Unidirectional TVS on I2C prevents it from working - Bi-directional one is ok

[T3sl4co1l]

Without schematic, obviously we don't know, but you can compare against these ideas and check or test whether it's relevant to your case.
https://electronics.stackexchange.com/questions/713381/correct-placement-of-series-ferrite-beads-to-avoid-dc-disconnect-during-power-cy/713473#713473

The solution for avoiding that tends to be the use of EMI ferrites to burn off the pulse energy into heat, and avoidance of using high Q capacitors on the input (like most MLCC capacitors)

Similar can be used around regulators to limit how sharp of a current the regulator is exposed to. Otherwise softstarting a regulator is also a good idea

Ferrites tend not to do much in power supplies, because of the low DCR (if it were higher, it could be used more easily as series damping) and saturation current (not much flux is absorbed).  Lossy caps are the most common go-to, especially as electrolytic or tantalum where the ESR comes for free.

Tim

[Njk]

I doubt it's necessary to select the part with documented standoff voltage equal to the max. working voltage (3.3 V or something). What's the point for that? Whatever the standoff voltage is, the breakdown voltage is always higher. And the clamping voltage is even more high. Moreover, the clamping voltage depends on current and that voltage typically far exceeds the absolute max. DC voltage for the IC you're going to protect anyway. The key assumption is that the overvoltage event is very brief. I doubt adding or subtracting several volts to the clamping voltage parameter of the part will make a difference as the current and duration are not exactly known. The IEC charts provide estimated values for modeled spikes but who knows what real spike will be encountered by the device. All that ESD protection stuff seems murky.

[Berni]

Ferrites tend not to do much in power supplies, because of the low DCR (if it were higher, it could be used more easily as series damping) and saturation current (not much flux is absorbed).  Lossy caps are the most common go-to, especially as electrolytic or tantalum where the ESR comes for free.

Tim

Yes the ferrites having near 0 DC resistance is exactly why they are good for this. They won't burn any significant amount of power due to resistance, but up above a MHz they have at least a few Ohms. Enough to eat up the energy trying to ring itself around a high Q capacitor. It can also be just enough AC impedance to hide away large low ESR capacitance from regulation loops to prevent them from going unstable. Using regular inductors as power filters is no good as they return most of the AC energy back. While increasing the ESR of capacitors could result in slightly worse ripple.

Tho admittedly it is not the primary reason i like to use lots of ferrites around power rails(but is a nice bonus). More of a reason is to keep the noise of switching regulators from escaping out, or to enhance the ripple rejection of linear regulators up into the MHz where they can't really do much. Other nice bonuses are also getting more control of the power rails return path in the ground plane(especially for switchmode converters), or as debug switches that can isolate a part of the circuit away from the rest to help find a short circuit fault or easily measure the current consumption.(They also light up nicely on a thermal camera if there is a short)

Might not be a perfect catch all solution for inrush spikes, but it does help in mitigating it while providing other benefits in the same cheap small component.

[thm_w]

I doubt it's necessary to select the part with documented standoff voltage equal to the max. working voltage (3.3 V or something). What's the point for that? Whatever the standoff voltage is, the breakdown voltage is always higher. And the clamping voltage is even more high. Moreover, the clamping voltage depends on current and that voltage typically far exceeds the absolute max. DC voltage for the IC you're going to protect anyway. The key assumption is that the overvoltage event is very brief. I doubt adding or subtracting several volts to the clamping voltage parameter of the part will make a difference as the current and duration are not exactly known. The IEC charts provide estimated values for modeled spikes but who knows what real spike will be encountered by the device. All that ESD protection stuff seems murky.

Not sure what you are trying to say here. Data sheets are linked in the OP: both diodes have a clamping voltage well above 3.3V tolerance, they are designed to be used on 3.3V systems (hence the 3v3 in the part number).
The original diode was 5.8V new was 4.2V minimum clamp.

The issue is the new diode has a snapback voltage of 2.6V or even less, as teslacoil points out, its not well documented.
This type of ESD diode is probably more appropriate for an ultra sensitive component where you want the voltage to rise as little as possible? They say in the datasheet for "USB2.0, HDMI, DisplayPort, eSATA and LVDS".

I know people complained recently about EMI shutting off their monitor signal, maybe a similar device was the cause?

[T3sl4co1l]

Yes the ferrites having near 0 DC resistance is exactly why they are good for this. They won't burn any significant amount of power due to resistance, but up above a MHz they have at least a few Ohms.

Maybe... when, is the question.  Saturation can be as little as 20mA, for small (0603-) high-impedance chips.  It can be 200mA or more for larger (1210+) chips, and some low-impedance parts (<30 ohm?) are actually made for filtering and saturate at high currents (8A+).

Since inrush can be 10s of A, it's very easy to saturate a bead, and then only a tiny delta of energy is removed per cycle; the ringdown envelope looks like a triangle rather than an exponential.

Tim

[Berni]

Maybe... when, is the question.  Saturation can be as little as 20mA, for small (0603-) high-impedance chips.  It can be 200mA or more for larger (1210+) chips, and some low-impedance parts (<30 ohm?) are actually made for filtering and saturate at high currents (8A+).

