PCB Autorouter Question

[PottedMeat]

Here is a picture of the Xbox360 motherboard:

http://dwl.xbox-scene.com/pictures/XeDK-smartxx/smartxx-xbx360-45.JPG

Whats with all the squiggly routed bus signals from the GPU and Samsung RAM ( at least thats what i think it is )? I know the signals have to satisfy timing / integrity / RF requirements, but it seems strange the way they zigzag and go to and away from the ram before connecting. The CPU - GPU connection looks like it has differential pairs ( i think ) and looks like the most direct path. Are those signals the result of an autorouter being told ' hook these ICs up at some frequency at these locations'?

 

[jmcoreymv]

Usually they route the signals like that when they want it to match a certain timing spec or if it is a differential pair they need both signals exactly the same length.

 

[Bassyhead]

I've seen this kind of routing on many other things before, like RAM sticks and many motherboards. Take a look inside your computer.

 

[TuxDave]

Originally posted by: jmcoreymv
Usually they route the signals like that when they want it to match a certain timing spec or if it is a differential pair they need both signals exactly the same length.

That's what I think too.

 

[RaynorWolfcastle]

Originally posted by: jmcoreymv
Usually they route the signals like that when they want it to match a certain timing spec or if it is a differential pair they need both signals exactly the same length.

Yep, that's almost certainly it

 

[blahblah99]

Originally posted by: RaynorWolfcastle

Originally posted by: jmcoreymv
Usually they route the signals like that when they want it to match a certain timing spec or if it is a differential pair they need both signals exactly the same length.

Yep, that's almost certainly it

Yes, that's exactly what the squiggly lines are for...

During autorouting, you specify a group of nets and give it a constraint such as a length or timing constraint (absolute or relative to the group) and the autorouter will try to meet those constraints by adding squigglys, zig zags, mazes, etc etc... A good example is the databus, which can be 32-bit wide on some systems, just like in the image shown above.

Sometimes, during timing analysis, there may be a clock/enable/bus signal that needs to be delayed by a few hundred picoseconds in order to meet timing, so that single net might be constrained so delay is at least x picoseconds... this is one of the reasons why some motherboards are better at overclocking than others, everything else being the same - some have more timing margins than others. It is also a cheap and free way to delay a signal rather than adding additional components or going through an fpga, but becomes impractical for delays in the nanoseconds (electrons travel through pcb copper at roughly 150ps/inch).

 

[Peter]

That'll be length matching on parallel busses. You want all signals to arrive at their destination at the exact same time, and at the frequencies we're having today, trace length already does matter bigtime.

 

[arcas]

Circuit boards aren't the only place you see this. If you go to any particle accelerator facility and look at the detector electronics, you'll very often see coils of wires, sometimes dozens of feet. It's the same principle. Very often you need the signal from an two or more detectors registering the same event to arrive at precisely the same time (within a ns or so).

Sometimes the detectors themselves have inherent speed differences. This could be because the photo-multiplier tubes have different speeds or could be due to the physical properties of the detectors themselves. For example, a lead-glass Cherenkov detector generally takes longer to register an event than a scintillator detector. In addition, your detectors are usually different distances from your racks of detector electronics so that even if the detectors are identical, their signals would arrive at slightly different times.

You can install electronic delays which are useful if you need really really long delays but more often you'll just adjust your cable lengths so that the signals arrive simultaneously.

 

[The Boston Dangler]

Interesting. This shows that on a PCB, equal length traces are more efficient than in-chip adjustments. On any late-model car, the plug wires going from the electronic ignition box to the spark plugs are as short as possible. If you plug them in incorrectly, the computer will *try* to compensate for the difference.

 

[blahblah99]

Originally posted by: The Boston Dangler
Interesting. This shows that on a PCB, equal length traces are more efficient than in-chip adjustments. On any late-model car, the plug wires going from the electronic ignition box to the spark plugs are as short as possible. If you plug them in incorrectly, the computer will *try* to compensate for the difference.

Not necessarily...even if in-chip adjustments were perfect, the pinout on the packaging will be enough to cause unequal trace lengths to the receiver.

Ignition wires are completely different... they run on a much lower frequency, not to mention higher voltage.

 

[The Boston Dangler]

Originally posted by: blahblah99

Originally posted by: The Boston Dangler
Interesting. This shows that on a PCB, equal length traces are more efficient than in-chip adjustments. On any late-model car, the plug wires going from the electronic ignition box to the spark plugs are as short as possible. If you plug them in incorrectly, the computer will *try* to compensate for the difference.

Not necessarily...even if in-chip adjustments were perfect, the pinout on the packaging will be enough to cause unequal trace lengths to the receiver.

Ignition wires are completely different... they run on a much lower frequency, not to mention higher voltage.

The voltage potential wouldn't effect the need for accurate timing of voltage arriving at it's destination.

Spectral output from ignition wires can be anywhere from 10kHz to over 10GHz.

I found that here.

 

[rivan]

"Pictures removed on request of Microsoft Corporation"

Bleh.