Fancy PCB materials--wisdom?

I can recall GM first making engine-control processors on 3/16" thick Alumina. Surprisingly strong... one engineer there demonstrated the strength by pounding a nail with the board ;-) ...Jim Thompson

Yup, I've used it before in a previous life. Alumina's dielectric constant is like 9.8 or something, though. I'm looking at Rogers RO4350 or RO3003. The 3003 material is Teflon based, but it looks like they've done something to it to get rid of the nasty room temperature glass transition. The 4350 stuff is hydrocarbon-based, and seems to be advertised by some board houses e.g. Sunstone, but its epsilon is around

Cheers

Phil Hobbs

Polysulfone? Glass transition temp is 185C. Low dielectric absorption. No idea what fab houses might still use it.

Just curious: Usually it's only the capacitance to neighboring traces and pads that is difficult to handle. Have you considered milling out any non-essential areas and just save whatever bridges are needed for stability and traces?

Polyimide, but Kr is around 3.8, not dramatically better than FR4.

FR4 is pretty good stuff. Leakage is almost ummeasurable. And it's cheap.

A small copper pad on FR4, say 20x20 mils, 062 thick, only has about

A parallel-plate cap made of FR4 does have a pretty radical TC, like

Yowser! dats a lot of bootstraps, (one is all I can usually handle.) I have no idea what to use, but it seems to be a trade off between vibration stabilty and some extra dielectric constant. (just thinking you could make it thinner.) Re Joerg's reply: How bad is the pcb dielectric? It's only a small fraction of the total space involved. (assuming thin traces and good layout.)

George H.

Are you sure that you've been looking at the right PTFE?

puts the glass transistion temperature of PTFE ar 130C, which lines up with my experience. Coax with a PTFE dielectric was perfectly stable at the melting point of solder.

For very high impedance and high voltage circuits we used to isolate the sensitive nodes with pins stuck through a PTFE collar that stuck into a circular hole in a regular printed circuit board. Farnell still carry them ITT CANNON TERMINAL, PTFE, TEST as part number 1347797 along with a bunch of similar parts.They aren't cheap.

PTFE doesn't soften at room temperature, but it has a phase transition where the CTE peaks at about +1000 ppm/K and epsilon drops by about 1% in a few degrees. Garden variety PTFE/woven glass substrates show the same behaviour.

Cheers

Phil Hobbs

FR4's epsilon is not very consistent from batch to batch, and it absorbs water. With multi-gigohm feedback resistors, the high frequency gain is set almost entirely by the stray capacitance.

Part of the goal is improved unit-to-unit consistency, part is stability, and part is just reducing the strays. In bootstraps, as you know, the high frequency noise is set by the input capacitance differentiating the first stage's voltage noise. The application isn't too cost-sensitive, but we do need to be able at least to get a 6-layer board fabbed.

Cheers

Phil Hobbs

There are really only two bootstraps, the main one and the one that bootstraps the drain of the main one. The main one has really low output impedance, so it can drive the shield as well as the inverting input of the op amp.

Keeping the strays low, stable, and consistent is easier with a more predictable material than FR4.

Cheers

Phil Hobbs

In this case the two places are across the gig-to-tens-of-gig feedback resistors and the input node. For noise, it's mostly the capacitance to the bootstrapped reference plane, but for stability it's the capacitance across the FB resistor.

Either way, the bootstrapping has to be effective at the 3e-3 level, so I'm reluctant to have a lot of big holes in the reference plane.

Cheers

Phil Hobbs

The reference plane could be on another layer and you can make the board almost as thick as you want. For horizontal paths you can still have milled out cuts. They do not necessarily have to go all the way through. For example, you could have the grinding bit's vertical travel stop just short of the reference plane. So you'd end up with trenches instead of cuts.

Interesting point. I'll keep that in mind.

Cheers

Phil Hobbs

Might want to also look at designing with unplated drilled slots. They can be made quite narrow methinks (0.6mm ~ 0.8mm, depending on the manufacturer). The slots should be generated with a specific G-code (G85) rather than creating overlapping holes directly through the NC drill file.

Rogers 6002 doesn't have the lowest Er, but it does have one of the best tempcos of Er (12ppm/K) over a normal operating range.

Thanks, I'll put that one into the mix. Any experience using it with a reasonably mainstream board house?

Cheers

Phil Hobbs

Nope, sorry. I see the Rogers materials at suppliers all the time though, so it should be okay (at some price, lead time and MOQ!). Sunstone does offer the 6002 material.

BTW, it should be possible to get _blind_ unplated drilled slots (so you could have a solid ground plane with critical traces sitting on isolated dielectric mesas), but I would certainly check that with the mfr before counting on it.

Yeah, with my limited experience with fancy materials, it seems as though in board fab, Sunstone is a reasonable equivalent to Digikey as a gauge of parts availability.

That would be artistic for sure--the customer could tell what a swoopy design it is just from looking at it. ;)

Cheers

Phil Hobbs

How about using a mesh instead of a solid plane?

I you want really low c, use a very thin kapton flex board over an air gap. The microwave people do that to keep dielectric losses down; they call it "suspended substrate."