Determining the level of protection


For a few different categories of interconnection from a PCB containing inputs and outputs, how much protection from ESD, overvoltage, and electrical surge should be applied?

Here are my standards:

  1. BNC or other physical connection logic/analog signal input from the external world on something that warrants being called an "instrument" (not some development board on my bench, but a final product that needs to impress a customer). This connection will be frequently connected and disconnected, and is likely to have bare wire adapters attached by the ordinary user while taking no ESD precautions.

This should have the maximum protection. Inputs should have a resistor sized for acceptable balance between input bandwidth and current limiting during ESD events and overvoltage application, dual diodes to the rails for primary ESD shunting away from the device, and another resistor between that diode and the device. The power rails need shunt protection from slower transients via a TVS and DC overvoltage protection via a zener or SCR crowbar.

Input network should be modeled with an RLC pulse source with component values and initial voltage conforming to the IEC 1000-4-2 ESD models, and verified that the protection and protected devices do not have their ratings exceeded. Actual testing should be performed as well.

  1. Multi-pin connectors such as D-sub and others that are intended to connect some other sensor or instrument to the "instrument", and that will not be changed frequently.

This one's a little more difficult. It might be very costly and take a lot of board area to put the full suite of protections on every pin in this case.

What I commonly see for these situations, is a single RC network. I wonder if the designers of these RC networks are certain that they can actually protect against standard ESD models? In my recollection from my recent SPICE experiments with ESD protection networks, RC networks just tended to shuffle charge from the external capacitor to the internal one, and causing the (presumed) protection diodes in the device to be protected to bear an excessively large surge current.

Thus, this protection usually can't meet the tougher 8kV and 15kV contact and HBM ESD models. In fact, I am not certain it is really intended to protect against ESD at all, but rather intended just to provide some noise filtering.

So perhaps most designers forgo thorough ESD protection on these multi-pin connectors, assuming that since they are not to be changed frequently, that there is little change of ESD damage occurring here?

Is this a wise practice on robust instrumentation?

  1. Connections from one board to another inside a chassis, assuming that reasonable protections are in place on each of the boards.

Here I think it is acceptible to provide no protections.

  1. Outputs from devices such as op-amps (analog) and logic chips (logic levels).

I rarely see ESD protections applied to these. Though I have tended to apply some protections to these as well. I have put SMD 0.2A fuses, to a pair of diodes to the rails. Then a resistor between my diodes and the device output pin. That way if there is an overvoltage applied, at least just the fuse blows instead of the whole output device.

Your input is of interest.

Good day!

Christopher R. Carlen
 Click to see the full signature
Reply to
Chris Carlen
Loading thread data ...

Never put fuses in. They fail, and someone has to open the box, pull out the board and replace them. There are other non destructive ways. Also you need a definitive spec to design to, a Mil spec etc. They give specific tests to do. With numbers. In Denver, you can easily get a 20K volt charge built up by static electricity and zap any touchable part.

Reply to
night soil dalits

How can such a thing be done without making a very large number of assumptions? Perhaps in product development there are statistical data sets which allow one to model the failure rates as a function of the level of protections? Who would do this, the board designer or some manager with a combination of business and statistical modeling skills?

That's why I'm inclined to use a zener for shunting DC overvoltage, where the TVS hasn't turned on yet and can't be relied upon to clamp before the device's power supply ratings are exceeded for instance, the

6.8V TVS might not turn on until 7.14V, whereas a 5.6V zener could absorb the intended fault current of say 0.2A (above which we just let the resistor smoke) while keeping well below 6.3V, so that the input holds below 7.0V as well.

We tend to still use a lot of linear supplies because of the lab environment with lots of low-level analog instruments strewn about the place, mixed up with all the digital stuff.

Hmm, are you saying that it might be wise to isolate the board's power planes with a LC filter between those and the power connections of the protection diodes?

What is a "coupling clamp?"

Are you saying that you short the input protections, so that the input is connected directly to the power supply of the device (or through the LC filter)?


Ah-ha! I just realized that is why the last board I looked at had the caps on the outside and the resistors on the inside. I thought "how's that supposed to work?" thinking about ESD (one of my obsessions, it seems). Since we never think about EMC regulations in this lab environment.

