I'm trying to protect inputs to an op-amp from the outside world and readin g up how ESD protection diodes/TVS can help. Datasheets may claim to withst and a particular standard e.g IEC 61000?4?2 (Level 4) which specs 30A peak in the first 1ns, then reduces 8A at 60ns. A small diode ob viously has hard time with this and so manufactures give an associated ESD clamping voltages, which might be 20V for larger diode (more capacitance) a nd 100V clamping for smaller diode. This sure helps, but makes a very low i mpedance spike that is way above absolute maximum ratings for most analog I Cs.
Do people use two ESD clamping diodes, first clamp to say 50V, then through a current limiting resistor to a second TVS, before finally going to the a nalog IC.
It looks like a small ESD diodes can be modeled as a ideal zener with 1 ohm dynamic resistance so I can get final clamping voltage.
ing up how ESD protection diodes/TVS can help. Datasheets may claim to with stand a particular standard e.g IEC 61000?4?2 (Level 4) whi ch specs 30A peak in the first 1ns, then reduces 8A at 60ns. A small diode obviously has hard time with this and so manufactures give an associated ES D clamping voltages, which might be 20V for larger diode (more capacitance) and 100V clamping for smaller diode. This sure helps, but makes a very low impedance spike that is way above absolute maximum ratings for most analog ICs.
gh a current limiting resistor to a second TVS, before finally going to the analog IC.
hm dynamic resistance so I can get final clamping voltage.
I use Schottky diodes to the power rails. So far they have worked fairly w ell, no returns for anything related to input failures.
Good rule is never let silicon see the outside world - always have something in between. Some series resistance and/or inductance will act with circuit capacitance to reduce slew rates and help the clamp diodes. This rule gets harder with RF and ultra fast stuff but you said op-amp so I guess you can add something in series and them clamp diodes. Look at Art of Electronics 3rd edition page 362.
Yes, that's a good way to do it, absorb the big hit then add as much resistance as you can afford (might be a lot for a signal input, but not for a high bandwidth input or an output), then either add more protection or use the device's own protection (most things are 2kV HBM, whatever that comes to after the clamp).
I don't think anything I've made has failed IEC 61000-4-2 testing due to damaged lines. More often it's something that I didn't make and didn't get a chance to improve, and/or due to upset within the circuit because of induction into sensitive traces or poorly filtered reset lines, that sort of thing. (In other words, a class B failure mode, which is usually acceptable.)
Seven Transistor Labs, LLC
Electrical Engineering Consultation and Design
When something works in thousands and thousands of units, it's superstition??? I thought that was pretty good evidence.
I think you have it backwards. Testing to a spec only proves the device will survive a test to the spec. Real world survival depends on how closely the spec matches the real world environment.
What you are saying is that measuring your gas mileage is less reliable than the tests the automaker does to the US government standards. It doesn't matter how much fuel is used on the road, what matters is only the fuel used in the lab.
ading up how ESD protection diodes/TVS can help. Datasheets may claim to wi thstand a particular standard e.g IEC 61000?4?2 (Level 4) w hich specs 30A peak in the first 1ns, then reduces 8A at 60ns. A small diod e obviously has hard time with this and so manufactures give an associated ESD clamping voltages, which might be 20V for larger diode (more capacitanc e) and 100V clamping for smaller diode. This sure helps, but makes a very l ow impedance spike that is way above absolute maximum ratings for most anal og ICs.
ough a current limiting resistor to a second TVS, before finally going to t he analog IC.
ohm dynamic resistance so I can get final clamping voltage.
Yup, is pg 362 where Win and Paul talk about depletion fets as input protection? If you can stand ~1k ohm in series then two lnd150s (gates and sources all tied together) is my new fav input protection. They current limit at about 1 mA and you figure (and can measure) that the protection diode on the opamp can handle the 1 mA.
Thousands? Maybe. I know products that shipped 10k's of units over more than a decade, without ever testing for emissions. They never got called on it. They even had customers complain about e.g. radio reception while operating.
If it doesn't have a number on it, it's just superstition.
Most customers are also disinclined to complain, unless they're heavily invested in it. If it's some throwaway part, they'll just buy another, which may or may not be a repeat sale, or lost repeat sales. From a business standpoint, it could be good or bad.
