Robust Microcontroller Input Protection

[JayDee]

Hi,
I'm re-visiting a project and want to design in robust input protection.
The inputs are mostly for slow digital switched inputs in an automotive environment. 
As ever..dont want to go crazy and over the top of parts (cost, board space and time of assembly) but dont see the point in making something that will be poorly protected.

The basic starting point is.
Microcontroller with a Capacitor to ground, series current limiter resistor and pull up resistor to 5V supply. 

I would like to clamp both constant and transient/ESD over voltages, thus looked at adding a TVS and/or Zener.
Do I need both? Looking at a 5V TVS specs ( for microcontroller inputs) they have a breakdown voltage (start of voltage clamp) at 5.6V and full voltage clamping at nearer 8V. This is obviously way over the Max voltage limit of the microcontroller inputs @ Vcc+0.3V  (so around 5.29V)

If I want to do this properly, do I need to run the TVS to deal with the high voltage short duration spikes/ESD, then use a Zener to bleed of any constant DC voltage over ~5.2V.
 

Any help appreciated.
Regards,
J.

[free_electron]

Search on the forum. I posted a whole page on this subject.
All you need is two resistors and a double-diode per pin connected to a hard reference made with a properly biased power zener.
Make sure to use long body resistors, create a moat under them, provide a shield ground that is only coupled with a ferrite bead to system ground and you will be good.
If you cpu does not have schmittrigger inputs you can add a schmittriger inbetween and feed that of as separate rail.
High energy , short pulses are best shunted using small caps to ground...

Someone needs to make an article about this stuff. This comes up ever so often and many people have no clue how to do this kind of basic thing..

If i have time i will whip one up... It's usefull since i'm preparing to write a book on practical PCB layout where this kind of stuff will be covered...

[TerminalJack505]

I think this is the post free_electron is referring to.

I'm doing an automotive circuit too so I'm looking at input protection circuits as well.  I wasn't going to do anything quite as extensive as free_electron's circuit, however. 

That circuit is meant for fast signals so you might be able to get by with something simpler.

[free_electron]

jep. that's the post allright. i'll whip up a sample layout later today so you can see more clearly how things are split.

[codeboy2k]

jep. that's the post allright. i'll whip up a sample layout later today so you can see more clearly how things are split.

I'd like to see the layout you do, but here's the circuit you drew up.  I did this to keep a reference for myself too.

I hope there's no mistakes.  Let me know if there are any changes to make, I'll update it.

Cheers!

EDIT: schematic diagram removed. See Version 2 here

[TerminalJack505]

jep. that's the post allright. i'll whip up a sample layout later today so you can see more clearly how things are split.

I'd like to see the layout you do, but here's the circuit you drew up.  I did this to keep a reference for myself too.

I hope there's no mistakes.  Let me know if there are any changes to make, I'll update it.

Cheers!

Nice! 

You might add a note about "the 1k resistor needs to be a fusible type for fire safety."  I assume he is referring to the 1k at the inverter's input.  Or did he mean the 10K resistor since it will be absorbing the bulk of the energy?

What kind of ESD voltages should this circuit be able to withstand?

[free_electron]

It's the 10k input resistor that needs to be long body. You don't want any flashovers there.

Note : this ONLY works if you have CMOS input schmittriggers. it will NOT work if you use TTL as you cannot make the input low enough to trip the circuit.
TTL chips source current from their inputs and all this series resistance prevents it from being pulled low.
-EDIT- I went back to my notes. the ground of the schmittrigger needs to be system ground. the supply of the schmittrigger is taken between ferrite bead and diode . i will adapt ascii schematics in a moment

The safety moats are imaginary lines. Think of it as the barbarians trying to raid the castle...

They have to hop over the body of the first resistor body ( which is a moat filled with water the barbarians have to get over ) 1206 - long body , i removed the step model so you could see clearly , then they hit the defense formed by the double diode trying to suck their energy to either chassis ground ( if they are negative) or the clamp voltage ( if they are positive ).

If they make it pass the second moat in the form of the ditch formed by the body of the second resistor . They will be all pooped-out by now and whatever 'oomph' they had is now a mere sliver that will be eaten by the clamp diodes in the schmittrigger chip.
The capacitors between CLAMP and SAFETY ground need to be able to handle a punch... it may be necessary to repeat that cap ever so often to shunt fast transients effieciently.
The same for the zener diode. if you have a long group of these inputs ( like a 32 input block ) repeat the cap and zener every 8 inputs and stagger them.
zener , 4 inputs , cap , 4 inputs , zener , 4 inputs , cap , 4 inputs etc ... If the zener or caps sits on the other side of the world , so to speak ( meaning far from where the strike occurs ) the inductance in the traces can still cause a large whop to appear.

