Recommend me some 3.3V ESD diodes

[HwAoRrDk]

I am overwhelmed by the selection available for ESD protection diodes. It seems like there are a billion and one choices.

It seems to me that the large number of variations are mainly to do with (apart from the primary specification of working/stand-off voltage) ratings for capacitance, leakage current, and clamping voltage. Some parts look to be specifically tailored with respect to these ratings for certain applications like USB, CAN, HDMI, etc.

I'm looking for something for protecting some inputs of simple switches/buttons that the user will interact with, which will be interfacing with a 3.3V microcontroller (or possibly I/O expander).

So, are the following assumptions about selection criteria correct?

• Capacitance: for my application, largely irrelevant? I don't think something with even several hundred pF isn't going to meaningfully affect the signal. In fact, for filtering of noise or switch bounce, it can only help, I guess?
• Leakage current: lower is better, but I don't think this is critical so long as it doesn't affect signal levels (e.g. overpowering a pull-up, for example). I think even a few tens of uA should be okay.
• Clamping voltage: again, lower is better? Although, I am confused why many parts have clamping voltages so far above the working voltage - e.g. 10V vs. 3.3V. Is my understanding correct that you need current limiting (e.g. with series resistance) so that within this 'middle ground' between V_RWM and V_CL, the MCU's in-built diodes don't get overwhelmed?

One related question: is it a good idea or not to use SMD resistor networks/arrays for series current limiting in the input protection? I'm concerned that the fine pitch between lines (e.g. 0.8mm for a 1206-sized quad-resistor part) is a weak point for high transient voltages arcing? Better to use discrete resistors?

[thm_w]

Your assumptions are correct.

The high clamping voltage is due to the spec being at a high peak current. Series resistance would help drop this way down.

If arcing to another input is a major concern, then it would justify not using a resistor array. But typically you'd have say, 4 user switches connected to an array, so if it does happen to arc from one switch input to another, its not really a concern.

I would just choose a commodity part, assuming you don't have to meet a specific regulatory test. eg ESD9X3.3ST5G, CESD3V3AP, ESDA6V1W5 (1, 2, 4 diode) were some I randomly see on LCSC

[srb1954]

One related question: is it a good idea or not to use SMD resistor networks/arrays for series current limiting in the input protection? I'm concerned that the fine pitch between lines (e.g. 0.8mm for a 1206-sized quad-resistor part) is a weak point for high transient voltages arcing? Better to use discrete resistors?

The bigger problem with SMD resistor networks is likely to be the ability of the small resistor element to withstand the peak surge power of the transient without damage. This factor is generally going to be a problem long before arcing between adjacent contacts occurs.

In fact, bit of arcing can be helpful as it diverts some of the transient energy away from the resistor element, just so long as the arcing isn't to a sensitive unprotected circuit. Some designers incorporate spark gaps into the PCB design just for this purpose.

The criteria for selecting your input protection resistors should firstly be based on their ability to withstand the expected peak power. The second criteria would be the maximum voltage allowable across the resistor element. Thirdly, would be the isolation voltage between elements if you are using a resistor network.

After evaluating the criteria you will probably be surprised at the size of resistor required and you might have no option but to use individual resistors.

[T3sl4co1l]

Yup, that.

So, a few things remaining --

1. Capacitance.  At your bandwidths, this doesn't matter; it does however relate to die size, and overall cost... to the extent that those are related.  An SMAJ part might be more expensive than a little TVS array.  Or maybe not, you can find some pretty boutique parts that are way more expensive than their capability would seem to suggest!  Anyway, to whatever extent you're motivated by cost -- shop around, of course.  No need to lock yourself into something special, unless it offers characteristics that you absolutely need.  You may find something with much lower capacitance -- or much higher for that matter -- that turns out to be economical.

Which...

2. Peak voltage.  Yeah, it's tested at high current -- ESD delivers a peak current over 10A, for just some nanoseconds.  It's very destructive, for very small (semiconductor) structures that are vulnerable to those currents and energies.  Particularly important is the transient or incremental resistance (approximately Vpk / Ipk; technically, dV/dI is the exact meaning), and the series inductance which includes trace and via length -- your layout is as important as the choice of device!

Now, because the clamping voltage is still relatively high (10s of V), and the incremental resistance is so low (ohms, or fractional ohms even) -- consider the TVS, while clamping, as a Thevenin equivalent source of that voltage and resistance.  Clearly, if you were to wire it directly to your MCU or GPIO or whatever -- it's not going to be happy, all that available current will just plow through whatever measly diodes and internal resistance those pin structures have.

