Posts by azonenberg@ioc.exchange
(DIR) Post #B752wxqtKvGuSLWv1E by azonenberg@ioc.exchange
2026-03-06T05:14:11Z
0 likes, 0 repeats
Oh yay, I thought my wife had gone to bed already but she was still up. Got her to click a few buttons for me and now I have a photo of the measurement setup and a clean VNA sweep with a wider bandwidth.Sure looks like the resonator is intact.
(DIR) Post #B752x6dOgSpDh6Hljc by azonenberg@ioc.exchange
2026-03-06T05:17:44Z
0 likes, 0 repeats
(some frequency shift from nominal is expected due to the forces of the probes disturbing things etc, plus the VNA is IIRC not wired to my GPSDO yet so it might have a slight drift in the timebase)
(DIR) Post #B752xFLIJ8gyhYswGe by azonenberg@ioc.exchange
2026-03-06T05:58:00Z
0 likes, 0 repeats
OK yeah the controller is definitely what fried. Measuring from VDD to ground I see zero current draw at 3.3V. It should be around 6 mA.
(DIR) Post #B752xNfRM7YoWMN8mO by azonenberg@ioc.exchange
2026-03-06T05:58:34Z
0 likes, 0 repeats
All of this is leading me to suspect a broken bond wire but before I do anything else I want to measure resistance from each pair of pins and confirm whether vdd or ground is the open circuit
(DIR) Post #B752xW2mDEysV3LtGS by azonenberg@ioc.exchange
2026-03-06T06:16:58Z
0 likes, 0 repeats
Sooo... ALL pins (measured at the little castellated U shapes) measure open circuit to any other.
(DIR) Post #B752xeRB0PFgX2pUPI by azonenberg@ioc.exchange
2026-03-06T06:19:54Z
0 likes, 0 repeats
Anyway, that's multiple pieces of evidence to suggest something catastrophic happened to the controller chip and that the resonator itself is fine.So I guess I'm at the point where I need to rip off the quartz crystal to see what the controller looks like.
(DIR) Post #B752xmEi0e24ih3e5I by azonenberg@ioc.exchange
2026-03-06T07:12:28Z
0 likes, 0 repeats
It looks like the tip end of the quartz is sitting on or slightly above that little protrusion on the case, and it's only attached by the two silver epoxy bits. So if I can get in at that end and pry up hopefully it'll snap or peel off at the epoxy bonds.
(DIR) Post #B752xvG6SpUmhcmO0W by azonenberg@ioc.exchange
2026-03-06T07:17:24Z
0 likes, 0 repeats
One final view with the polarizer on before I start removing it.
(DIR) Post #B752y2zjhZ9mhBBQbg by azonenberg@ioc.exchange
2026-03-06T07:18:14Z
0 likes, 0 repeats
Getting ready to pop it
(DIR) Post #B752yCDtOIpPJgsMQC by azonenberg@ioc.exchange
2026-03-06T07:19:47Z
0 likes, 0 repeats
I just realized I'm an idiot. Looking back at the original photo I don't think those side castellations I was probing actually connect to the internal pins at all.Meaning my open circuit measurements aren't valid. Too late.
(DIR) Post #B752yL8uEDBeypc0Om by azonenberg@ioc.exchange
2026-03-06T07:22:11Z
0 likes, 0 repeats
Anyway the quartz element came out without significant visible damage (so I could hypothetically re epoxy it back in although I'm sure it would have parametric shifts lol)Here's the controller bondout
(DIR) Post #B752yTcygHzlIPk8Nk by azonenberg@ioc.exchange
2026-03-06T07:30:48Z
0 likes, 0 repeats
And here's the die under the 20x objective.All of the wire bonds look *intact* which makes me wonder if maybe the problem *was* the epoxy bonding to the quartz shaking loose and not making good contact?Looks like a 3 metal process, planarized but fairly large feature sizes (I'd ballpark a 350nm class node).
(DIR) Post #B752ycLaGKQgL4fs1I by azonenberg@ioc.exchange
2026-03-06T07:35:03Z
0 likes, 0 repeats
Wasn't sure if the 100x objective would be able to reach with the die so deep in the cavity of the package but it just barely clears.100x scan running now then will do a 100x with focus stacking.
(DIR) Post #B752ykAt9ycycDjRSa by azonenberg@ioc.exchange
2026-03-06T08:00:51Z
0 likes, 0 repeats
Found the problem! Need to get a good angle on it with the stereo microscope but there's a failed bond wire.
(DIR) Post #B752yshnRVhj4bBprc by azonenberg@ioc.exchange
2026-03-06T08:13:11Z
0 likes, 0 repeats
Here's two of the good bonds. These are second bonds from a gold ball bonder, which means the wire is being fed through a ceramic capillary needle then pushed down against the bonding surface and ultrasonically vibrated while heat and pressure are applied to weld it in place.You can see the circular indentation on the pad where the capillary landed, and the flattened, squished part of the gold wire.Most importantly, using the focal plane as a cue to depth, the bond wire is making good contact with the surface.
(DIR) Post #B752z1V0kPpCLYHFgW by azonenberg@ioc.exchange
2026-03-06T08:13:59Z
0 likes, 0 repeats
Here's the failed one seen from the side under the stereo microscope. The defect is more obvious to the naked eye, the camera doesn't show it too well.
(DIR) Post #B752z9eWQwBbdMmxcW by azonenberg@ioc.exchange
2026-03-06T08:17:01Z
0 likes, 0 repeats
And here's the failed one top down under 20x objective, with the focal plane at two different steps.With the focal plane at the pad surface, you can't see the wire at all. With the wire tip in focus, the pad is very blurry.That means they're at very different heights and not touching. So we have an open circuit failure here.Given that the board was sonicated, this is the longest bond wire by far (the one crossing across the controller die to the opposite side of the package), and there's no support for the wire like IC molding compound would provide, it seems reasonable that it probably shook around violently during the cleaning process and built up enough momentum to snap the bond at the wire-to-package interface.
(DIR) Post #B752zHPDbihLgDgqsS by azonenberg@ioc.exchange
2026-03-06T08:18:12Z
0 likes, 0 repeats
Anyway, I think we're done here - the failure analysis was successful, we've visualized the defect and it's consistent with the suspected origin of the damage.I'lll do a blog-formatted writeup at some point soon and share the link when it's ready, but as far as the actual lab work is concerned I have nothing more to add.
(DIR) Post #B752zQ1RZU1oP5dUZM by azonenberg@ioc.exchange
2026-03-06T08:24:51Z
0 likes, 0 repeats
(Other folks are saying maybe it was a manufacturing defect and that sonication should not have damaged the bond wire, but we'll probably never know)
(DIR) Post #B77aJsL5gX3mp26k9A by azonenberg@ioc.exchange
2026-06-08T06:18:42Z
0 likes, 0 repeats
@wren6991 slop, i assume.Although... I have been seriously thinking about one day making a JIT shader compiler in libscopehal that if you e.g. chain two simple math functions end to end, and nothing is using the intermediate results, will be able to concatenate the filter kernels to avoid allocating buffers and storing/loading from memory needlessly.This would be probably done by having prebuilt sub-kernels that read/write from predefined local variable names or something, then doing simple string concatenation on the blocks, then invoking glslc at run time to generate SPIR-V from the compiiled kernel then feeding it to vulkan.