As for how much this can withstand ... hard to say. depends on humidity and other factors. But it can pack a serious punch. we hade these in big telecoms racks. zero field failures... given that some of these racks get lightning strikes coming in ... ( ok , theres gas surge arresters along the way but still )

[Baliszoft]

As i always say, for slow digital signals, a cap for esd and a high series resistor do the trick. They dont really use more complicated way in high volume production. They shave off every possible cents.

[free_electron]

That depends. For consumer stuff yes.

For telecoms stuff ....  no.. there it needs to be bulletproof.


Here's a board i did years ago ( 2001 )
48 line ADSL 2+ line card. There is 48 houses connected to this board. each getting up to 24 Megabit /s second ADSL2+ a chassis holds 16 of these blades , 4 chassis per rack , that are housed in arrays of 8 racks side by side in a column... in buildings holding hundreds of these columns ...

Knowing that each blade costs about 400$ ... you don't mess around with "a resistor and a cap will be good enough..."
The first lightning strike closeby would fry the whole room and i would by on my ass in the gutter flat out broke and without a job and no prospect of ever finding one again ... i would be burned , literally and figuratively speaking...

I have never had to design consumer crap , and i refuse to. I know it is endless penny pinching and i can't work that way. If i make something it needs to as good as can be , be robust and work for 30 years +. If i make something for myself i still abide to those standards. If someone else doesn't like it , it is their prerogative to strip it down , but i take my hands off at that point. If it's worth doing, it's worth doing it right. In all other cases : buy it from wingpangpong..

[JayDee]

Excellent, thanks for the info guy, I'll have a look through in more detail tomorrow,
Thanks, J.

[TerminalJack505]

free_electron, good to see that this was used in an actual high-end product.

Was this circuit tested for compliance against IEC61000-4-2 (ESD) and IEC61000-4-5 (surge/lightning)? With the circuit powered-up and powered-down?

I'm just curious if it can meet the ESD requirements when powered-down since the rise time is so fast.

[codeboy2k]

You might add a note about "the 1k resistor needs to be a fusible type for fire safety."

ok, but I'll note the 10k that way. We wouldn't expect a lot of current there, just high voltage transient spikes, but  it would certainly help to protect against someone putting the mains onto the input pins. (more reasonably, a long run of the cable fraying in the plenum after a mouse eats away at the insulation and causes the input to be shunted to the mains cable beside it in the same plenum tray)

... I assume he is referring to the 1k at the inverter's input.  Or did he mean the 10K resistor since it will be absorbing the bulk of the energy?

it's the 10k, as FE has already pointed out,

It's the 10k input resistor that needs to be long body. You don't want any flashovers there.

but maybe, when money is no object and protection is highest priority, then perhaps even the 1k can be a long body resistor, for that little bit of extra protection.

Note : this ONLY works if you have CMOS input schmittriggers. it will NOT work if you use TTL as you cannot make the input low enough to trip the circuit. TTL chips source current from their inputs and all this series resistance prevents it from being pulled low.

I'll add a note to the schmitt trigger.

-EDIT- I went back to my notes. the ground of the schmittrigger needs to be system ground. the supply of the schmittrigger is taken between ferrite bead and diode . i will adapt ascii schematics in a moment

Yes, that makes sense actually, since the safety ground will carry large transients, you don't want those near the pins of the schmitt trigger, I'll update the drawing I made.  What about the bypass cap on the schmitt trigger power pin, you had it going to safety ground, I suppose it now goes to system ground too?  And the supply of the schmitt trigger doesn't need it's own ferrite bead anymore, since it's taken between the diode and the existing ferrite. It won't hurt, but I think it's not necessary anymore.

The capacitors between CLAMP and SAFETY ground need to be able to handle a punch... it may be necessary to repeat that cap ever so often to shunt fast transients effieciently.
The same for the zener diode. if you have a long group of these inputs ( like a 32 input block ) repeat the cap and zener every 8 inputs and stagger them.
zener , 4 inputs , cap , 4 inputs , zener , 4 inputs , cap , 4 inputs etc ... If the zener or caps sits on the other side of the world , so to speak ( meaning far from where the strike occurs ) the inductance in the traces can still cause a large whop to appear.

basically give the spike lots of places to dissipate quickly.  I'll note it on my diagram too for reference here.

EDIT:  updated schematic, v2 here now.

[SeanB]

FE, that did not help where I used to live. Lighning hit 40km away in the mountains, travelled along the overhead wiring ( blew them to vapour at the nodes along the way as well), hopped over the incoming gas dischargers and surge arrestors, landed in the frame and fried everything. Only survivor was the backup lead acid battery, the rest had to be cut out as it was literally welded into the frames. Everyy phobne rang once, then died. Took a week to replace the exchange, they literally went to a town with a similar unit being commissioned, but not cut in, and pulled it out and trucked the whole lot over to replace it. 1 week of 24/7 shifts by the technical staff and they were runing again. The donor town had to wait 4 months for the new one to be built again, so they still had service and the operators had a job for 4 months.

