Low MOSFET IDSS current

Sorry, forgot to say, this has to be a P-channel

Oh. This is for a 2N7000/7002, n-channel:

I think that's Win's data. A few pA.

I'd guess that a small high-threshold p-channel would be similar. BSS84? They're just not production tested down there.

Many thanks.

I have been looking at a number of datasheets and I discarded the BSS84 as I'd found a datasheet that gave an IDSS of 15uA and 60uA at two different temperatures.

On closer inspection I see that this was at VDS of 50V.

Another datasheet, this time for a BSS84L gives an IDSS max of 100nA, but at VDS of just 25V.

So it looks as if I need to be more specific about the manufacturer and the specific part rather than assuming all those with the same generic number were the same!

Most mosfets have remarkably-low leakage currents at room temperature. But it's expensive to test at low currents, and there's little demand for very low-current specs, so the datasheets list a much higher value, they're willing to test and stand behind. Mike, I recommend you purchase parts and vet them yourself for the application. Perform an incoming test, we call it.

Here's a cute simple thing you can do with a power MOSFET. Connect it with a series resistor to brightly drive an LED. Next use a pot and test lead to apply a voltage to the gate that runs the LED at a partial brightness. Remove the lead so the gate is sitting disconnected, but holding the voltage you applied. Observe the LED remain at its set brightness, so the gate is remaining at that voltage, perhaps for hours, stored on the gate capacitance. Leakage I = C dV/dt.

Many thanks.

Given Johns links to actual IDS currents, I was coming to the conclusion this might be down to the cost of testing. I'm aware that testing in a generic fixture down to sub uA is not a run of the mill expectation.

I'm not in a position to test every component that could run into 1,000 or more, although batch testing might be a compromise.

The BSS84L is about 5-10x the cost of similarly sized FETs, but hey!

You can do interesting things with a 2N7000:

Briefly touch the gate to B+, ground, or the fet drain to get three brightness levels. The gate leakage can't be mant electrons per second.

There is a simple push-push on/off switch effect.

It might be possible to demonstrate electron quantization somehow.

I test a few parts and assume that future purchases will be similar, at leat within a factor of 10 or so. Semiconductors tend to be very consistent.

Batch testing should be fine.

Nobody does AQL any more?

It's not practical to test standard parts for AQL. At a few FITs, you'd have to test hundreds of thousands of reels of resistors to verify specified quality.

We test a couple of parts on each reel to make sure they are what we ordered.

There are a very few parts, like some lasers, where we test every one. But that's not an AQL measurement exactly either.

The specification is what they're tested for. If you want some tiny current specified, the test takes longer, and you'll have to contact the factory for a special order. Or, test 'em yourself.

Off-the-shelf high speed testing: cheap, and sloppy.

More info here:

A common DVM, on a voltage range, can measure low currents.

This is very wrong. AQL is a sampling method for incoming inspection. It allows you to pick a sample size for a desired probability that acceptance or rejection of the lot is correct. It saves the cost and time of having to test the entire lot.

FIT is "Failure in Time" and is a reliability term. It is defined as a failure rate of 1 per billion hours. It obviously applies after the parts have already been accepted and placed in service.

There is a wealth of information available on both on the web.

OK, what sample size to you have to have to verify that you're meeting

And is used based on the now completely discredited reliability estimation method of MIL-HDBK-217. The idea that the failure rate of a system can be calculated as the sum of tabulated, invariant failure rates of the individual parts is no better than a random number generator.

I had to use it back in the early '80s in my satellite radio job, but even the managers there (who were all engineers) agreed it was a joke.

"Hey! If I leave out the input protection circuitry, I can reduce the failure rate by 600 FITs! And if I get rid of one of these parallelled MOSFETs, I can save another hundred!" (Not.)

Cheers

Phil Hobbs

Some of our customers want an MTBF calculation, based on the mil hbk or Bellcore numbers, so we do it for them. Their quality people are invariably lunatics so we do what we need to to keep them off our case. DGMS on SPC.

Most of our products have field failure rates far below the calculations. When one doesn't, there's a reason. Like "it's heavy and customers tend to drop it."

Quality people are invariably lunatics. If they were sane to begin with, the job drives them nuts very rapidly. (It's a bit like social workers--the ones who are actually sympathetic can't stand the job.)

Part of the reason, I think, is that their only performance metric is adherence to arbitrary rules, like six-sigma, continuous improvement, and whatever the current version is of ISO-%$&#@@!!.

Continuous improvement may be the stupidest of the lot--if you had one failure in the previous arbitrarily-chosen measurement period, God help you if you have two in this period. And God help you again if you ever have zero failures--because then you aren't allowed any more, ever.

Parts nowadays are very reliable if you don't mistreat them.

Cheers

Phil Hobbs

John used the term, not me.

It has nothing to do with AQL.

Let's talk about AQL instead of FIT.

That's not a problem for us. A quality audit team shows up and I always ask them...

Do you know what we make? Have you ever seen one? Do you know how many we ship? Do you know the field failure rate?

Of course, all the answers are "no." All they care about is their cult rituals.

But it seems as if some of their other vendors do have serious DOA and field failure rates. But the rituals won't help that.

Yes. Incoming inspection is not reasonable (parts on reels, complex chips) or necessary. There is no way we'd do incoming testing on an