So, limit the current, by simply placing a series resistor between TVS and logic pin.  The goal is not to completely prevent ESD current into the device -- that's simply not necessary; the goal is to reduce current to safe levels, so the device can recover without damage (class C test criteria), or without power cycling (class B), or without any upset at all (class A).

What current levels are safe, isn't always so easy to find; most logic is good for 10 to 100mA.  You'll find 74HC family logic rated for around 100mA (above which, CMOS latchup occurs -- effectively the power supply gets shorted out, the affected chip gets very hot (maybe to destruction), and at least a power cycle is required for recovery).  Most MCUs are somewhere around 0mA (crap ones(!?), ADC inputs) or +/- 5 to 10mA (most GPIOs, e.g. STM32; may or may not include both directions on 5V-tolerant pins, RTFDS! ).

Note the series resistor introduces another lowpass filter, between the TVS (which as you note is already some capacitance) and the digital pin.  So this again limits bandwidth, and obviously it directly limits current output from the pin, if using it to drive loads instead -- you may need to prefer smaller value resistances for such situations, which requires a tight compromise between ESD performance and output capacity.

Alternatively, add a buffer with better ratings (like a 74HC bus interface or something), which might even be socketed so it can be expendable while protecting the MCU behind it.

There are many strategies to brain-storm about, for these sorts of things of course; but until you need the bandwidth, yeah, nothing crazy here.


TVSs -- there are two types at low voltages.  Well, kinda three.
1. Clamp diodes -- not a TVS at all, or at least not purely.  When the pulse is brief (like ESD or EFT, but not like lightning-induced surge), use the power supply to your advantage -- the amount of charge delivered is small relative to the total filtering on a typical supply, so diverting the pulse into it, can deal with it just fine.  Typically, you want at least 1uF total, with a modest fraction of it near the clamp diode itself.  BAT54S and BAV99 are typical choices, as well as clamp diode arrays (2 and 4 port), which often have a VCC-clamping TVS in them as well (so they can be used alone, unpowered, in many cases, or double as VCC clamping too!).

2. TVS diodes.  Some exist for 3.3V, but they suck.  The problem is, low voltage zeners are true zener diodes.  (And a generic TVS is just a beefy, well made zener diode.) Zener breakdown has a slow knee, below which leakage current is rather high, and above which, incremental resistance isn't very low.  So, under pulse conditions, a "3.3V" zener or TVS, might still clamp at the same 8 or 10V that a "5V" part does.

So, you mostly see TVSs down to 5V, where they are quite reasonable.

The sweet spot between zener and avalanche breakdown, is around 6V.  The knee is sharpest there.  So, parts with nominal ratings from say 5 to 9V are pretty good, having fairly low leakage combined with sharp clamping.  In higher ratings, the exponential response of avalanche takes over; it's still pretty sharp, giving a peak clamping voltage of say 30% above nominal rated voltage.

3. Snapback diodes.  These are a special multilayer device, sort of a hybrid between TVS diodes and BJTs or SCRs.  There's some semiconductor magic involved in their construction, but the result is a characteristic that can be very sharp -- indeed, the effect by itself has dominant negative resistance -- even up to very high currents.  By balancing that characteristic with internal device resistance, they can make a very stiff (low dynamic resistance) device, while still getting low leakage, and low nominal voltage ratings.  Example:
https://semtech.my.salesforce.com/sfc/p/#E0000000JelG/a/44000000MCod/Fs.eSBQs1U.304Nkc5rxRZ3Lk6_1A4.ZX_npjU1R.C8

Without a whole lot of reason to pick otherwise, I would guess the clamp diodes will be adequate for your purpose -- but there are other options available.

Tim

[HwAoRrDk]

Thanks all for the information.

The high clamping voltage is due to the spec being at a high peak current. Series resistance would help drop this way down.

Ah, I see that now I look at the datasheets more closely - e.g. 7V @ 1A, 20V @ 18A.

So, I take it that it's advisable to have some series resistance both in front of the ESD diode, as well as after it?

But typically you'd have say, 4 user switches connected to an array, so if it does happen to arc from one switch input to another, its not really a concern.

Good point, yes, I hadn't thought of that. It would be stupid to run some non-external signals through the same resistor network. But if all signal lines present are from the same origin, I suppose it doesn't really matter.

The criteria for selecting your input protection resistors should firstly be based on their ability to withstand the expected peak power. The second criteria would be the maximum voltage allowable across the resistor element. Thirdly, would be the isolation voltage between elements if you are using a resistor network.