Like I have said before, I am in an environemnt where I am almost called to task (not by the customers, who usually appreciate the quality, but by the software engineer) for spending time to design in ESD, EMC, etc protections. We have a situation in which a software engineer specifies hardware, and thinks that "hobby grade" is just fine, since it's just for a lab, and will never be sold. It is getting me a bit frustrated since I am the one who has to spend the time redesigning things to eliminate the crosstalk, to put in the glue logic that they didn't realize they'd need, etc., etc. If they'd just have let me do it carefully from the start, then there would have been no problems.

But then the next project comes along and it's the same attitude, "just hack it together quickly with COTS hardware. It's easy, just connect the dots"

I will see what they have in the way of clamping arrays. I have some Littlefuse devices, which are a bit expensive, but nothing to worry about here where such a small relative cost increase is of little concern.

Yes, I haven't figured out a fool-proof method here. The main difficulty is the need to preserve the low output impedance, so the output needs to be almost directly connected to the outside world.

But if the user might accidentally

Yeah. I think it's a matter of eliminating 90% of the faults in cases like these.

Thanks for the input.

Good day!

Christopher R. Carlen
 Click to see the full signature
Reply to
Chris Carlen

its often a tradeoff between reliability and cost, usually requiring a detailed Total-Cost-of-Ownership calculation.


IMO SCR crowbars are of very limited use. They came about because a common failure mode of series-pass regulators was to go short, dropping full unregulated volts across the load, which was generally highly destructive. Uncommon nowadays, as SMPS dont tend to fail that way (although some do eg buck). Plus of course they only do something once the product has died. I'll be impressed if you can make one clamp a transient.

there are plenty of nice little SMT "L" filters (muRata etc) that have an input L and a feedthru-type C, which can feed your clamping array. Some of these are dirt cheap too.

yeah, coupling clamps, ESD guns, the full monty. I have a far more brutal test, involving bypassing the coupling clamp, and directly frazzling inputs, outputs etc. with a showering arc generator (sch posted to abse a week or two ago). If you can pass that, you'll piss in at the testing lab.

Mind you, it was a bit scary discovering weaknesses that way - slap it on the output of a (competitors) VFD, and *BANG* !! Fun at 1kW, downright scary at 1MW. But its a great feeling when the DUT doesnt skip a beat as you spark-erode your name across the output terminals.

how much does it cost when the unit croaks? customers *hate* that.

that'd be my guess. Or it is just an ineffective design - there are plenty of those around. Especially when it comes to EMC, ESD etc.

There are a number of mfgs that make n-tuples of clamping arrays etc. for exactly this purpose. And ferrite slabs with holes that the leads of the D-connectors pass thru etc. muRata is a good start.

most manufacturers also produce cheap shit. Because consumers cant see past the here-and-now and love low-cost products. Best not to think about how often they need replacing.


usually. Depends on mfg/assy practices. I did one board where a particular micro pin went directly to a connector, that turned out to be a convenient hand-hold. Our first run of 200 pcbs had 3-4 micros with this pin dead despite some pretty good anti-static measures, so in went a small cap and a series R. no more died.

its unclear, but I presume these go to the outside world, hence the over-voltage concern. Fuses are a pain, as its often easy to set up a fault scenario that wont kill the fuse, but will kill the chip - often just the right voltage will do it. But if the user might accidentally connect up mains, fuses are often the cheapest solution (but look at rupture current rating)

Cheers Terry

Reply to
Terry Given

historical data helps a lot here. Often its not too hard to work out roughly what it costs to rip a unit out of service, return for repair and re-install. the installation category (wrt transients) will give you an idea of the likelihood of getting zapped, as will the environment (dry = static zapped, humid = condensation problems etc).

IMO designers need to do this stuff, in conjunction with marketing. You know how many you will be making, what the unit and additional parts cost etc. Marketing can tell you (in theory :) what the cost to the customer of a failure is - perhaps 150 workers sit around doing nothing until it works again, perhaps it makes little or no difference. Mess with the operation of a dairy factory or steel mill, it'll cost upwards of $100,000 per hour.

zeners tend to have much steeper slopes than TVS do. BTW, a 6V8 zener will have much the same problem as a 6V8 TVS, only worse. why not a 5V6 TVS ?!

But you're on the right track.

fair call. the other problem with lab supplies is the twiddly knobs, which can get twiddled unfavourably. I have fond memories of a tech using my test setup to charge his car battery after work one day. Next morning, I turn it on and stick 14V@10A up the arse of my 5V circuit. which is why I now always turn all I,V knobs to min before re-starting circuits.

no, not at all. in place of input R.

