Low noise small signal MOSFET

A clean 2N7000 has gate leakage in the ballpark of 1 fA. But mosfets are a bit noisy, so you should use a big source resistor.

We just now need a very low-noise 50 nA current source. It's into a summing point, so a resistor would do. If we use a low-value resistance, like 10M, and a low voltage, the resistor Johnson noise current is horrible. 10 volts across 200M looks OK, but the resistor can't have shot or excess noise, and nobody makes 200M metal films. We are testing various resistors to see how noisy they are.

What are you making?

John

Well, as you know, ATE machines being what they are, testing for very low-current leakage is expensive and slow, and isn't done unless the manufacturer thinks there's a market for it.

I'd suggest you consider a JFET, many of these have very low worst-case gate-current specs, because there is a market. Alternately you can just ignore the specs.

As for noise, JFETs also have lower e_n than mosfets. BTW, did you try a larger-value source resistor? That reduces the effect of e_n.

The 10uA shouldn't be any problem. Consider a ~100k resistor, voltage source and the voltage feedback loop to keep the current constant.

Vladimir Vassilevsky DSP and Mixed Signal Design Consultant

That analysis is flawed. 'hfe' isn't proportional to your current, unless you somehow have decided to start with an accurate sub-microampere base current, and amplify it with a grounded-emitter transistor.

The base current is a (maybe) half-percent of the emitter current, and its drifts are only a fraction of that.

So, what are the REAL requirements? What is your accuracy target, what is the output voltage range, are you OK with a calibration procedure? And, what power is available? Iit's easy to do this kind of thing with a good voltage reference and sense resistor, but it takes a few volts, and the output compliance range has limits that depend on the power supplies.

If you want something with written specs talk to this company:

The example specs 10pA typical but I believe they can provide special parts for you.

If nothing suits your needs you could consider peltier cooling. Maybe use one of the modules from the laserdiode guys.

How about an actual specification? You can get BJTs that are more stable than most resistors.

Cheers

Phil Hobbs

have a

would

electrons?

Very helpful. Thanks.

--------------------------------------- Posted through

ve a

s <

s?

You might consider an electrometer-grade J-FET. The 2n4117a was an old stand-by.

But, 1pA max. leakage needed to make a 10uA current implies you need 1 part in 10^7 precision. Really? Wow.

A less demanding application might just use a Howland and your favorite op amp. Cascode a FET for high speed or compliance.

--James Arthur

The data sheet spec tels you what the TESTing parameter is. The lower the current that has to be tested, the longer it takes: lower the "limit" by an order of magnitude and (without nasty tricks that will most likely not be used), the test time is increased by an order of magnitude. Too damn expensive to take more than a second to test ONE parameter. Iffen you wants two no, than tesstem yersself.