Microstrip filter fabrication problem - any suggestions?

[A_G_Lent]

Hi everyone,

New poster here, so hopefully this is in the right place...

I am looking for any suggestions on tracing down a fabrication problem we have with a microstrip filter on our boards. It's a 6GHz hairpin filter, on Rogers 4350B, gold/nickel plated copper. We've made thousands of these boards without any problems, but the latest batch has come back with a much higher insertion loss on the filter than we've seen before (around 6dB IL, rather than the typical 3.5dB). The filter bandwidth / centre frequency is unchanged between the batches, it's just a matter of increased insertion loss. There are several of the filters across an A5-sized board, and they are all similarly affected (so on an old batch all the filters are good, and on a new batch all the filters have high IL).

The PCB manufacturer is claiming that everything is normal, but nevertheless we see that the new boards just behave differently for some reason. They (and we) are reluctant to just rebuild unless we understand what has changed.

The PCB manufacturer have measured (by cross-sectional analysis) the trace/gap widths (no obvious problem), the plating depth (no problem), and the substrate thickness (no obvious problem). I have the CofC for the substrate material, which seems valid.

In short, I am a bit stuck on ideas for what to chase down next - it doesn't look like anything I can see on the surface of the board. I wondered if there was some subtle anisotropy of the R4350B panels, so if our board was made "with" the top-level weave it might perform better than if it was made "against" it, or vice-versa. But I can't find any suggestion that R4350B behaves like this. Or perhaps there is some PCB processing step that has changed the substrate adversely in the new batch?

Playing around with our Sonnet simulation of the filter, it looks like the only things that might give the effect we see are a substrate with a much higher tan_delta, or a much lower conductivity metallisation. But in both cases, it would take a big change (x5 tan_delta, /5 conductivity) to replicate what we are seeing. [ Our first thought was that overetching might be to blame, but (a) that's not what we measure under the microscope, and (b) in Sonnet at least the filter isn't super-sensitive to that. ]

Has anyone out there seen similar problems before? Any tips on how to debug this?

Thanks!

 

[FvM]

What's the PCB stackup? Basic microstrip (double side plated Rogers)?
Do you actually see losses or might it be mismatch? (Consistency of S11 and S21 measurements)

 

[BigBoss]

Metallic losses have second order effects on the attenuation.Tand D has a dominant on the overall attenuation.
Question : How the manufacturer measured Tand D ?? Did they measure it or they're talking with dielectric manufacturers' claims and lot ( manufacturing ) reports ???
If everything is same, the results must be same as before.
Did you do a sensitivity simulation ?? I mean, can you do a simulation by changing Tand D as a variable ??

 

[biff44]

it could be the plating thicknesses are way off. Like if the nickle layer was too thick, and the gold on top too thin, that would kill the resonator q.

It is rare, but i have see boards where the copper traces have microcracts in them...making it like it is not a continuous piece of metal. As i remember, the plating was discolored right over where the microcracks were.

you say the filters are still centered, so the board dielectric moving....or the wrong dielctric material used is not an issue.

 

[volker@muehlhaus]

It's a 6GHz hairpin filter, on Rogers 4350B, gold/nickel plated copper.

Given the tight tolerances and quality control of RO4350, I would suspect that the substrate itself is OK and only the postprocessing (nickel/gold plating) went wrong.

I never had that problem myself, but colleagues told me that nickel plating is really critical for loss. What if the nickel is too thick and/or the gold too thin?