The "L's" are really ferrite beads, so tend not to bug signals of interest, but present a decent Z to fast transients. I've seen (and done) a number of designs where the input protection is simply a cap and a pair of clamp diodes, where the cap is as large as it can be without buggering up normal operation.

That said, I have often done this:

Vcc----[100k]---+-----+-----to +ve clamp diode Cathode] | | [C] [Zener] | |


which is almost exactly what you suggest. Its a good way to slap a HUGE cap on an output that otherwise would be unhappy (eg RS485)

its a device about 1m long. Flat metal plate, about 6" above a ground plane. Piano hinge on long axis, fixed to a half-width plate, piano hinge, another half-width plate - a base witha double-hinged lid. The (grounded) transient generator (Showering Arc Generator) output (usually N-type coax connector) connects to the coupling clamp. Lift the lid, lay your cables on the bottom plate, lower the lid. Voila, lots of capacitive coupling, hence the name. DUT is connected to dummy (or real) load via these cables. Watch it go bonkers, and maybe even bang if the protection aint good enough.

I forget the relevant standard....

Nope, no changes to DUT at all. I attach a wire with a multimeter probe (IOW insulated so I dont get zapped) to the SAG output. Then I probe directly to the DUT I/O terminals, making evil little 5mm arcs (hence the ability to write ones name on the metalwork). much nastier than using a coupling clamp.

Same idea as DIY EMC testing keeping an extra 3dB below the limit, to ensure you pass the real (expensive) test first time.

This is very common. I have learned from bitter experience to always do these things properly. you might not need to comply with EMC regulations, but if your doodad flips out and screws up multiple experiments and/or fails, thats still a bad thing.

A corollary is: always use a proper PCB. My first employer had a little DIY pcb setup, and we did our own 2-layer non-PTH prototype boards. Hey, its just a prototype right? wrong. There are enough things to screw up without wasting a week due to a shitty home-made via, or a solder bridge. And the economics are easy, I could burn $500 before lunch looking for a problem, the cost of getting a real PCB made.

the phrase "f*ck off" is useful in these situations. OTOH you learn a lot of what not to do, and every design gets better. With a bit of practice, its just as easy to do it properly.

IMO s/w "engineers" are often the biggest idiots. They very rarely think about non-ideal operation - eg buffer overflows. These are the guys who often wont even use parity on RS232, refuse to implement CRCs, pick bit-oriented protocols (one bit = one command) which are really just noise sample-and-go-bonkers etc.

And its invariably s/w that is the holdup. Usually because testing is an afterthought, as is design. Many are little more than over-qualified typists. Those who immediately start coding are usually the worst.

Ask him how well the s/w works when the hardware dies.

A less-impolite phrase is "which part of NO dont you understand?"

ya gotta hate that. Beat him with the Principle of Reciprocity - apply all his arguments re. hardware to software, and watch him rebut the lot. Then point out what you have done :)

Or better yet, pull rank. Go higher in the chain of command, with a cost-benefit analysis.

This is a very common problem. We had a GM like that - as soon as he saw a motor turn, he reckoned the product was finished and so tried to produce/sell it. So we learned to leave that step until last, and even went so far as to disable that aspect of the software....

it often crops up with testers too.

you shouldnt need a fuse for ESD/EMC. Only some d*****ad attaching mains etc.

I did one which was a user +24V supply, that had to cope with 480Vac up its bum. A stompy TVS and a ceramic fuse (80kA rupture current) was the cheapest option by far. We didnt make it user-serviceable (why shouldnt they pay for their stupidity), but it worked well.

You're welcome.

Cheers Terry

Reply to
Terry Given

SCR is for overvoltage protection, not ESD.

The fuse blows if an external voltage source greater than the power rails to which the diodes are connected, is applied to the output. It isn't intended to protect the op-amp from a shorted output.

Good day!

Christopher R. Carlen
 Click to see the full signature
Reply to
Chris Carlen

Like someone else said, I agree the SCR is a waste of time for ESD transient.

There are devices called 'transorbs' which might be more appropriate than zeners.

Also on the market. You would probably take a different approach with the GPIB connector on a $100,000 vector network analyser than you would on the RS232 port on a PC.

From experience, a lot of devices are taken apart without suitable static precautions. I see it all the time from people who should know better.

What makes the fuse blow if you put too much on the output of an op-amp?

Think of a project where a medical doctor could use the instrument. Then loan it to him, and tell him it is delicate. If it can survive a medic, it can survive anything.

My boss says it is hard to build something a nurse can not break, but impossible to build something a doctor can not break.

Reply to

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.