 10Mohms that are readily available.

.... Phil

The other choices tend to be more pricey. But whether FET or not, best not to saddle a 0603 with several hundred volts. It's not kosher.

The "Slim-Mox" page you posted earlier are that type. The term was similar to "kleenex". "Flatso" covered most of them as they were on flat Silicon Oxide substrate and had thick Metal Oxide resistance media on them.

OK, so it's about guarding the high impedance signal path. Easiest is not to use a resistor divider at all, but to use the high resistance as input to an op amp current/voltage converter. The op amp (-) input is at pseudo-ground, and can be easily surrounded by a grounded guard ring on the printed wiring board (but this might mean you need to use large-package chips for your amplifiers).

Since you claim the input is DC, you can bypass the 10M resistor with a capacitor, to lower the Johnson noise a bit. Power-on transient through the capacitor can be an issue; diode-clamping the (-) node of the op amp to GND will be worth considering.

If you do really want to use a resistor attenuator, a follower amplifier and a driven shield electrode will be useful (the follower can be just a FET). Then, stray resistance will only be excited by small constant (not signal) voltage, and stray capacitance will result in nil current.

It was exactly this scheme that Philbrick used, for the 'operational parametric amplifier' (?P2), also AD310 (?) from Analog Devices.

The high-Z (-) node was a varicap bias, the (+) node was not such a high impedance.

The benefit wasn't high voltage accuracy, but low input current (with better aging than any subsequent MOS amplifiers offer).

At 10 megs, none of that is necessary. At 10G, maybe.

John

Choppers are pretty good.

John