The ohnosecond

Not to change the subject (what? me? never!) but we're working on an ADC board that will have ranges down into the millivolts, and a couple tens of picoamperes of leakage into a critical node will gobble up our error budget. One of my guys wants to actively guard the critical node

- quite a rambling node, lots of parts involved - with the guard driven from the follower opamp that buffers the signal into the ADC.

So thoughts are...

If the guard trace is on layer 1, under the solder mask, and leakage is on the pcb surface, the guard doesn't help much.

Unless there's leakage on the pcb surface itself, under the mask.

Or unless there's some vertical conductivity of the mask itself, allowing it to reach up and slurp any leakage trying to creep over the guard trace itself.

(I kinda doubt either of the last two is the case.)

Maybe the main virtue of guarding is that it enforces spacing.

We measured one randomly-selected, much-handled bare board with a crude leakage-test setup (9 volt battery and a Fluke DVM) and couldn't find any trace leakages at all, with a sensitivity to about 1e12 ohms.

The Fluke handheld DVM must itself have pA leakages... impressive!

Clean/bake/coat may be a superior way to get low leakage if there is indeed a problem. Except for the mess.

John

Don't put solder mask over the guard trace, and expect extra inspection/rework around those pins?

The ohnosecond is a very short increment of time. It is the time from applying the wrong voltage until the device no longer "meets its specifications".

I know that 7500V/inch is a good rule for long term voltage gradient on a PCB but I can't find a good reference for a rule for shorter amounts of time. Does anyone know of one?

My PCB may have a high voltage connected for perhaps a few minutes and I don't want the PCB to be too huge.

BTW: Clamping the voltage is not an option. The source impedance is near zero and popping the fuse is considered a failure.

I read in sci.electronics.design that Ken Smith wrote (in ) about 'The ohnosecond', on Sun, 6 Nov 2005:

For you, UL standards usually have decided opinions on the matter. For others, IEC 60950-1, IEC 60065, IEC 60335-1 etc., etc.

On Sun, 6 Nov 2005 17:18:52 +0000 (UTC), snipped-for-privacy@green.rahul.net (Ken Smith) wroth:

I base all my measurements on the FemptoFuck.

Jim

I was of the understanding that the ohnosecond was the amount of time taken from switch-on to when you realise you've done something wrong. In this case you would have to add human sensory response and reaction time. Also, there has always been debate as to whether or not the ohnosecond is taken from before or after the actual expletive.

Dave :)

Not really. Remember this is the PCB material its self that we are worried about. The failure of a PCB at high voltage can take quite a while. It starts with a few ions moving a few copper atoms along the gradient. Slowly but slowly a little finger of copper grows where there wasn't one in the past. As the finger grows the process slowly speeds up.

I have just such parts on the PCB. Now the question is "how many volts per inch on the PCB material".

They are far from perfect. You have to run both sides of the LED in the opto out to the outside world if that is where the signal comes from. If you apply 1000V between said LED's legs, it takes very little time before it isn't a opto-isolator any more. In the process, it can make a surprising amount of light btw.

Among the signals I'm protecting are ones that move at up to 1MHz. Relays are not a option, unless I invest in a MEMS relay.

Try this with most that you can buy. Apply 12V and see that your equipment works right. Disconnect the 12V and replace it with 440V AC. Diconnect the 440VAC and reapply 12V.

I have to run wires in and out of the product. Without connections to the outside world, things would be easier but that isn't the world I live in.

In article , John Larkin wrote: [...]

You may want to leave the mask off.

In the extreme case, you can use a TO-92 JFET and a teflon standoff to keep the signal off the PCB totally.

There is always some leakage everywhere. The surface of a clean dry PCB under a solder mask is not very leaky. If the process is dirty, you can get pA-nA of leakage.

I haven't been able to locate anything about 1 minute time frame.

snipped-for-privacy@green.rahul.net (Ken Smith) wrote in news:dklt6a$ui5$ snipped-for-privacy@blue.rahul.net:

I've seen PCBs with slots cut in them to provide greater HV resistance. It makes the path much longer.

I read in sci.electronics.design that Ken Smith wrote (in ) about 'The ohnosecond', on Sun, 6 Nov 2005:

Clause 5.2.2 of IEC 60950-1, for example, requires the hi-pot test to last 60 s, which is about 1 minute. (;-)

You would also need to read bits of the long clause 2.10 and probably Annex G as well.

But having done that you will know more about clearances and voltages than you ever expected.

On 6 Nov 2005 21:57:33 GMT, Jim Yanik wroth:

And I've seen high voltage opto's in DIP psckages where the leads were spread out just a little on each side to give an extra .1" between input and output. The came that way from the manufacturer.

Jim

I'd use IEC 60065. There's a nice graph.

Graham

I really don't understand your long / short term idea.

You seem to think you might 'get away' with reduced insulation for short periods.

A truly daft idea. Wait till it goes bang and you'll understand why.

Graham

Yup. 0.4" spacing in place of 0.3".

Originally to meet VDE specs IIRC. I use them. Helps keep pri and sec sides apart ( isolation barrier ).

Graham

I almost sounds as though a PCB isn't the right tool for this particular job. Is there any way you could build the relevant bit of circuit using a cluster of PTFE standoffs like the old-fashioned tagstrip?

I'd say anything between 1 to 10 seconds. The damage is typically instantaneous but at around 5 seconds it starts to dawn on you why the circuit is suddenly dead.

In my final year at university, my project was designing a PCI card. So I was running my PC with the casing open & an exposed motherboard. & the PCI card had wire-wrapped wires dangling all over. While probing an input pin on a chip on the card the PC suddenly went dead. It took 5 seconds to spot the loose 12V wire dangling from the card onto my motherboard. The poor old motherboard has never booted since.

And the PC was a Motorola StarMax Mac clone with a PowerPC 603e. Considering that the clone licensing was ended a year later the machine was quite rare.

Generally speaking ohno = 5. ;->

Cheers.

Thanks, I'll check it out.