Don't blow your Teensy 4.1 Ethernet PHY when using passive PoE

[mamdos]

Hot-plugging a passive PoE connection will blow your teensy PHY. Even though the PHY is DC-isolated from the jack, the unbalanced condition that occurs while hot plugging a powered cable will cause a transient spike of the order of your PoE voltage on your PHY. If you actually make this mistake, only the PHY is damaged and you can fix your teensy by replacing it (PN:dp83825irmqr). Robin very generously sent me some spares they had around and I was able to bring my dead teensies back to life. Thanks Robin!!!

The solution to this is easy: TSV diodes on the ethernet data lines on the PHY side + small series resistors. Implementing this on the next teensy revision or the ethernet adapter board or on your custom PCB will protect your PHY from hotplugging passive PoE.

This application note explains all this in detail. http://ww1.microchip.com/downloads/e.../00002157B.pdf

[bicycleguy]

Thanks for finding this, very interesting. I use a Netgear GS305P POE that I'm always connecting and disconnecting test equipment, rPi and computers to, as if it was a regular ethernet connection. While the simplistic visual aids in the instructions show plugging in the power last no specific mention is made of the necessity. So far no damage to my stuff, but I will now take precautions, thanks.

Looking at schematics of the Raspberry Pi 3+ and 4 that were designed for POE, I see no TSV. In fact looking at 4 devices so far I see no protection, unless it is built into the magJack or hidden somewhere.

[mamdos]

You probably wont have this problem with active PoE because the power isn't enabled until negotiation, so hotplug events wont happen the same way they would with passive PoE.

[bicycleguy]

@mamdos,
Thanks, wondered what passive meant.

[PaulStoffregen]

@mamdos - Can you tell me which hardware you're using for passive PoE? If I wanted to experiment with various diodes and resistors, which gear should I buy to give me the PHY-destroying scenario you had?

[mamdos]

Hey, just saw you responded. I came back to report that the adding a parallel TSV diode (P/N: UCLAMP3301D.TCT with no resistor) as described in the article didn't work for me. I blew another PHY after hotplugging with a load. I'm going to hook thinks up to a scope to get more measurements.

My setup: I've designed a custom baseboard for the Teensy 4.1. I'm using the ETH1–460 PoE magnetics, and FDMQ8205A for the rectifying bridge (because "PoE" is what happens when you fail at making a standard ). I've attached the schematic below. This schematic is for the "fixed" board and it contains series resistors in addition to the TSVs. For my tests, I've only been patching the first revision of the board by soldering just the TSV on the teensy ethernet header.
For the PoE injectors, I'm using these https://www.tycononline.com/GigE-4Pa...45V_p_361.html and a 54V power supply. The load is mostly capacitive. I'll post results once I figure out a solution.

[mamdos]

Some updates. I took the teensy out and probed the data lines on one of the pairs. I used a 6" and a 100ft CAT 5 to see the effect of input LC and I tested with and without the UCLAMP diodes. Here are the results. As you can see the TSV diodes make a huge difference, but not enough to protect the teensy. Without the diodes, the transient spikes at the PHY are in the order of 70V!!! the diodes help bring those down to 9V. Some series resistance should hopefully provide some protection. Thats up next.

[Spyy]

Hi,

any news concerning this topic...i am using a PoE solution from Silvertel together with my T4.1...

Thank you

Torsten

[mamdos]

I've been pulled into other project atm but please keep in mind that these events only occur if you use passive PoE AND you are hotplugging cables. If you power up/down from your supply and/or use a power negotiating PoE you should be fine. Hope thats helpful

[mamdos]

Quick update: I've added 33ohm series resistors to each data line on the phy side as recommended in the application note. Unfortunately, I see 7V spikes on the Phy, which is above the maximum rating of the chip. On the bright side, I haven't been able to get the chip to fail after repeated hotplugs, but I obviously cant comment on the long-term effects of asserting this form of stress. At this point, I'm moving on. I'll update this thread if anything fails.

