Blew two ICs with seemingly benign clamping diodes

[timecrystal]

I was using an RC circuit to low pass filter the digital inputs to an IC. This was working fine but then I got the bright idea of clamping the output of the RC circuit to Vcc using a diode to avoid the capacitor discharging a destructive amount of current through the IC when VCC shuts off too quickly. Unfortunately this totally destroyed both the ICs I did this with and I have no idea for the life of me why or how. Even more puzzling, this destroyed the IC whose data sheet recommended doing this. See point 11.3 in this datasheet https://assets.nexperia.com/documents/data-sheet/74HC_HCT123.pdf

For reference the RC values are 10K and 0.01uF, the input voltages are TTL level so 0<=5. The diode I used was MMSD4148. Does anyone know why this would happen? I think it must be some IC voodoo magic because as far I as can tell the diode should have no effect (the reverse leakage current is in the nanoamp range).

[timecrystal]

For reference this is the circuit I used

[timecrystal]

Um, have to ask a seemingly stupid question, is your clamp 5V the same rail as the IC 5V? And is the diode connected in the wrong direction? If all is well, there should be no reason why it ruins the IC.

Yes the clamp rail is the same as the IC's supply rail. If the markings on my diode are correct (bar means cathode) then yes it was always connected in the right direction, that is to say the bar was attached to +5V.

Just like you I am stumped... There is nothing else in this sub-circuit, the inputs come directly from an IC and the output goes directly to another IC so it should be buffered on both ends. This is reproducible, at least I've ruined 4 ICs so far and saved the rest by desoldering the diode.

Edit: it's not the same diode causing these issues. 2 boards with four different diodes all caused the same destruction when put in this configuration.

[bdunham7]

Which IC is blowing up--is it the one for which this circuit is the input?

If you don't want to do any measuring, put the diode back on backwards and see if it still blows up.

[timecrystal]

Which IC is blowing up--is it the one for which this circuit is the input?

If you don't want to do any measuring, put the diode back on backwards and see if it still blows up.

Yes, the one receiving the signal.

The diodes were extra paranoia but they are not necessary since the capacitors are much smaller than the total capacitance filtering Vcc. My curiosity is high with this one so I'll have to break out the oscope to get to the bottom of this.

One thing I did not mention is that the input is a reset line so it's possible there is special circuitry in the IC to handle that.

[timecrystal]

Okay I measured the input pin on power-on and power-down. On power-on the input pin goes to +5V for about 20ms then goes back low (this is a reset pin and this is normal). On power-down, about 200us after +5V starts to decay, the input pin immediately goes high all the way to +5V then starts exponentially decaying as well to something that looks like 0.6V. About 20ms after the input pin starts to decay it seems to linearly decrease to 0v, taking about 538us.

The reason that the pin goes high on power down is that the input to the RC filter goes high on power-down (it's the output of a not gate and its input is the reset# line which goes low immediately on power-down). If the positive output of the NOT gate is essentially a rail to +5V, you would expect the clamped input pin and the +5V rail to be about the same.

I've attached the power-up and power-down scopes. The yellow trace is the measurement of the input pin, the blue trace is the measurement of the +5V rail.

All of this looks normal so still I'm wondering how this chip could have blown.

[Andy Watson]

What exactly was the RC filtering? Is it possible that the input could (very briefly) be undershooting 0V?

[timecrystal]

What exactly was the RC filtering? Is it possible that the input could (very briefly) be undershooting 0V?

300R and 10nF. I've measured the input multiple times over power-on and power-off cycles and it never undershot 0V. The input was a AHCT digital output, so it was basically a rail to either +5V and 0V, so there is little chance it could go under 0V.

After some more testing I'm starting to believe that I got extremely unlucky and the original ICs that I think blew were actually bad from the start... "Once you eliminate the impossible, whatever remains, no matter how improbable, must be the truth." as the Sherlock Holmes quote goes. To confirm I'll have to resolder in the diodes onto one of working boards and I'll to see if those ICs start malfunctioning... I'm hesitant to do so but I have to remove all uncertainty from this design... I will report back.

