1. Home
  2. Electronics Design
  3. What price impedance match?

What price impedance match?

There have been several threads recently about impedance matching PCB traces.

But how do you get that impedance to match?

You can do wonderful things with cad programs. Problem is that boards aren't made by cad programs. They're made with inconsistent materials and process variations.

How accurately do you want your impedances controlled? What's the trace width specification from your board vendor? How about the thickness spec on each layer? The dielectric constant spec?

If you add it all up, you may find you can't get there from here.

If you want specified impedance, you have to specify it. And pay for the testing at the vendor and at incoming inspection...and provide test point access.

Call up your vendor and tell them you've got 12 layers and want impedance control on six of them. See what that does to the cost quote. You might want to rethink your layup.

If you specify it, it's gonna be expensive. If you don't, you're gonna get screwed. I've seen boards fail because the vendor switched to a fiberglass with tighter weave. I've seen boards fail because they used too much pressure when laminating the board stack. I've seen boards fail because a process engineer switched a copper layer from one piece of fiberglass to the adjacent fiberglass. Same (nominal) thickness everywhere, the board just didn't work. Also remember that purchasing agents get paid to reduce costs. If the original engineer is gone and there's nobody to ask, it's easy for them to save a few bucks on the board... and shut down the production line.

Engineers need to sit down with the board vendor long before the first prototype is ever made. There's a lot more to it than pressing the "route" button on the cad program. It's an analog problem.

I've not personally specified a board in 20 years. Today, if you had 100 ohm traces on a bunch of layers, What impedance variations would you expect over the life of the product due to pcb manufacturing tolerances?

Reply to
spamme0
Loading thread data ...

I do controlled-impedance stuff all the time, and I specify stackup, not impedances, and usually work in FR-4 at 4 to 8 layers.

Last board I did - laid it out myself! - is a little pulse generator. I did it as a 4-layer board with the 1-2 dielectric spec'd as 12 mils, which is very common for board houses. Layer 1 traces TDR'd right on the money, 52 ohms. I did have a small capacitive glitch at the edge-launch SMA, but not enough to hurt. I was in a rush so guessed at the ground plane cutout under the SMA pads, and missed by a little.

ftp://jjlarkin.lmi.net/J245_R5.gif

Anyhow, it works, makes 50 ps edges and nice 2 GHz square waves. Shipping two on Thursday, $3500 each.

If you can tolerate up to, say, 5% impedance errors, specifying stackup is good enough and saves a bunch of money. 5% is plenty good enough for digital stuff and, I'd guess, most RF. Hell, the RF boys think that a VSWR of 1.5:1 is good!

It does pay to call the fab house and get their preferred stackups, and maybe a scrap board.

I do put test traces on my boards so I can TDR all the layers that matter. That closes the learning loop.

John

Reply to
John Larkin

I had the occasion to do an ECL board for a stupidly fast video DAC we needed to bench. [Competitors part.] I found the tolerances on the dielectric were pretty wide. I ended up spec-ing a really thick dielectric to keep the error down, at the expense of a wider trace.

One interesting tidbit. We were using HP ATE (9480, 9840, some number like that). HP build the DUT boards. Well, the HP salesman loaned us a scope with TDR built in. [Very sexy at the time.] Sure enough, the HP traces were about 40% off target impedance. Mine were on the money, but nobody would build a board as wasteful in space as that bench test board.

So all this said, I think the original post has some merit. Of course, there is lots of stripline out there, so perhaps with a good relationship between vendor and design team, the impedance variation can be tamed.

OT, but at what point in time will it be necessary to call HP by Agilent? Will there be a day when you say HP and people think made in China computers?

Reply to
miso

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.