[PaulStoffregen]

Do the 33 ohm resistors impact use with long cables? Wouldn't that be effectively driving a 100 ohm twisted pair from a 166 ohm impedance? (assuming the PHY chip really is doing some analog magic to get close to 100 ohms)

I'm debating whether to update the PCB and we ship with the ethernet kit, maybe even add more parts. Looks like 2 of those diodes could fit on the bottom side of the PCB. Not sure how I feel about series resistors. Also not sure if increasing the kit cost makes sense?

[Spyy]

I've been pulled into other project atm but please keep in mind that these events only occur if you use passive PoE AND you are hotplugging cables. If you power up/down from your supply and/or use a power negotiating PoE you should be fine. Hope thats helpful

Hi,

other project..sounds good;-) no problem...you are right that you have to have two things together to happen....but that you hotplug a PoE device is quite common i think...this is the big advantage of this tech....

Thank you very much for your effords an for doing this research...

@Paul: Please let us know here about progress about this topic...thank you

Torsten

[TeensyWolf]

As the 4.1 PHY is a TI part, these might be better than the microchip docs?

https://www.ti.com/lit/ug/tidues1a/tidues1a.pdf

Here is an older doc:
https://www.ti.com/lit/an/slva233a/slva233a.pdf

I don't like the idea of adding 33 ohm series resistance to the lines, TI suggests a different configuration, with caps, ESDS312s and some resistors (in the Bob Smith). There are also some LC03-3.3BTG between the RJ45 and the magnetic, but how to connect when using a magjack?

If anyone has 2 x DP83825I chips they can spare, let me know. I didn't blow it on POE, but soldering bridge on 6 pin.

[MarkT]

You can reduce the value of the resistors if the TVS diode has a low enough effective resistance - a divider is formed.
So chunkier TVS will allow smaller resistors - assuming I understand the app note (equations 1 and 2).

[Nominal Animal]

I haven't seen any magjacks that'd expose the RJ45 side pins for TVS use (LC03-3.3BTG).

If you look at say Abracon ARJM11 104 model RJ45 with magnetics – noting that compared to TI tidues1a appnote, the schematic needs to be horizontally mirrored –, it exposes the cable-side center taps, but the PHY side center taps are tied to pin 5 of the connector (meaning the 100nF capacitors C58-C60 cannot be used with this one).

I checked a few RJ45 at LCSC with integrated magnetics, and none are compatible with PoE. On the other hand, suitable external magnetics (like Hanrun HR641680E) seem to be cheap and easily available; there are also at least couple of suitable, shielded RJ45 connectors without magnetics. Interestingly, there are also several isolated power modules capable of providing 5V from 37-57VDC, the cheapest being TDPOWER TD6-48S05W which is about 82% efficient at this voltage range (and can handle continuous 18-72V). So, in addition to the two full bridge rectifiers and capacitors, this might yield a way to implement PoE for Teensies by just replacing the Ethernet module.

Me no EE, though, and know nothing. If I did, I might try to draw a schematic for review at EasyEda.

[TeensyWolf]

You can reduce the value of the resistors if the TVS diode has a low enough effective resistance - a divider is formed.
So chunkier TVS will allow smaller resistors - assuming I understand the app note (equations 1 and 2).

So taking the 2 equations for this TVS: https://www.ti.com/product/ESDS312

Eq1 Vclamp=5V@1A & Vw = 3.3V, Rs = 1.7ohms

Eq2 Vmax=57V Rp> 12.98ohms, so a 13ohm is suggested here. This feels better than 33ohms, but Microchip appnote also says 0-5ohms would not affect ethernet data.

[HackNFly]

I'm relatively new to this. I have a teensy 4.1 and pjrc ethernet kit I'm using for a laboratory project. Under what conditions would the chip be damaged? If I use an external passive poe splitter before the pjrc ethernet kit would I be OK? Or do I still have to add some circuitry to be safe?