[timecrystal]

I will report back.

Wow so I soldered new diodes to one of my working boards and... the ICs do not malfunction. Who'd have guessed that?

I do not like this result but I cannot argue against it. The only explanation that fits all the evidence observed is that the ICs were bad to begin with, they just also happened to be the first two boards I soldered, they also just happened to be the only ICs with clamp diodes attached to their inputs. No I very much do not like this result but every other board works fine with the diodes soldered in.

The only other possible explanation that I can come up with is perhaps some static electric discharge happened with these boards when I first handled them. That's similarly unlikely but what other explanation is there?

[Manul]

Well, in my opinion you are pushing luck here. With such a filter you have around 40us (!) per volt rise time, while this IC does not have schmitt trigger on reset pin. Actually datasheet states maximum input transition rise and fall rate around 100ns / V. I'm not stating, that this is what kills the IC, but it could, because there is shoot through in CMOS chips between around 0.3 and 0.6 Vcc. Maybe shoot through in this IC is quite high.

[timecrystal]

Well, in my opinion you are pushing luck here. With such a filter you have around 40us (!) per volt rise time, while this IC does not have schmitt trigger on reset pin. Actually datasheet states maximum input transition rise and fall rate around 100ns / V. I'm not stating, that this is what kills the IC, but it could, because there is shoot through in CMOS chips between around 0.3 and 0.6 Vcc. Maybe shoot through in this IC is quite high.

Hmm can you expand? Actually the 300R/10nF RC filter is on the reset line of this IC https://www.nxp.com/docs/en/data-sheet/SC16C550B.pdf. I believe it does have a ST on the reset line. It's more likely this IC was actually bad.

On the 74HC123 clear pin, I have a 10K/10nF filter, and I don't believe that input is a schmitt trigger input. If the rise time is very slow I don't fully understand how that could damage the IC. Could you explain that a bit more? Admittedly I did not match specs on maximum transition rate. At the same time, even if it were the RC filter, in my testing it only ever failed when the diode was attached... or is that a red herring?

[Manul]

As you know CMOS logic contains P and N type FETs. They connected in different, sometimes complicated chains, but the idea is (in contrast to TTL chips) that CMOS do not normally use pullups / pulldowns, but instead the turn on of high or low side mosfet. They both have some gate threshold and in the region of around middle conduct both at the same time creating a short. This wastes power, create heat and in some cases might be even damaging. That is the whole reason for having rise/fall requirements, meaning that rise/fall can not be too slow. Very slow rise/fall extends the period of time in which both mosfets conduct and create a what is called a shoot through.

You exceeded the recomended rise/fall timing not by a little, but by around 400 times.

I really do not know if that can actually kill this IC, because I never tried, I just always observe and respect the timings. Anyway, it is for you to investigate. You may even try to deliberately make it even slower and see if it kills the IC consistently.

[timecrystal]

ery slow rise/fall extends the period of time in which both mosfets conduct and create a what is called a shoot through.

You exceeded the recomended rise/fall timing not by a little, but by around 400 times.

I really do not know if that can actually kill this IC, because I never tried, I just always observe and respect the timings. Anyway, it is for you to investigate. You may even try to deliberately make it even slower and see if it kills the IC consistently.

Wow thanks for this knowledge, this is definitely a design aspect I overlooked. This IC indeed has CMOS inputs for the clear pin. In fact I was able to bring both of these ICs "back to life" by just removing the capacitor. I think the slow rise of the clear pin when powered on was mucking up some internal state but without the capacitor it just works (The other IC types are indeed dead).

Just curious, I know the that the capacitor voltage is V = Vbegin + (Vend - Vbegin) * (1 - e ^ (-t/RC)) but how did you come up with your 40us estimation for the rise time? Would the time it takes to go from Vil to Vih be a good estimation?

[Berni]

Yep analog in between levels can indeed mess up chips big time.

