Signals with curved routing...

This is of course a more complex question then a simple answer. A good article on the rounding bends is from Howard Johnson called "Who's Afraid of the Big, Bad Bend?" that can be found at the following url (http://www.sigcon.com/Pubs/edn/bigbadbend.htm). The bottom line is that a right angle bend will have some parasitic capacitance that does effect the signal integrity. Howard Johnson argues that this is significant until we get until we get to 10Gbps. I have seen designs that are worried about the accumulated effect of issues such as right angle bends even down to the 1-2Gbps range. I focus on the right angle bend versus just the curved to address the issue that the curve primarily addresses, the parasitic capacitor. The majority of this parasitic capacitor can be eliminated by using two 45 degree bends instead of a 90 degree bend. Studies have been done (I can't find the reference) that show that by curving versus using a 45 degree show a minimal gains in signal integrity at the sub 10Gbps range. Again, some designers will curve to gain all the margin they can. I have found that curving creates additional challenges. Since adding the curve is an additional step, if you have to change the route during the layout process, it adds additional time re-curving the signals each time there is a change. Second, I find most length balancing algorithms don't correctly calculate the length of a curve. This means that if you are balancing the differential pair very tightly, adding curves might throw off the calculations and you might actually aggravate the signal integrity issues instead of helping them. I would be interested if any else has information on the accuracy of the current length calculating algorithms on curves. The bottom line for digital designs is that it is not the difficult to use 45 degree bends versus 90 degree bends and you gain most of the signal integrity gains versus going full curving. If you feel you need to curve due to tightening margins or very high frequencies, be aware that it will take more layout time and potentially complicate balancing of the signals.

C Shiva said:

And why we can use ground shielding for this signals like a clock signal?

 

Good question.  Appears to be a rule applied for no engineering reason.  Use a field solver to see its effects.  Either it screws with the impedance of the line or it has no function.  The bends appear to be applied for a similar reason: none.  Way too many bad rules out there when just a few good ones are needed.

hi, just wanna share my ideas about this issue. In our design we curve
lines because of the some manufacturing issues. 90 deg bends have the
tendency of over etching (clot of etch in the corners) and may short
other signals near to it. We use ground shield for differential pairs in
order to isolate them to noise (crosstalk) since these are high speed
signals. usually we do this on LVDS, USB, SSTL, DDR signals (this
unwanted noise will be terminated to GND). As a result, signal integrity
can be achieved. hope this can help. good day, eric

redwire wrote:
>
> *C Shiva:*
> And why we can use ground shielding for this signals like a clock
> signal?
>
>
>
> Good question. Appears to be a rule applied for no engineering
> reason. Use a field solver to see its effects. Either it screws with
> the impedance of the line or it has no function. The bends appear to
> be applied for a similar reason: none. Way too many bad rules out
> there when just a few good ones are needed.
>
>
>
>
> You received this email because you subscribed to notifications for
> the Cadence PCB Design Forum. To unsubscribe, log in, go to forums,
> and change your forum subscriptions.
>
> http://www.cadence.com/community/forums/ForumSubscriptions.aspx