Ah yes, I forget that for something like a 1206-size resistor network, their ratings are half that of an 0805 resistor (e.g. 63mW, 50V versus 125mW, 100V). Hmm, probably not the best choice then. I think perhaps discrete resistors will be best, and a smaller array component only if the layout can't be done without it.

Without a whole lot of reason to pick otherwise, I would guess the clamp diodes will be adequate for your purpose -- but there are other options available.

I think I might stick with TVS/ESD diodes, as I find sometimes clamping diode pairs a pain to route on a 2 layer board, as you have to run both power and ground to them (a lot easier on a 4-layer board - just slap some vias through to the power planes), whereas with a TVS that only requires a ground, it's no trouble if you have a ground pour everywhere anyway.

[T3sl4co1l]

So, I take it that it's advisable to have some series resistance both in front of the ESD diode, as well as after it?

Resistance in front doesn't do all that much -- HBM ESD for instance is 1.5kohm source resistance (give or take -- the RC model is a crude approximation of the equivalent inductance in the real system, or transmission line effects for that matter), so you need 1. a whole lot of resistance to do anything, that 2. needs to withstand a metric shit ton of voltage and 3. power and energy.

You can actually find chip resistors rated for this; the rub is they only approach that rating when embedded in epoxy or other dielectric potting (or oil or whatever).

For the same reason, shunt capacitance isn't all that useful, except for sufficiently large values -- again, in the ballpark of 1uF.  (The source is equivalent 100pF or so, so 10kV divides down to 10V into a 100nF cap.  The 1uF figure was for power supplies, where you'll want that even lower i.e. < 1V.)

Resistance out front might still be useful for other reasons -- if it's the best place to apply filtering, there you go; or, maybe it's useful to limit current from accidental sources (like cross-wiring inside an industrial control panel supplied with 24V -- your 5V input might need to withstand +/-24V for some duration).

You could even do a combination of things, to target high reliability in extreme conditions -- maybe a series resistor to handle momentary or sustained overvoltage, or maybe a depletion MOSFET to provide current limiting (to the same end, but dissipating less peak power than the resistor method).  That'll be sensitive to ESD and surge, so maybe you put something big and dumb outside that, a MOV or TVS -- and in the case of the MOV, the relatively high clamping voltage (typically 2-3 times the nominal rating) will be tolerable thanks to the voltage rating of the series element (DMOS are available in the hundreds of volts, or are there some over 1kV too, I forget?).  The clamping part means it's still subject to cross-wiring failure -- but you might have to cross-wire it to fuckin' 400VAC mains to actually get there!  You can make such a circuit withstand 120VAC cross-wiring, and high magnitudes of induced lightning surge, meanwhile passing the intended signal at acceptable bandwidth (some kHz; MHz would be pushing it), maybe like, an industrial control I/O at 4-20mA or 0-5 or 10 or 24V, etc.

Oh, I suppose some noise floor will be spent in the process (added resistance is added noise voltage/current), but getting a solid like 16 bits of dynamic range should be fine, still.

I think I might stick with TVS/ESD diodes, as I find sometimes clamping diode pairs a pain to route on a 2 layer board, as you have to run both power and ground to them (a lot easier on a 4-layer board - just slap some vias through to the power planes), whereas with a TVS that only requires a ground, it's no trouble if you have a ground pour everywhere anyway.

Good point, there's some economy of scale for them, on a 2-layer board -- if you have a bunch of them in one place, or a bus, it's worth routing.  One-offs, not so much.

Which, if you include zener arrays too (independent, or common-anode or whatever, also the clamp+TVS packs if the nonlinear capacitance is tolerable), the scale where it's worthwhile to do, could be fairly high, considering (half a dozen, say)... hmm!

Tim

[Siwastaja]

With 1uF cheaply available in a small package, though, that is a reasonable option. Provides a lot of filtration for HW debouncing as well.

Also useful with near-DC analog inputs. 1uF provides low impedance for the ADC.

Hence 1uF may be an appealing solution especially if you already have the 1uF cap in your BOM.

[harerod]

So, how about this simple circuit?
Assuming that the load on 3V3 is high enough to dissipate the ESD.
The layout simply follows the schematic.

[Siwastaja]

Remember, with schottkys to the rails, what is actually taking the ESD is the 3V3 power supply capacitor bank. It may be just a few 100n caps. In principle, this is no different to just having the capacitor at the IO pin, except that diodes isolate the capacitance from affecting fast AC signals as long as they are within rails. But if you don't need fast AC signals, or even better, want to filter them anyway, then just use the capacitor at the IO pin. It can be the same BOM reuse part you use for power decoupling. Smaller size is better, larger C is better. a 0603 or 0402 1uF would be a fairly obvious choice.