But i don't think this would actually blow up chips. More of a problem is that this weird logic in between state can propagate trough multiple logic gates deeper into the logic of a chip and confuse it, making it behave in all sorts of weird ways possibly until being power cycled, all those logic sections shorting out and pulling the excess supply current. So the rise in supply current might not be just a single transistor but spread across many of them. The transistors are so tiny that they likely are not even capable conducting a destructive amount of current. Similar to IO pins on chips. You are not supposed to short a logic high output pin to ground, but pretty much all chips will simply just send the maximum output current out of the pin while getting warm, but no damage happens. This is because small FET transistors can saturate there channel at very low currents, once the channel is saturated the transistor effectively becomes a constant current source. So if this current is low enough there is so little power being dissipated that the heat can spread out onto the silicon die before the transistor actually overheats. You can buy MOSFETs that do the same sort of self current limiting thing, typically these are ones with a very high Rdson since that is a sign of a small narrow channel.

Since this is a common problem, chip manufacturers also came up with a solution. Those are schmitt trigger inputs on digital chips. Chips that advertise this feature in the datasheet have extra buffer circuits placed on the input pins. They implement a hysteresis comparator in this buffer that cleans up the analog waveform into a sharp digital signal by quickly snapping to a 1 or 0 state when a voltage reaches a certain threshold. These chips will not mess up no mater what weird analog signal you feed in. If the chip you need doesn't have this feature then you can just buy a hex buffer gate chip that features schmitt trigger inputs and place that before your sensitive input to clean it up. This also generally helps in input protection since a voltage surge on the input is likely to stop at the buffer chip to avoid blowing up a more important chip down the line.

[Manul]

Just curious, I know the that the capacitor voltage is V = Vbegin + (Vend - Vbegin) * (1 - e ^ (-t/RC)) but how did you come up with your 40us estimation for the rise time? Would the time it takes to go from Vil to Vih be a good estimation?

The actual rise time should be around 200us (if we think from 10% to 90%). That was rough calculation of 2x RC time constant. And 40us is how much time it takes for signal to rise 1V. Datasheet provides this metric, so I used it (time per 1V rise).

[perieanuo]

I was using an RC circuit to low pass filter the digital inputs to an IC. This was working fine but then I got the bright idea of clamping the output of the RC circuit to Vcc using a diode to avoid the capacitor discharging a destructive amount of current through the IC when VCC shuts off too quickly. Unfortunately this totally destroyed both the ICs I did this with and I have no idea for the life of me why or how. Even more puzzling, this destroyed the IC whose data sheet recommended doing this. See point 11.3 in this datasheet https://assets.nexperia.com/documents/data-sheet/74HC_HCT123.pdf

For reference the RC values are 10K and 0.01uF, the input voltages are TTL level so 0<=5. The diode I used was MMSD4148. Does anyone know why this would happen? I think it must be some IC voodoo magic because as far I as can tell the diode should have no effect (the reverse leakage current is in the nanoamp range).

lot of replies, i won't repeat other oppin,ions
if you clamp to +5V, clamp also to GND (with your clamp diode to Vcc you offered DC path for negative spike on capacitor)
that's nothing wrong with the diode you add, just add the second clamp
hope you get the schematic idea

[timecrystal]

Thanks to all for helping me debug. It turns out the diodes were red herrings. The real issue here was the slow rise time on clear pin of the 74HC123. The other ICs were actually damaged but empirically it was not from the diodes, I think it was another unrelated issue.

As for the diodes, well I've removed them from my design. The diodes I was using only offered 215mA of max clamp current, while the IC pins themselves offer 20mA. This doesn't seem like much of a difference, I would have to use diodes that can push 1A or 10A to make a significant difference but it's just overkill for this design. Instead I am using small capacitors, considering the relatively large filtering caps I have on the +5V rail. My device is attached to another device's power supply which is why I wanted the extra protection since I can't predict the total current consumption but according to my calculations the total consumption would need to be more than 20A to discharge one of my caps filtering Vcc fast enough to overcurrent the internal IC clamp diodes.

I should have considered the CMOS shoot through but I naively defaulted to assuming digital was actually digital and I've only ever used schmitt triggers for their noise filtering / hysteresis function. I legitimately learned something valuable here and for that I am grateful. Thank you to everyone who contributed, especially Manul!