Since inrush can be 10s of A, it's very easy to saturate a bead, and then only a tiny delta of energy is removed per cycle; the ringdown envelope looks like a triangle rather than an exponential.

Tim

Yep that's why i said it is not a perfect solution for inrush spikes.

Ferrites designed for power filtering are designed to operate a sizable amount of DC current, hence why they have amp ratings. Even after they do reach saturation they will be dissipating around the ringing zero crossings, then even in the saturation they still have some DC resistance that becomes significant at large currents due to I^2*R (the non low ESR caps are still well under 1 Ohm too). Tho if you are actually ringing with multiple up and down swings then you need more damping anyway. You want to suck up enough energy on the way up so that you don't even get a significant overshoot, let alone full on exponential decay ringing.

Ferrites are also not a solution if you have like 10000uF of capacitance and expect it to solve your inrush problem(the ferrite for that would be enormous). Just that with a sensible amount of input capacitance (like you might have with MLCCs before a switching regulator) a ferrite will likely remedy the problem enough for it to not be a significant problem. You might want a ferrite there for filtering and EMI reasons, but even if the ferrite is not great at damping inrush, it might still be enough damping to fix the overshoot, so you can have 1 component do multiple jobs EMI and inrush.

In a lot of cases just putting a 0.1 Ohm resistor in series with the input is already enough to fix it, but those burn more power at DC, so they tend to put the resistor in series with the capacitors.

There is a similar problem with SD card sockets as some SD cards have a significant amount of power supply capacitance inside the card. So if you apply permanent 3.3V power you could have the card monetarily drag down your supply rail enough for your MCU to crash. I found that a ferrite in series with the power helps nicely with that too. There inrush limiting is admittedly a primary job of the ferrite, but it works and i likely already have one in the BOM somewhere as a cheap component. So i just reuse a existing part and avoid increasing the BOM line count.

[Njk]

Not sure what you are trying to say here. Data sheets are linked in the OP: both diodes have a clamping voltage well above 3.3V tolerance, they are designed to be used on 3.3V systems (hence the 3v3 in the part number).

I think the problem was created artificially, by choosing the part type too optimistically. It is not stated in the NXP data sheet for PESD3V3X2UT that the parts "are designed to be used on 3.3V systems", this is your interpretation. The vendor is not aware of the definition for your particular "3.3V system". So the data sheet merely states that the reverse stand-off voltage (V_RWM) is of 3.3 V max, at 25 degrees C. As no further information about the test method for this parameter is provided, we can only assume that the protection will not be triggered and will not start leaking at 3.3 VDC. Seems fine but what will happen at a little bit higher voltage, say 3.4 or 3.5 V or at different temperatures? The document does not specify that. It's also reasonable to assume that every mfg. and measurement processes are associated with the tolerance so the actual parameter value may vary within the tolerance window. That's the reason for putting some engineering margins in the design.

The other consideration. In the past, the margins were typically provided by every reputable vendor. For instance, if the part is rated for 10 A, it actually could withstand 11,2 A. So the integrators were provided with 12% safety margin for free. Not because the vendors were less greedy back then, but because it was the only method to guarantee that every part meets the rating. As the mfg. process matures, that goal can be achieved by technology and now the new part works fine at 10 A and blows up at 10.001 A. I think it's reasonable to expect something like that in the near future. You can't force the vendors to continue with the traditional approach without resorting to a nonmarket mechanism such as regulations. Perhaps this case is a good example. I'm not sure a parameter like (V_RWM) is a subject for any regulation. Actually, the responsibility for margins provision is shifting from the vendors to the integrators. And worse, that creates a temptation for vendors to start manipulating a little with the numbers and the definitions, for the sake of marketing.

For one reason or another, there is no perfection in the world. Just don't use the part at 100% of its rating, leave some margin to improve the design reliability.

In that particular case, why do not choose a part with (V_RWM) one step higher, e.g. 5 V? That will provide reasonable assurance that no shit will happen at 3.3V, and also will take care about all the usual suspects like the PSU output voltage tolerance, ringing on the bus, etc.

[thm_w]

Not sure what you are trying to say here. Data sheets are linked in the OP: both diodes have a clamping voltage well above 3.3V tolerance, they are designed to be used on 3.3V systems (hence the 3v3 in the part number).

I think the problem was created artificially, by choosing the part type too optimistically. It is not stated in the NXP data sheet for PESD3V3X2UT that the parts "are designed to be used on 3.3V systems", this is your interpretation.

I don't disagree with your other points but NXP straight up recommend the part for USB2/HDMI which are 3.3V systems. As noted above.

[Njk]

OK, not for an I2C system that was mentioned by OP

Edit: Just looked into HDMI v 1.4 specification. Table 4-22 Required Operating Conditions for HDMI Interface:
Termination Supply Voltage, AVcc: 3.3 Volts +/- 5 %

No wonder all the recommendations are mentioned in the informative part of the data sheet, not in the normative one