[free_electron]

ah.. overhead wiring. yes , that is a direct strike .. hard to stop that one.
i'm talking underground trunks. lightning heats ground about 2 meters above where the cable sits. this still induces a massive jolt in the cable.

All i know is this stuff was tested according to CISPR and ITU. and it lives. We shot it many times with esd guns at 15 kilovolts. no problemo. ( i don't rememeber what the energy setting was though. 15 kilovolts doesn't say a lot. 15 kiovolts from 15 pf is different than 15 kilovolts from 1 uf   .. ) By the way , this is not my design. i learned this from the telecom guru's at alcatel. they had a dedicated lab for this kind of stuff. They explained the moatign system , how to make sure there is enough crep , how to open up the soldermask for the esd guard ring ( drat , i forgot to do that in my quickly-slapped-together-layout : there should be NO soldermask around the testpins , nor underneath the first resistor. you need bare pcb. the reason is high-voltage creep.

long body resistor : 1206 is fine. that gives you enough creep.
Try to use thin film and not thick films. thin film is metal sputered and laser cut. thick film is printed ink and laser cut. thick film flashes over quicker than thin film.
there are 1206 resistors with high standoff voltages. they are typically blue in color ( as opposed ot a black strip , they have a blue strip on the alox base ) vishay and others make those. the blue material is an extra isolation layer to shield the resistive element. it is applied AFTER the laser trimming to seal the element from humidity.


you can omit the schmittrigger if you know the processor or whatever digital chip you use has these in the input stage. i just added those for completeness. if you omit the extra chip then the processor or chip is powere BEFORE the ferrite ( and not from between ferrite and resistor to the diode )

the ferrites are high-impedances for the pulses. there too you want long-bodies.
any connector shield needs to be plugged to the safety ground as well.
like a usb connector. it has 4 wings connected to the metal body and 4 pins : vbus d+ d- and gnd. the wings go to safety ground , the gnd pin goes to system ground ! try to route this connection (gnd pin) on in an inner layer. and create another bare copper ring around the pins.

the whole idea of creating bare copper is so the esd zap is attracted there and doesn't shoot into system ground


Once you are past the 10k moat there isn't much energy anymore. then you cross the 1k moat and the esd protection devices will eat the rest.

it is important the capacitors at the double diode can handle large voltages. you can get away with 10nf to 100nf but grab 250 volts . whatever spike goes rthrough the diode needs to be absorbed by these ( the zener will react , but the very narrow spikes will be shunted mainly by the capacitors. )  thos caps will probably be 1206 as well. in telecoms we used 2512 bodies ans some other weird number 32-something ( can't remeber , i have to look it up. we used wide electrode caps as well.
just like you have a 1206 capacitor ( 12 long 6 wide ) there is an 0612 cap ( 6 long , 12 wide , so the electrodes are wider but the distance between is shorter. this is for pulse currents.)

there is a lot to say about selecting the correct kind of parts .. most egnineers and hobbyists pick whatever they find in the drawer.. there is a reason for having all these different kinds of constructions , materials , body shapes , laser cutting differences ( l -cut , r-cut , modifed r-cut , meander-cut and plenty others )

someone ought to write a book about selecting passive components for function... you can easily fill a 500 page book on resistors and caps alone ..

[notsob]

I once viewed a microwave repeater station that had a direct lightning hit. The copper 'lightning' bars inside the room had been vapourised and had coated everything inside like a spray paint job

[free_electron]

nothing will stop a direct hit. that is for sure.
but an induced hit can be stopped.

[TerminalJack505]

Thanks for the circuit, free_electron.  I might base my automotive input protection circuit on it.

So far, the attached schematic is what I've come up with.  The MLV is $1.50 in one-off quantities so doing something like you've done might be cheaper.  The nice thing about your circuit is that it easily handles the one condition that is the cause of the most expense in my circuit--the load dump condition.

So far as my circuit goes, the MLV is going to clip or shunt the majority of transient conditions and absorb most of the energy.  The 4k7 resistor, the zener and the 100n cap should catch the rest.  The zener and the 4k7 will clamp negative voltages (negative transients left after the MLV and reverse battery condition.)  The 5V zener will clip to the MCU's input voltage with the 4k7 limiting current into the zener.  The 4k7 and the cap will also act as a lowpass filter.  (My signals are very low speed.  Basically DC.)

The 10k is a pull-down.  My inputs will either be +12V or hi-z.  Finally, the 1k will limit current through the MCU's protection diodes when the input is negative.

