Ugly Tezro power

[Elf]

Curious, I heard reports of similar voltage regulator issues from an O300 or O350 owner as well. Maybe these are just getting to that age!

With regards to mine, I have replacement parts for that whole block, from the LT switching controller, to the switching MOSFETs, to the capacitors. I suspect more than just a bad capacitor given how one of the MOSFETs seems to be stuck and overheating, so I am hoping that just replacing it all will do the trick.

I ordered the parts some time ago, but with everything going on I have not had the chance to pull the board and put it under the knife (so to speak). It is a big board that I am sure will be annoying to pull, so I am not very keen on it, but it is definitely on the to-do list since I am reluctant to use the machine until the issue is resolved.

 

[weblacky]

Well, I’m pretty new at PCB level semiconductor repair. A part of me understands what your saying, another part of me proposes a different reason for your observations. So I’m kind of in limbo.

I know yours is not happening consistently, but these MOSFETs are being directly controlled. My brain tells me you should check the signal at the gate of the stuck mosfet before blaming it. Perhaps your controller is commanding it on nearly all the time to compensate for something. I mean, from what I know, more ON time means higher voltage, under normal circumstances. If your controller was trying to raise voltage by the maximum amount of gate signal increase, it would likely look like this, right?

so because the MOSFET hasn’t shorted, I think it might be working fine and it’s being told to do this?

I've not heard of a sticking semiconductor that didn’t exploded shortly afterwards and could reset, so I’m having trouble accepting a semiconductor failure. I would assume the controller is overcompensating (aren’t we all) for feedback voltage being low and cannot run enough to make voltage rise.

But the only way to know for sure would be to scope out the signal and a couple points and see, without equipment to evaluate the cap insitu (which I hope to have in the near future) we wouldn’t know without a minimum of cap replacement (using hot tweezers) and see.

Also those caps are extremely high quality so I’m even more disappointed, I looked them up. Panasonic SVP SMD, 2000 hr service life, low ESR, high ripple tolerable. They still make them!

 

[Elf]

Ah, I did not mean "stuck on" in the sense that the MOSFET itself was shorted, but "stuck on" in the sense that it may be turned on at an inappropriate time by the controller.

If you look at the photo, there are two MOSFETs (FDD6670A, FDD6612A) that appear to be in a push-pull configuration, with the source of one connected to the drain of the other, and that source/drain connection also leads to the inductor.

This actually lines up pretty well with what is provided as the "typical application" in the LTC3728 datasheet, where you can note the push pull arrangement a bit more easily:

If the cycles of TG2 and BG2 gate drives overlap, for example, then this could cause the MOSFETs to overheat. Oddly in my thermal image above, it was only the lower side MOSFET which was significantly overheating. In that case especially, it is unlikely that trying to prop up too low a voltage (too high a PWM duty cycle on the switching output) would be the cause of overheating of that low side MOSFET. The voltage rail itself showed very little ripple, though I do not have a scope trace of it handy.

Unfortunately the events were so intermittent that I gave up trying to catch one with any more than a multimeter left on. Per Murphy's law, the issue almost never happened when I had the whole unit hooked up on the bench.

Edit: I described this as a "push pull" arrangement due to the half-bridge of MOSFETs, but it is not to be confused with an actual push-pull topology DC-DC converter. It is a synchronous buck converter topology.

 

[weblacky]

Yes, I've actually read this datasheet before as user on IrixNet is dealing with a low voltage O350 backplane and it uses a 2-pair mosfet design like this (same chip). I actually offered the suggestion that before he condemns the ICs, that perhaps there is a short on his dragging down the voltage and maxing out current with limiting (as his is extremely low...like 0.188v, instantly), but we never got any farther.

The overlap is an interesting idea, to occur, would the mosfet gate bias threshold need to have drifted/moved? Also isn't the low end mosfet the one that connects to ground? So if high-current was being drawn (ohms law on low voltage) wouldn't the mosfet guarding ground take the actual hammer brunt of the on/off cut-off during load? It's been my personal observation (in another buck DC mosfet pair controller case) that the "lower" (if I'm indicating right) of the pair that connects to ground really heats up on high current draw (due to a dead short after the converter, in my observed case) while the top one was not near as hot. In my observed case, I thought the lower mosfet was bad...but a new mosfet did the same thing...until I actually low-voltage bench PSU short-checked the DC output track of the converter and found the DC converter circuit was just feeding a HUGE dead short (in that case). But nothing was wrong with the converter, feeding a dead short causes low voltage/high amp and the lower mosfet takes on more heat/current.

