Low voltage rail on Cisco switch

[Otto W]

Hello,

I'm diagnosing a Cisco Catalyst 2960G ethernet switch that all of a sudden went lights out. Internal fan still spins and the PSU has been tested to work.

It appears that the various voltage rails do not come from the PSU board (which I somehow presumed), but instead, there's this step-down circuitry on the main board. The PSU seems to deliver only 12VDC to the main board. There are four other wires coming from the PSU, whose purpose I haven't yet quite figured out (orange 0V, yellow 3.1V, lavender 0V, brown 0V).



The rails have conveniently test points, and they measure as follows:

+12V (TP18) - OK
+1.5V (TP51) - OK
+2.5V (TP49) - Low, 0.8V
+1.2V (TP56) - OK

On the right side of the big Delta inductors there are Renesas R2J20601 MOSFET drivers. The one dedicated to the 2.5V rail doesn't really get warm, while the others get noticeably hot. I'm leaning towards this part being faulty because A) no short or other fault is pulling 2.5V down because the MOSFET driver isn't getting hot, and B) it can't get the 2.5V rail up with idle load. Since the Renesas part isn't exactly the easiest to find and replace, I'm hoping to hear comments on whether my theory makes sense. Is this a typical or even possible failure mode on a MOSFET driver? I'm not familiar with this type of rail configuration, and I'd be happy to hear if someone has alternative ideas on what could be the problem.

Thanks!

[TheMG]

Use oscilloscope to check for switching waveform at the PWM pin of the DrMOS chip. Also check to make sure the DISBL# pin is high (enabled).

It might not be the DrMOS, the SC2446 step-down controller and the components surrounding it could be the problem.

Also, don't rule out a short circuit or overload on the rail, if foldback or hiccup current limiting is employed, the converter could actually run cooler not hotter when grossly overloaded or shorted (oscilloscope will easily tell you if the controller is operating in such a foldback mode, as you would see short bursts of switching instead of continuous switching).

It appears that the various voltage rails do not come from the PSU board (which I somehow presumed), but instead, there's this step-down circuitry on the main board. The PSU seems to deliver only 12VDC to the main board.

As a side-note, this has been a very common arrangement in most modern devices. Using a single output power supply lowers cost by being able to use common or off the shelf power supplies with common output voltages. It's also less wiring between the power supply and mainboard, and also improves rail stability and efficiency by having the step-down converters closer to the circuitry they are powering, since a lot of these digital ICs require low voltages at relatively high currents.

[perieanuo]

hi,
yellow shouldn't be 3v3?
3v1 seems unusual voltage, at this equipment 0.2V may count dor something
[edit2]
but if your 2v5 reg is powered by 12V rail, regarding the fact control ic is a pain in the a:s:s to change due to his big gnd tab, try to change in and out caps (or exchange them with his cousins from adjacent rails)
the big inductor i doubt it's problematic

[Otto W]

Use oscilloscope to check for switching waveform at the PWM pin of the DrMOS chip. Also check to make sure the DISBL# pin is high (enabled).

Thanks for the tips. Looks like we may be up to something here...

DISBL# tries to get high, but something's not letting it:


PWM on the DrMOS follows the pattern:


While DISBL# is up the PWM signal looks correct.

It might not be the DrMOS, the SC2446 step-down controller and the components surrounding it could be the problem.

Also, don't rule out a short circuit or overload on the rail, if foldback or hiccup current limiting is employed, the converter could actually run cooler not hotter when grossly overloaded or shorted (oscilloscope will easily tell you if the controller is operating in such a foldback mode, as you would see short bursts of switching instead of continuous switching).

Is the aforementioned what a foldback mode would look like? Looks like some sort of boot loop / protective cutoff.

I measured resistance from the rail test points to ground, and they are as follows:

12V - GND = 151 Ω
1.5V - GND = 138 Ω
2.5V - GND = 312 Ω
1.2V - GND = 3 Ω

At least there's no dead shorts, but the 1.2V seems oddly low.

It appears that the various voltage rails do not come from the PSU board (which I somehow presumed), but instead, there's this step-down circuitry on the main board. The PSU seems to deliver only 12VDC to the main board.

