[robespierre]

Weblacky, I don't disagree that there needs to be a way to repair SGI equipment, including producing schematics and board views. The problem is that for some devices like the Z EYE, each time you take it apart you likely reduce the population of intact units. Forget that piece of Scotch tape in the correct place? Zap. I'm still working out how to clip the damn thing back together without a vise.

Some of your other points feel wrong to me. Is there a reference for ceramic capacitors "aging"? I think I understand MLCC failures and that isn't one I'm familiar with.

[weblacky]

Yo,
I can provide easy reference to MLCC again and de-aging...it's real:

https://www.digikey.com/en/pdf/k/kemet/c...-to-expect

https://product.tdk.com/info/en/contact/...74018.html

Now because this reset naturally occurs during initial circuit construction, most completed circuits cannot be simply placed into a reflow oven for a widespread cure. There are parts that simply won't survive the slow, global, reheat. However individual caps/parts CAN be de-aged by simple solder tweezer reflow (not removal) as long as the substrate gets to around 210C for like 2 seconds.

Then 24 hours is required until the MLCC will once again register a stable capacitance (near original value too). This really happens in decades. However a 20 year old SGI will definitely have this and the original de-rating isn't enough compared to the DC bias derating you already do. So capacitance actually goes down and the DC bias slides into unacceptable.

This cure is a known and real, but individual reflow is needed. Hot air is considered VERY risky because it will heat too fast and cause layer separation or cracking of the MLCC. They claim for small numbers, a simple reheat of the solder (not touching the SMD CAP is possible) reflow (letting the solder bridge heat up the part) and then release to cool is enough to perform the reset.

Now some caps are like the frigg'n size of the period on a sentence, is that going to be hand-reflowed too? Hmmm...not by me, But large ceramics I think should be top priority during this phase. Big packages and values are often used on power rails. So regaining the original capacity is more important than on a purely logic area. For lots of small caps in a dense area I might try a "low heat" hot air to try to deage them without full reflow...anything might be better than nothing.

Also, onboard measurement and perhaps even replacement maybe a worse or better option? Yes MLCCs have come a long way...but this hasn't really changed from what I'm reading.

I'm not advocating diagrams because that will never happen. But even high-res PCB pictures with component values and labels (maybe even replacement part numbers) is what I'd like to see the community move towards. I don't want people to ask what this charred/black thing was, I'd rather they can look at info and know what it was without asking.

[robespierre]

Neither of those guides (nor another that I found in an AVX catalog) refer to aging as a source of failure, only as a factor in evaluating component measurements for acceptance purposes.
Look at the graphs! After 10,000 hours (~ 1 year), the change in value is 5% for a common medium-stability dielectric (X7R). Ever wonder why it's called a "medium-stability dielectric"?

Q: What is the tolerance for X7R?

The change is also exponential, so as time goes on, its effect is less and less. This is not the profile of a wear-out mechanism.

[weblacky]

I think you may have missed the forest for the trees. Those hours not operational hours, the graph claims since "last heat". MLCC age doesn't normally factor into manufacturer part acceptance (unless stock is VERY old) because the formulation is supposed to "rebound" back in tolerance upon heating, so no one cares if the inventory of MLCCs is a like 2-3 years old. That's why they don't have machines to "test" MLCCs in the placement machine before use. It shouldn't matter, soldering should restore a 10+ year old MLCC in the packaging back to acceptable.

So since last reset of the dielectric, by soldering. 20 years is 175,200 hours...and growing.

Class II X7Rs start at 10% and grow larger in tolerance ratings. Also please don't make an assumption that SGIs (in general) contain X7R caps, those were invented in the early 2000's from my research (X7R patents seem to start around 2004+), assume an older rating...which is much more likely for 1996 = O2 timeline. So worse....

Ceramic cap failure happens on modern consumer electronics now, let alone in the past. If you look at the graph and factor in original DC bias derating for correct circuit operation, after 2+ decades, the caps can easily be well out of an original design tolerance (20%) and then some (eating into base capacitance assumption during design), also fact is that many MLCC caps have a DC bias so bad, that they need 50%+ derating, so what's left? The circuits were designed with DC Bias Derating + a stated tolerance as originally installed. The capacitance isn't going to grow...it also states aging is logarithmic. I think that point speaks for itself.