I guess I blamed the filtering...but as well...it could be an intermittent (IC-based) high-current draw that's pulling down the voltage rail naturally? Hard to fathom but checking the other side of the DC rail isn't a huge problem if it's a real short. If not a static short...then...ughh yeah...unknown right now.

 

[Elf]

Take a look at shoot through current in a push pull arrangement of switches: if both the high and low switches are on at overlapping intervals, you are shorting the input supply to the ground for a short amount of time. To prevent this there is a dead time given between turning one off and the other on, since the devices can turn off somewhat slowly (relatively speaking). The only thing that does not fit this explanation is I would expect both to be hot.

Also the role of the low side MOSFET in a synchronous buck converter is not that straightforward; it essentially takes the place of the schottky diode you would see in a simple (non-synchronous) buck converter. But this is a simple version of a much deeper explanation that is probably better served by a power electronics textbook

It could be that some part of the controller has failed, or that some characteristic of the MOSFET has changed unfavorably, or something else entirely. A short on the output side is possible, but the voltage drops slowly, which would make me put that further down the list of likelihoods.

In any case, lacking the ability to reliably catch it in action to troubleshoot (and being reluctant to cause the issue further), I will just replace the whole section and see how it does!

 

[weblacky]

Yeah, I very much look forward to your report on those replacements!

My Tezro goes down slowly as well (just does it all the time now), I was only mentioning a similar circuit in a different case. I'm hopeful I can break the log jam with new testing equipment, fingers crossed for end of year, as that's the time I was told to call back the business that was going sell their old equipment to me (but backed out for 6 months due to reorg) and see if they will to now sell me their old/unused tools.

 

[weblacky]

Yo,
I'm nearing getting to this 1.8v VRM project and I have a slightly different take now that I've seen some more stuff on what MIGHT be happening.

I'm wondering if this isn't a failure of the above diagram D3 or D4 Zenors (or similar diodes on the circuit). I say this because I saw an OVER VOLTAGE of this kind of system due to a failure of the flyback diode on the MOSFET gate on one of the MOSFETs (never figured out it if was for UPPER or LOWER MOSFET Gate diode, assumed UPPER), replacing both diodes (needed to due to change in available parts, I had to upgrade both diodes to ensure they fire at the same time, but only one failed) solved the issue completely. The datasheet mentioned that the role of these diodes is to "clean up" the gate firing signal of the MOSFETS. Since this is situational/conditional it's an intermittent issue as it's a filter.

When they have a diode failure the MOSFET can be "triggered" incorrectly from normal wiggle that occurs on the gate signal line. I saw this in person on a Fuel PIMM VRM I worked on.

You can only detect the diode failure out of circuit, but I did found that to be the case. Though this was instantaneous (the above VRM I mentioned)...not a slowly ramping down thing. For you, this might make sense (instant). For Mine...I still think it's a cap issue because to goes down but about 15-20 seconds from fine to low.

I also think doing a MLCC de-aging in all the surrounding caps is as something that should be done as well.

But in the case that this ringing on the gate line, that is no longer squelched once you "get out of tolerance", the LDO regulator tries to get BACK into conformance and the signaling does a runaway and never ends well.

This could be a workable idea. On mine, I'll be desoldering all diodes and doing out of circuit tests to confirm and the cap first. On yours I'm thinking actual semiconductor damage.

 

[Elf]

Let me know what you find! Fixing this is still sort of a low priority for me, especially with the amount of work necessary to remove the board, but I will get around to it at some point.

 

[stormy]

Hey @Elf I've seen you mention a few of your projects (even your custom Indy psu) being low priority, I'm left wondering, what is your high priority project? Besides this community of course

 

[Elf]

Hi! During the majority of the year I was barely using computers at all outside of work, and had a lot of construction projects and time spent increasing my skillset in that area: buying heavy equipment and tools, doing dirt work, regrading, concrete, plumbing, and learning masonry (brick, block, and stone). That will all start up again later this year as well.

While the weather has gotten worse and the days are shorter, I have gotten back to indoor projects and I do have some SGI centric work in progress which I will release later. However I've also been working through a thick stack of books on technical subjects, and revisiting my music instrument (drums, guitar, bass, keyboard), and production (recording, mixing, mastering) skills, as well as taking some music theory and sound design classes.

