Protecting a 22bit ADC input with low leakage diodes

You haven't really provided specs for what you need. Simple diode clamps like these can't give you a sharp transition between clamping and zero leakage. How much of the 500mV input range can you afford to give up to that soft transition? Or can the input signal stray beyond this range without causing component or signal problems? What temperature range will your device need to work in? Is this a one-off (hand tweaked) or is it for production?

If this transition range is tighter than can be achieved by this simple clamp, you'll have to consider an active clamp, at which point you'll have to think more about things like clamp speed.

HTH--

Leakage is monotonic with voltage. Whatever it is at 75V, it's less at

1V. And it's zero at 0V.

--
Tim Wescott 
Control systems, embedded software and circuit design 
I'm looking for work!  See my website if you're interested 
http://www.wescottdesign.com

Move reverse voltage makes more leakage current. Increased temperature really increases leakage.

T/c's are just wire, granted higher resistance wire than copper. A really long run might have hundreds of ohms, a short run a few ohms. Omega has reference data online.

Transistors tend to have less leakage than diodes.

A BFT25 c-b junction leaks femtoamps.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

Don't the collectors pins do better than the emitters for that purpose?

--
 Thanks, 
    - Win

It probably doesn't matter in this application, with a low-impedance t/c signal. Using the emitter of the PNP dumps the sink current into ground, which eliminates a clamp supply.

Using the emitter of the NPN reduces the negative clamp voltage a bit.

I don't know which ADC he's using, but there may be a latchup hazard of its input swings below ground. A capacitor would attenuate spikes some.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

At the top end, I just need to clamp it so it doesn't go above 5V, which is easily achieved by connecting the upper BAS116 to +0.6V. That will clamp the signal to +1.2V while leaking "zero" if the signal is below +0.6V.

At the bottom end I would like to clamp it so it doesn't go below

-0.6V, which is easily achieved with one BAS116. However I would like to achieve something like -0.3V but that is not possible (with only P-N junctions).

The ADCs I am using are ADS1118 and MCP3550 which both have absolute max negative Vin of -0.3V. That is the bit which concerns me a bit, but if the current is limited to a few mA it may be ok.

The protection is really only for static (but there are 100nF ceramic caps signal-gnd anyway) and installer wiring errors to another terminal which outputs +20V.

So I think my simple scheme will be OK, subject to getting the diode leakage low enough.

Frank Miles wrote:

Sure, but something was telling me the TC voltage comes out of quite a high Z.

OTOH I guess thermopiles could not be used to power underwater submarine detection stuff if they could source only a few uA :)

That is ingenious!

Your BFT25 is a 2 GHz transistor. Is it more special than e.g. BC847C/BC858C? We already use those. I think the BFT25 has a much smaller chip though.

I think there are trade offs.. B. Pease talks about diode/transistor leakage here,

George H.

Well, umm, that's what I do.

The smaller chip is probably why the BFT25 has such low leakage. For your clamping application, you don't need fA leakage but you want a hunky clamp, so I'd use something bigger. The 847 type is probably fine.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

Why should it be? Just have to use ultra-low leakage diodes for 22-bit performance.

How about these?

500 picoamps max at 75V and 25C which would put your leakage current at around a volt Vr into a range that you'd hardly be able to measure. You can buy duals with series connection so you only need one device per line.

--
Regards, Joerg 

http://www.analogconsultants.com/

Thermocouples and t/c reference junction sensors are sufficiently inaccurate that a 10 or 12-bit ADC is good enough. 1 few nA of leakage won't matter.

The zillion-bit internal PGA delta-sigma ADCs are handy for t/c measurement, but the clamp won't be the accuracy limit.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

Interesting part, thanks! It's slow to turn off - any idea if its turn-on is also slow? {I'm thinking of JL's transient tests on diode turn-on}

If your source resistance is as low as a thermocouple, do you really need to keep clamp leakage all that low? Couldn't you tolerate, say, 1uA without issues? 0.3V between "no leakage" and "fully clamped" is probably not achievable with a simple diode clamp. The 0.1uF would help a lot with static - though if we assume a source capacitance of 100pF and protecting to 0.3V you'll only be able to tolerate ~(10^-7/10^-10)*0.3=

30V, presumably inadequate.

Yes, usually the clamp isn't the limiting factor. Once you get into the single-digit nanoamps things such as encapsulating or potting become important. Else just a wee bit of condensation or even breath can throw measurements off. Flux clean-up also becomes very important.

--
Regards, Joerg 

http://www.analogconsultants.com/

It's usually best to provide helper rails 1V or so beyond the supply rails or desired limits and then add a resistor from clamp point to the ADC. That way the ADC's ESD diodes will never see current beyond what they could take and the clamping diodes won't leak much.

--
Regards, Joerg 

http://www.analogconsultants.com/

Ah, make that 300V (still inadequate)

r t n

Turn-off is stored charge. You've got to inject the charge carriers to get the diode to conduct in the first place , but that tends to be less of a pr oblem.

--
Bill Sloman, Sydney

Consider that 1M-ohm / (2^22) is in the neighborhood of 1/4 ohm. As JL said, a thermocouple is just a hunk of wire (well, two hunks of wire spot- welded together), but they might be hanging out on the end of a long cable. If the total resistance is higher than 1/4 ohm, then you've lost your 22-bit accuracy.

If you've never designed circuits to this sort of accuracy, keep in mind that you have to start taking a whole lot more stuff into account than you've ever had to before. Just as an example, consider that a typical "jellybean" op-amp has a GBW of 10MHz, which means that GBW/(2^22) happens at around 2.5Hz, so at some frequency lower than that you need to be throwing the "op-amps have an infinite gain" assumption out the window. Moreover, I wouldn't trust that any candidate op-amp that I've chosen actually has a DC gain higher than 1M or so -- so even at DC you can't necessarily trust your assumptions.

--
Tim Wescott 
Control systems, embedded software and circuit design 
I'm looking for work!  See my website if you're interested 
http://www.wescottdesign.com

That's neat - thanks. They are very pricey however - about 30x more than a BAS116.