'Design' vs 'Specification' dielectric constant

[radar_macgyver]

When attempting to do a design using RO4350B, I noticed that the data sheet specifies two different epsilon-R (dielectric constants) for the material, one at 3.48 +/- 0.05, the other at 3.66. The latter is termed "Design dielectric constant", and there are a couple of app notes indicating this is the value one must use when designing transmission line components.

Various software that includes libraries of dielectrics (eg: Genesys) list the dielectric constant for RO4350B at 3.48. I can also reverse-engineer eval PCBs from, say, Mini-circuits and find that they have used Er of 3.48 for their designs. Why, then, does one need this new value of Er? It's almost 5% off the original value. For a simple transmission line, a 5% error is likely not a big deal, but it would dramatically change the operating characteristics of, say, a hairpin filter.

What am I missing? I'm quite the noob regarding microwave layouts, as one can tell.

[darrell]

It's mostly the IPC clamped stripline test method not being appropriate for these materials. The design value should be closer to the true value.

Lots of detail here:
https://www.rogerscorp.com/documents/2441/acm/articles/The-Influence-of-Test-Method-Conductor-Profile-and-Substrate-Anisotropy-on-the-Permittivity-Values-Required-for-Accurate-Modeling-of-High-Frequency-Planar-Circuits.pdf

[radar_macgyver]

Thanks for the information. I guess that means I should use Er = 3.66 when designing my board, and that RO4350B due to its anisotropic properties, might not be the best material if I intend to fab a narrow-band filter.

[darrell]

I think you will get a closer match to simulation if you use Er=3.48. Edit: 3.66 is what I use.

Narrow band filters on RO4350B or any epoxy type laminate are probably a bad idea. Uncertainty in Er can really stack up with process, temperature, aging, measurement methodology, etc.

[jungle vegetable]

Also laminate thickness tolerances (up to +-10%) could have comparable or larger impact as the difference between these two Er values
So if you want better precision than that, you might need to work with your fab on prototypes and process repeatability

[radar_macgyver]

Thanks for the inputs, looks like I have a lot to learn.

[darrell]

I think you will get a closer match to simulation if you use Er=3.48.

I happened to look at the RO4350B spec sheet today and realized I had swapped the two numbers in my head. I use 3.66 for design. Sorry about that.

[radar_macgyver]

Thanks for clarifying, I was about ready to write it off as 'another bit of RF voodoo I will never understand'

I wonder why the libraries in Genesys still indicate 3.48 for RO4350.