New to high speed hardware design, should I worry

[jonwilhelmjr]

My manager just told me I will be in charge of a 5Ghz bandwidth ethernet passthrough board. He says it's just traces. So everyone knows I'm a fairly green engineer, my pcd experience is under 10 Mhz. Mostly arduino like designs with microchip and atmel 8 bit micros. My question for everyone should I be freaking out or is there books/videos I should binge read or watch? How do I give myself a crash course in GHz pcb layout? Is it just a manner of controlling the trace impedance and stay away from harmonics 1-10? Any wisdom or suggestions is appreciated. I look forward to this learning experience and I hope I don't hit a brick wall when running into this.

[gnuarm]

Not sure what you mean about staying away from "harmonics 1-10"  But if there are no chips on the board the main issues in running differential traces, which i assume you are working with at 5 GHz, is to control the impedance and maintain matched length.  If the board is just running traces between connectors and you have control over the placement, put the connectors parallel and assign pins that do not require crossovers.  Then matched length is simple.  Vias are also of concern at 5 GHz.  So these traces are ok.

      /------
-----/ /-----
------/


These traces are not

       o------
------/-------
-----o

Different lengths and unmatched vias, or vias at all if you can avoid them.

Some may talk about avoiding sharp corners which creates a tiny lumped capacitance.  This may not be a big factor at 5 GHz, but that is possibly the threshold of it being significant, so to be on the safe side use 45 degree corners which have less of a problem.  Or you can get all RF like and use round corners.  Just make sure everything is symmetrical on the two lines of the differential pair.

There should be plenty of references on running differential controlled impedance traces.  It has particular requirements, but it's not rocket science.

[tggzzz]

How big will the board be, how long are the traces, how long are the existing traces, and what are the connectors?

You will find many references about traces, impedance mismatches, reflections etc. I'm not going to repeat them.

You may need to be careful about the choice of PCB material: FR4 is lossy and the fibreglass weave can cause problems, especially different boards have the weave in different position relative to the tracks.

[tom66]

Make sure you understand PCB stackup and the consequences of the design there.

[jonwilhelmjr]

Not sure what you mean about staying away from "harmonics 1-10"

As for the harmonics, see here, I believe he referenced how you do not want a trace length to equal the wavelength of your fundamental or  harmonics 1-10. The reasoning is due to the reflections will cause noise. I may have misunderstood and need to review the video again.

[jonwilhelmjr]

I have a meeting today on the project. I will reply with more details once i know more.

[tggzzz]

Not sure what you mean about staying away from "harmonics 1-10"

As for the harmonics, see here, I believe he referenced how you do not want a trace length to equal the wavelength of your fundamental or  harmonics 1-10. The reasoning is due to the reflections will cause noise. I may have misunderstood and need to review the video again.

Reflections cause inter symbol interference which closes the eye pattern, not noise. There is no need for the length to be related to harmonics for it to cause trouble.

Google for Bogotin's rules of thumb.

[gnuarm]

Not sure what you mean about staying away from "harmonics 1-10"

As for the harmonics, see here, I believe he referenced how you do not want a trace length to equal the wavelength of your fundamental or  harmonics 1-10. The reasoning is due to the reflections will cause noise. I may have misunderstood and need to review the video again.

43 minute video, so I only watched the first few seconds.  I never heard anything about harmonics.  As tggzzz said, reflections are bad without worrying about multiple reflections.  Getting rid of the first round of reflections is a matter of impedance matching.  If you don't have the first round it doesn't matter what your trace length is since you won't get a second or third round trip. 

The video starts off with a really horrible example.  If I have time later I might watch the video to see how he made it look so bad.  It's also interesting that his data rate is not very high compared to the round trip delay.  Yours will be the opposite, high data rate and relatively longer path delay depending on your board size.

[nctnico]

My manager just told me I will be in charge of a 5Ghz bandwidth ethernet passthrough board. He says it's just traces. So everyone knows I'm a fairly green engineer, my pcd experience is under 10 Mhz. Mostly arduino like designs with microchip and atmel 8 bit micros. My question for everyone should I be freaking out or is there books/videos I should binge read or watch? How do I give myself a crash course in GHz pcb layout? Is it just a manner of controlling the trace impedance and stay away from harmonics 1-10? Any wisdom or suggestions is appreciated. I look forward to this learning experience and I hope I don't hit a brick wall when running into this.

Basically: controlled impedance and no stubs. I recommend to use a field solver (like Sonnet Lite) to model the geometry including the solder mask. Online tools are usually way off.

[basinstreetdesign]

Get this book:  A Handbook of Black Magic by Howard Johnson and Martin Graham
(it's available on Amazon dot com)

Take a weekend and read it from cover to cover.  Then read the sections you are most worried about.  It will open your eyes.

[Bassman59]

Get this book:  A Handbook of Black Magic by Howard Johnson and Martin Graham
(it's available on Amazon dot com)

Take a weekend and read it from cover to cover.  Then read the sections you are most worried about.  It will open your eyes.

ABSOLUTELY.

And get the boss to pay for it!