Strictly speaking, just passing doesn't give you a number, only a minimum. It's a pretty reasonable minimum, though, and, say, 30kV ESD is really, really uncommon in practice. Even in dry cold winter. More than uncommon enough that you'll not see it in thousands of units.
Fortunately, EMC tests are unusually representative, being a simpler environment than, say, testing an automobile at some very well defined temperature, air mixture, pressure, speed, acceleration and so forth.
False equivalency. This would be more like, the automaker measuring failure rates on their millions-strong fleet, versus an enthusiast group counting failure rates among their thousands-strong subset.
Just basic statistics.
Seven Transistor Labs, LLC
Electrical Engineering Consultation and Design
V_DS_MAX of LND150 is only 500V. Better use a TBU, this device is specifically designed to do the job. You can get 850V ones. My fav is a 90V GDT + a 50mA TBU + a TVS + 200Ohms in series. A nuke-resistant combo. ;-)
So?? Emissions are something that can be a problem that is never solved or more importantly, the source never identified. Ask any ham radio operator ... They fell the pain.
What are you talking about?? This is not in any way responsive to the disc ussion.
My product goes into million dollar systems where they very much WOULD comp lain if there were ANY problem. In fact, they asked about returning some 5
0 or so units once. They were failing a lot of units during their integrat ion testing. While we were negotiating the shipping, they must have figure d out what they were doing wrong with the installation or tests because the y dropped the return. The point is they will complain when they have a pro blem.
And that number tells you nothing other than it passed the test. The devic e can still be destroyed in the field by discharges that exceed the tests. You make it sound like passing the test is some guarantee of not having pr oblems. That is the issue we differ on.
The purpose of the standard is to provide some minimum level of protection. Saying that a product is "superstition" if it hasn't been put through tes ting is pure BS. The product can have protection even if it was never test ed.
That does not contradict my statement.
Your example is a false equivalency. You are talking about two groups meas uring the same thing. No one else is measuring my product so your example is meaningless.
My point is testing to a spec in a lab is fine, but if units are in the fie ld and never fail, they must be adequately designed to work in that environ ment. If the failure rate is better than one in many thousands, that is ce rtainly good enough for the vast majority of applications, aerospace and li fe support not included.
I once found a design flaw in a military radio as it was being put into pro duction. The flaw was on paper, a small timing violation on a memory part. I pointed it out but because the design had passed "acceptance test" it w as not rectified. Passing test should never be confused with working. In this case they may never know if it was a problem or not. Timing flaws can fail only when everything is at its worst, voltage, temperature and proces s and that one routine that bangs hard on the data bus is used.
Good! You agree it's superstition unless measured.
(The general population of hams is also a relevant example. Proof of this comes in many forms, one example being nasty antenna designs on the market -- people are actually buying them; let alone all the hacky plans out there for free -- that make claims that violate physics, or thermodynamics even. The most common error being poor tuning and balance, so the feedline is part of the system, giving much more bandwidth/directionality than the intentional element size alone would suggest. There are very smart hams out there, who know to test and verify these things -- just not very many. The
90/10 rule applies, as it does in any population.)
Good, you have responsive customers. Valuable!
The purpose of the standard is to provide some minimum level of protection. Saying that a product is "superstition" if it hasn't been put through testing is pure BS. The product can have protection even if it was never tested.
Can you design a device that will withstand the Sun going supernova*?
If not, then you can't guarantee anything. No way, no how, absolutely nothing. There is not, and never can be, a probability of zero.
*Well, our Sun exactly, probably won't, so if nothing else, using insurance to solve the problem instead, it should be pretty cheap to get someone to underwrite that...
Now, if you want to talk failure rates, statistics -- we can have a meaningful discussion! There is no absolute ESD, only an expected value over some period of time, or likelihood per event. This is what the test proves. 15kV (air) happens to be pretty representative of the worst you'll see in winter. As a result, it's near enough of a "guarantee" to the average person.
Or if it's in the spec sheet or contract that it withstand 15kV ESD and not a volt more, that's literally just passing a test. Whether or not that's a problem to the customer, is up to them, and them alone -- it's their fault if their spec turned out to be too low. Guarantee doesn't fit into that at all!