 

[stormy]

Hi! During the majority of the year I was barely using computers at all outside of work, and had a lot of construction projects and time spent increasing my skillset in that area: buying heavy equipment and tools, doing dirt work, regrading, concrete, plumbing, and learning masonry (brick, block, and stone). That will all start up again later this year as well.

While the weather has gotten worse and the days are shorter, I have gotten back to indoor projects and I do have some SGI centric work in progress which I will release later. However I've also been working through a thick stack of books on technical subjects, and revisiting my music instrument (drums, guitar, bass, keyboard), and production (recording, mixing, mastering) skills, as well as taking some music theory and sound design classes.

You truly are multi-talented =)

 

[Elf]

Thanks

Unfortunately for those relying on me I have way more things to do than I have time to do it in, by several orders of magnitude, and some things can only be done during certain times of the year (subject to weather and daylight and so on). I have more discretion in the winter months, but being bored or trying to find something to fill the time with is definitely not an issue I will have for the foreseeable future. As a result you won't often find me just playing around with things these days, unless it is focused on learning a particular skill.

With SGI projects because they tend to rank a bit lower as "nice to haves" or "leisure time activities," and with there being no unfocused "leisure time" in my life anymore, more than a desire to be secretive I've stopped talking about what I want to do because I don't want anyone to get their hopes up or to rely on any of it being completed. Better just to show something if it is ready. I should have at least one or two things ready for people in the near term though!

 

[stormy]

I don't think, correction, I *know* I couldn't do things the way you do. I'm very much a one task at a time guy, my max is about three otherwise it actually makes me depressed lol. I like to start and finish projects as quickly as possible, I think because usually I'm doing projects of things I don't often do and therefore if I bench the project for a while, I'd have completely forgotten where I was up to and probably just throw the whole thing in the bin than relearn it from the start again. I see that your skillset is majorly in long term project management and wouldn't be surprised if your day job entailed that

 

[Elf]

There are similar things going on at work I will admit

Outside of (paid) work I do have a lot of projects that might linger in various uncompleted states for quite some time, but, they still weigh on my mind and I do eventually get back to them; I haven't truly abandoned too much yet.

 

[kln_nurv]

The interesting thing is that MrWeedser has fitted a standard ATX power supply into his Tezro and it works!

Quoting his post on IRIX Network: "

Just get a normal PSU for ATX, they work perfectly.
Mine is running with a standard corsair HX520, but you can get whatever psu you want. Go with a super efficient enermax or whatever you prefer.

The problem is the 10 Pin CPU connector besides the 24 Pin, because standard ATX PSUs don't have it, they are 8 pin maximum.
My solution was to put in the PCIe connector in the missing 2 pins, see the attached picture."

I am also trying this with a slightly disgruntled tezro. Powerup is blocked by 12v #2 having 0v. Would you happen to know if the substitute PSU must have separate 12v rails? This PSU is 750w and can do 60A on its single 12v rail. What could be the issue?

 

[Elf]

If I recall correctly from when I was poking around, it looks like the rails are separated on the output side and separately current monitored, but originate from the same point on the forward converter inside the unit. So it's likely that having them separate (as would be implied by others getting ATX supplies to run) is not strictly required.

I would try to measure the voltage yourself during power up, as a troubleshooting step. The environmental monitoring chips that are responsible for those alerts are also known to go bad.

 

[kln_nurv]

If I recall correctly from when I was poking around, it looks like the rails are separated on the output side and separately current monitored, but originate from the same point on the forward converter inside the unit. So it's likely that having them separate (as would be implied by others getting ATX supplies to run) is not strictly required.

I would try to measure the voltage yourself during power up, as a troubleshooting step. The environmental monitoring chips that are responsible for those alerts are also known to go bad.

Thanks. I'll try that. What is the level of risk of trying to force powerup with env=off? All fans spin well so I don't expect any cooling issues.

 

[Elf]

I would think lower risk if the ATX supply has been tested and is known to function well, if you are doing so to monitor the voltages yourself

 

[marcdat]

Hi there,

The release buttons for opening the side covers of my TEZRO have been lost, I can't open them to remove the power supply.

Can somebody help me please?

Thanks in advance!