asymmetric differential stripline

Plug it into ATLC? ;-)

Differential calculators are hard enough to come by, let alone the various special cased ones (CPW, asymmetric..). Most calculator programs are based on formulas from the same publications, which come from simplified approximations (usually using logs and polynomials), obtained under modest assumptions (like trace width

Too much work! Posting to s.e.d. is easier.

Too many people use the single-microstrip equation from the Motorola ECL handbook, without limit checking. Results can be ludicrous.

Appcad is nice, but doesn't do diff pairs, much less lopsided ones.

I have Wadell, and it's crazy. A half-page formula will have terms that are expanded on three more pages.

I trust ATLC and ATLC2, but they are a nuisance to run. I can ignore losses for short digital runs.

Fortunately, I can tolerate a 20% or probably more impedance mismatch, a little drool or overshoot, and most signal edges will be in the 1ns or so speed range, so things aren't extreme. One DDR3 differential clock will be fast, but it's very short and we can keep that all on layer 1.

Designing PCB stackups is a lot of work. Gorilla Circuits has an excellent 67 page PDF doc all about this, but I shouldn't post it because it's "confidential." Maybe available on request.

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

Looks symmetric to me. Give me the core and prepreg info and I'll calculate clearance/width values for you.

Using the values given (assuming inner dielectric constant of 4.3 and

1oz copper) a clearance and width of 9 mil and 4 mil gives an edge coupled differential impedance of 100 ohms on inner layers 3,4,7,8.

For simple geometries

looks OK.

It is asymmetric: H1 H2.

There will be two signal layers between two plane layers, with 6 mil dielectrics and 1 oz traces. So any trace sees 6 mils of dielectric to one plane and about 13 to the other.

=============================== plane

---- ---- sig

---- ---- sig

=============================== plane

(not literally. Traces will be orthogonal to reduce crosstalk)

That's close to the Saturn solution.

Thanks

--

John Larkin         Highland Technology, Inc 
picosecond timing   precision measurement  

jlarkin att highlandtechnology dott com 
http://www.highlandtechnology.com

If the free tools can't handle that just use the average of the two gaps on either side, and calculate the spacing and width using an impedance about 5% higher than the target.

IIRR 100R differential pairs are hard to get with buried stripline, unless you use a low dielectric constant material for your boards

I had a text that gave elaborate formula for buried strip-line impedances, but it looks as if it is still in Nijmegen

seems to present something similar.

The simple equations for characteristic impedance tend to be good only around 50R. The formula used on the website I found is messy enough to do better, but doesn't involve any hyperbolic functions, which are what seem to be required for anything serious.

--
Bill Sloman, Sydney

ound 50R. The formula used on the website I found is messy enough to do bet ter, but doesn't involve any hyperbolic functions, which are what seem to b e required for anything serious.

ultracad had a good calculator but I see it is no longer free.

m

I recall it making some really bad errors. It didn't range check.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

Nonsense, pretty much all recent high speed digital serial buses are 100 ohm differential (Serdes, PCIe, SATA etc.), and are built using materials with dielectrics in the low 4's.

There is no need to calculate any of this. Your board manufacturer will advise on the stackup (based on the available core/prepreg material they have) and calculate any spacing/width requirements for all the impedances you need on all the signal layers.

I had a look at the Saturn tool John is using, and it looks to give results fairly close to the commercial tool I use, for the few cases I tried.

The dumb way to make a 100 ohm diff pair is to use two 50 ohm traces spaced fairly far apart. That's not very far for a stripline pair on a thin inner layer.

Most board houses will do 4 mil traces without complaining much, so 50 ohm traces inside a 10-layer FR4 board are OK. We need 4 mil traces already, to do BGA breakout.

I have a zillion various-controlled-impedance traces on a big 10-layer board. We'd rather get the impedances right on the original layout, without consulting an outside party about every one. And we want to be PCB-vendor agnostic. For digital-to-digital signals, all you need to do is get pretty close.

That's encouraging. The Saturn thing sure looks nice.

--

John Larkin         Highland Technology, Inc 

lunatic fringe electronics

ess you use a low dielectric constant material for your boards

Oops. Paired 50R lines would be fine. I do recall not being able to get bur ied 75R lines.

es, but it looks as if it is still in Nijmegen

around 50R. The formula used on the website I found is messy enough to do b etter, but doesn't involve any hyperbolic functions, which are what seem to be required for anything serious.

In other words they can be relied on to do the calculation for you - which they mostly can. On the other hand I've got a painful memory of a board hou se scrambling a carefully worked out layer order because they through that it might make the board warp - it didn't when they finally did it right - w recking all the carefully worked out transmission line impedances in the pr ocess.

It was particularly irritating because the board got passed to different en gineer for debugging - it had been designed by a sub-contracted engineer - and whenever I walked past his bench I'd point out that he had the board la yer print-outs organised in the wrong order. It took him weeks to get aroun d to checking that he didn't and only then did it become obvious why the bo ard was working so badly. There were other problems - even with all the cri tical connections made with sub-minature coax it didn't really work.

--
Bill Sloman, Sydney