If you're concerned about MLCCs shorting (and you should be), having an cap be LESS than it's designed tolerance + value means you're operating it out of expected norms.

That's not too different than tants that have increased chances of failure due to running them close to their own ratings. In this case, the MLCCs are slowly coming down...naturally and while voltage rating MAY not change on an MLCC, the effect of running an "undersized" cap for both ESR, ripple, and capacitance is a problem itself.

Now it's still not too common, but I'm bring up everything to do with vintage/aged electronics. This is real and matters. A ceramic cap can fail shorted and short out a power rail, just like anything else. The cap can be "refreshed" with a manual operation. Will we get all caps...no. But when working in a power area or decoupling caps on a power rail, it means it may actually pay in the long-term (keyword here) to gently reflow those MLCCs when also addressing other cap chemistries as well.

Fair?

[Fascia]

There was a request for more disassembled photos of the O2 cam, so I'm just posting them here for posterity for the curious.

Some notes: That cable has ZERO slack inside the case, and it reach all the way across the board to plug in. The camera electronics are housed in that metal enclosure, and it has to be wrenched out of the camera to take apart and unplug the cable. By sheer luck and cable tensile strength I was able to pull it apart without breaking it, but it wasn't easy. The way this came apart tells me it definitely wasn't meant to be serviceable. I suppose if someone had a bad camera they figured they'd just ship a new one. Cheaper than engineering this to be easy to disassemble.

   

   

   

   

   

   

   

[jan-jaap]

Talking about repairs at the board level: I had two broken GE16-4 Geometry Engines. One passed irsaudit, but even the simplest demos put distorted geometry on screen. The other complained about flash. I asked one of my colleagues to replace the flash chip with the one from the GE16 with broken geometry, he was busy, but finally got tired of having these boards around, so:



I'll see sometime soon whether it works now.

[robespierre]

That cable has ZERO slack inside the case, and it reach all the way across the board to plug in. The camera electronics are housed in that metal enclosure, and it has to be wrenched out of the camera to take apart and unplug the cable. By sheer luck and cable tensile strength I was able to pull it apart without breaking it, but it wasn't easy. The way this came apart tells me it definitely wasn't meant to be serviceable.

After some more time struggling with it, I think I understand how it can be serviced. There are very specific steps that must be done in a certain order. Appropriate tools are also required. I needed to use a hook probe and an orangewood stick, which are common technical tools.

Fascia, you should pay particular attention to the pictures above because they reveal a serious issue with the cable: its outer shielding is a loose metal "wool" instead of a tight braid, and it frays and shorts out anything it touches. That's why there was a square of Scotch tape over the board covering your burned capacitor (not sure if it melted when the tantalum ignited; I don't see it in any of your pictures). That tape is required to prevent destruction because of the way the outer braid is uncovered inside the device. It would have been nice if Logitech (or whoever actually assembled this thing) had used heat-shrink insulation there, but they chose not to do so.

Addendum: it also means that air ionization was used in the factory that produced these units, since that is the only situation where Scotch tape can safely be applied to electronics (it generates static when dispensed).

[Fascia]

Yeah, that frayed shielding didn't look right to me. But seeing the tape it seemed it was intentional. I wonder if years of the cable being shifted and tugged could help the shielding at the end work its way out and the frays to get beyond the protection of the tape.

[robespierre]

I think the idea was that the outer braid would press against the upper metal cover, and the inner braid is connected by the faston to the lower metal cover. They also press together, of course, but that might not be low enough impedance. To improve the design we could connect the outer braid to a drain wire, solder the drain wire to the upper cover, and then surround the outer braid with heat-shrink. That would maintain the grounding scheme without risking shorting to the board.

To apply the heat-shrink, all of the wires in the 26-way connector can be removed, and then pushed through heat-shrink tubing; or the easy way is to use self-adhesive heat-shrink tape.