[free_electron]

keep in mind that with  the 10k pull down and the 4k7 input resistor you can never make more than 2/3 of the input voltge ..

if you need a pull down or pull up it should be pace BEFORE the 4k7 ( where the transil sits )

apart from that your circuitry is also a good approach. use a 4v7 or 4v3 zener and not a 5v1 .. you don;t wan tot trip the top diode in the mcu. i see a lot of designs where they abuse the esd diodes as clip/clamp diodes. that is NOT their intended purpose unless they are specifically designed for that ( ATMEL for instance does that , they even have an appnote about it , but don't expect that from all parts ! 99% are note designed or spec'd, or have never been tested for that ! )

Keep in mind that zeners are slow to turnon. ( that's why i bias them in my circuit. they are already conducting weakly ) i only need 1 zener that can be shared amongst inputs. i can have one big fat zener protect 4 or 8 inputs. saves board space and cost.

There's different ways to skin this cat. the key to protection is to control where the current flows. at all cost keep it OUT of your actual system. shoot it in the shield groudn , burn it in the transil , whatever .. but do not send it into the input pin of the chip , nor shoot it in the power or ground pin of the chip.  that is the only rule that needs abiding. if you keep in mind that you are dealing with fast transients : give them a low impedance path. they contain high frequency components so they don't like coils , but do like small caps. make ample room and creep area and you are golden.

for esd : bare the copper ring so it has something to zap to.

[free_electron]

one thing that maybe needs some more clarification

the circuitry, principles, layout philisophy  i presented here does several things.
- ESD protection
- absorb zaps and pulsed overloads wihtout destruction
- avoid brown-outs or system resets to load dumps , spikes or whatever

the circuitry will survive this undamaged.

 in case of a hard overvoltage somethgin will give ... let's say a rodent is attracted to a bunch of yummy isolation , makes a feast of it and ends up frying itself and in the process creating a hard short between power rail and the input. 220 volts or 380 volts at the pin.
something will give ! and it needs to be the first resisotr ( that 10k)

even at 1 kilovolt the current will only be 100mA. (1 kilovolt / 10k) the zener or diode has no problem holding that. even if 10 inputs were shorted to 1 kilovolt. it's a friggin 1 or 3 watt zener ins SMA body. it won't even heat up fast enough since something wil give...

The power dissipation in the 10k will be remendously high. 100 watts. for a 1206 .. that's going to fry .. almost instantaneous, so that is why you need safety resistors there. (sometimes called fusable resistors) these will open , will not catch fire , will not arc and they will hold their rated voltage when fused open.

just want ed to make that aspect clear... as to the why how and what of having safety resistors there. of course if you don't expect anything over 50 volts continuous ... for 50 volts you dissiapte 250mW in the 1206. no problem. it won't fry. if you want to hold 100 volts: use 47k ...
P = IxIxr .. doubling the voltage is quadrupling resistance.. don't get caught by that one !

[TerminalJack505]

I appreciate all the advice, free_electron.  If I use my circuit then I'll make the changes you suggested.

I'll look into using yours as well.  I shouldn't need the Schmitt trigger and I'm not going to have an earth ground, obviously.  Other than that it should be usable pretty much as-is.

[Baliszoft]

For telecoms stuff ....  no.. there it needs to be bulletproof.

Thats a different story i admit. I thought it was just a plain (and local) input signal.

[codeboy2k]

There's lots of good safety porn in this thread

[TerminalJack505]

Any thoughts on using the attached circuit as automotive input protection? 

It uses a cheaper TransGuard MLV and now has a 8k2 resistor to help absorb a big chunk of the energy.

[free_electron]

get rid of the 8k2 and change the 1k to a 10k.

i don;lt know about putting series resisors before transil type devices... it may make these slow to react...
the idea if these protection devices is that they form a thyristor that ignites when a certain voltage thereshold is crossed. the problem with a thryistor is that you need to send CURRENT into the gate to kick it into action. so if you put a series resistor there the voltage to trigger will be reaached but that resistor may limit the current enough that the thyristor doesn't fire properly ...

the other thing is that those devices are designed to go in hard short. when the pulse decays the current through them drops and once the current is low enough they come out of conduction.
so there you have anothe rproblem. the series resistance may limit the current enough that the transil never stays in conduction do it doesnt really drain the energy.. it just sits there ding nothing at all , you might as well remove it..

just my quick thoughts.. i'd need to try this on a curve tracer and see how those things react under current limited conditions.

[TerminalJack505]

So like the attached circuit then?

Do you think I could get away with using a resistor up front like that if I use TVS diodes? 

I'm really just trying to take some stress off of the part for load dump situations where the voltage can spike to 100V for 400ms.  The MLV (even the 0.3J part) according to the datasheet can handle automotive load dumps so I should be okay with the circuit as attached.