As a side-note, this has been a very common arrangement in most modern devices. Using a single output power supply lowers cost by being able to use common or off the shelf power supplies with common output voltages. It's also less wiring between the power supply and mainboard, and also improves rail stability and efficiency by having the step-down converters closer to the circuitry they are powering, since a lot of these digital ICs require low voltages at relatively high currents.

You're probably right. I've mostly dealt with ATX PSU's. Although now that I think of it, motherboards have step-down and regulation right next to the CPU

[Otto W]

yellow shouldn't be 3v3?
3v1 seems unusual voltage, at this equipment 0.2V may count dor something

I honestly don't know what it should be. I've been trying to look for a pinout for the PSU (model: Delta DPSN-125CP), but no luck there. The PSU was tested in another switch, so there shouldn't be a problem on that board.

but if your 2v5 reg is powered by 12V rail, regarding the fact control ic is a pain in the a:s:s to change due to his big gnd tab, try to change in and out caps (or exchange them with his cousins from adjacent rails)
the big inductor i doubt it's problematic

The caps before the DrMOS measured in-place give a solid 4200µF for each rail. And the cap arsenal on the 2.5V rail after the inductor to ground shows 1300µF.

[TheMG]

What does the output voltage look like on the scope?

[Otto W]

What does the output voltage look like on the scope?

It's a rather steady 0.8V. I was expecting some sort of sawtooth shape after realizing the boot cycle, but there's none. Or then it's flattened out by the caps. I suppose I could put some small load on the rail and a sawtooth should show up, if my theory is correct.

[perieanuo]

so no luck with caps, replace/reflow renesas
load on 2v5 seems normal

[Otto W]

I had some time to further investigate, and here's what I found out.



So the DISBL# pin on the Renesas is directly connected to GDL1 (pin 24, gate drive output #1) on the SC2446. Probing around the other pins of the SC2446, I can see that COMP1 (pin 5, marked with red in the picture) does the following:



The adjacent SC2446's show a steady 2.6V and 3.0V on COMP1.

On behalf of COMP1 the datasheet says "The Error Amplifier Output for Step-down Controller 1. This pin is used for loop compensation." From the block diagram it looks like the only inputs for the Error Amplifier opamp (EA1) comes from the "Soft-Start And Overload Hiccup Control" module. Reference voltage (REF) is steady 0.5V, and AVCC for the IC measures OK (12V) as well.



I'm now looking at the inputs of the soft-start and OL hiccup control block, and what I don't quite understand is why I'm getting a different reading on SS1/EN1 with my DMM (Fluke 189) and my scope (Siglent SDS 1102CML). Scope shows 1.5V for SS1/EN1 on each SC2446, while a DMM reads 1.9V on the one that's acting up, and 4.2V on the other two.

[perieanuo]

good,
so for start cut connection for dsbl and force him to enable the rail

[Otto W]

If I hook 4.1V from a lab supply to DISBL# to force Renesas enabled, the PWM signal stays the same. There's no visible effect at the 2.5V test point and the rail stays at 0.8V.

The current sense pins on the on the SC2446 (CS1+/CS1-) measure 6.2mV, which (if I'm reading the above block diagram correctly) does not indicate overcurrent.

I also tried lifting the 2.5V rail manually by feeding 2.5V from a lab supply. My supply can go up to 2A, but the rail pulls the voltage down to 1.4V. To me this looks like something is shorted or drawing too much current from the rail. I'm starting to run out of ideas, so any help would be appreciated.

[TheMG]

Your bench power supply might not put out enough current. The low voltage rails on devices like that switch can actually need quite a few amps.

If you have a multi output bench supply, or more than one bench supply, you can put them in parallel to get some more current.

1.4V at 2A is not enough to indicate a problem, unless some chip is getting abnormally hot.

[skipzoid]

I've had a simular problem and after pocking about couldnt really find anything that lept out that showed a fault.

Then........ Whilst pocking about, the probe lead got caught under the SFP PCB at the front and as i pulled it free it disloded the board nearest the edge of the machine. I had a console lead plugged in and that has a TX/RX led on it and started blinking away - opened up term and there was text output and it was in the process of booting.

looking at the 2.5v rail and it was correct where previously it had been at the same 0.8v you had.

I've since taken both SFP boards out and cleaned the connectors and the switch is back in the rack doing its job. most odd.