As for protection without test, yes, you can design things in. This is easier to do for, say, DC operating point -- managing automotive input (reverse, swell, load dump and transients), say, than for very fast events like ESD.
It is perfectly fair to note that, I don't know what the failure level is of my protection circuits -- they might've passed (at whatever acceptance level), but that doesn't tell me where they are likely to fail (number of shots * energy, say).
We're all, always, just guessing into the void. The point is to be guessing less, to calculate the things that are easy to calculate, and to know the error bars on the things that are more guess than know, rather than just seeing what sticks.
Or not even that. Just throw everything and do what sticks. That can work too. Using customers as beta testers. Lots of consumer crap is shipped on that model. People still seem to buy it, so it's a valid business decision.
Just not the kind I prefer.
(Not implying you are doing THAT, by the way.)
LOL! So your customers don't know either? :-) If it's not even in the spec, you are completely forgiven -- literally, and explicitly: no one knows, and no one cares!
The downside is when no one realizes that it is actually a problem; I'm currently working with a large electrical manufacturer, who until I brought it up at a meeting, never considered the effect of nearby distribution lines interfering with their product (which, so far, had been designed and specified explicitly for single line use, but I think there was always intent that customers would use the product on multiple lines). One of those "unknown unknowns" things.
Yeah, it's at least, what, ~3 sigma, which is about as good as any customer service gets, for example.
Military understands statistics, if they find about that many fail acceptance testing, that's probably just right, to spec. (But in that case, the customer knows -- they're testing!)
Ah yes, lovely. And there's the matter of "not tested, guaranteed by design" -- but they never give you the error bars on those numbers...
In the above case, you could've tested the radio over all extremes: temperature, voltage (however applicable), maybe aging (accelerated aging test?). Thing is, that's a lot of expensive, per-unit testing. You're testing a per-unit variable, not confirming a design process. You should be buying parts that have that baked into /their/ testing process, and designing to that. Which sounds like it was the failure step, a design oversight that is -- or compromise, perhaps, which can be more insidious!
Seven Transistor Labs, LLC
Electrical Engineering Consultation and Design
You seem to think you are good at this, but you aren't at all. Putting wor ds in my mouth isn't useful to make your point.
I have no idea what point you are trying to make. Do you?
Responsive and no complaints. So the product is meeting all the user requi rements including ESD and EMC.
I think you are helping my point. The test is just one way of meeting cust omer expectations.
Not sure what you are trying to say. You use a lot of words that dance aro und the subject without making a point.
What is your point???
Uh, "acceptance testing" for commercial goods is testing at the point of re ceiving the goods. In purchase agreements the product can be returned unti l the acceptance test is completed and passed which must be done in some ti me interval. That doesn't mean it is the only way of testing a product. I n my case my units are tested in the system. We offer a warranty which cov ers any failure in goods and workmanship for a year so they don't need the expense of formal acceptance testing at the time of reception. They order boards as they ship systems.
Yeah, people often don't understand testing. They think testing is some as surance and it is far from that. Testing can show that a defect is present , but it can never show the absence of defects other than specific defects the test is designed to cover.
BTW, these are military radios and they are tested at high temperatures, ev ery unit. I don't know if they test at the low end as well. I didn't work in the test department.
Often, the cause of an RFI problem is very hard to locate: this example comes to mind <
where a very savvy customer exceeded the warranty on the device long before finding its flaws. No amount of measurement of RFI would catch that particular flaw. It was a transient with the interlock switches, and test protocols for RF output don't include a monkey introducing transients...
Good design doesn't just quiet customer complaints, it keeps quality up over entire industries. The best practices might involve accurate measurement, but also should cover all of the odd use cases (and that cannot be done, IMHO, with standard procedures, though field tests and design reviews have a chance).
Oddly, that is why I like a particular kind of (usually antique) chair: there are Windsor chair designs that got thoroughly developed over decades of chairmaking by dozens of craftsmen, But, I have to find 'em as antiques, because the spindles cannot easily be fabricated from dried boards. For best results, one takes green wood through splitting and rough turning in one operation, then final assembly when some of the parts (not all) are dry.
There's a few craftsmen making good chairs of that kind, but mass production can't really do it. Testing quality in is just an impossibility.