Re: [BP] analysis addressing points from the 9/17 channel meeting
Following on on Shannon's thread, if anyone has any proposed models, please
forward to me so that I may include with my presentation for comparison of
data.
john
-----Original Message-----
From: owner-stds-802-3-blade@listserv.ieee.org
[mailto:owner-stds-802-3-blade@listserv.ieee.org] On Behalf Of SAWYER
Shannon
Sent: Wednesday, September 22, 2004 10:15 PM
To: STDS-802-3-BLADE@listserv.ieee.org
Subject: [BP] analysis addressing points from the 9/17 channel meeting
All,
Regarding the SDD11 mask discussion on the 9/17 channel telecon, it wasn't
clear to me or others the reason for the flatness of the SDD11 above 400MHz,
or the -12dB loss above 400MHz. I believe this addresses those two
questions.
I've spent some time analyzing 40in linecard-backplane-linecard
(6mil-4.75mil-6mil trace width) channel measurements and compared them to
measurements of my own cleanSMA-stripline-cleanSMA boards of the same
laminate in the frequency and time domains. My cleanSMA launch is a balanced
L and C of 6ohm (stimulated with a 30ps step). I brought the .s4p files into
ADS and stimulated them with differential step functions of various rise
times to see the transient effects. I found the impedance over time of the
backplane channel ramping up significantly between the discontinuities until
it settled at steady state.
Comparing the frequency and time domain plots, I've found that the flatness
of the SDD11 on the backplane channel response is there after ~2GHz, and it
is produced primarily by the non-frequency dependant DC losses of the long
4.75mil traces and secondarily by the reactance of the connector
discontinuities. Since the linecards were backdrilled, I suspect that some
of the excess L and C stems from the difference in conductor diameter to
trace width, or improper ground via stitch placement around the conductor or
trace.
To break apart these two issues I measured S Parameters of a cleanSMA ->
32in of both 4mil and 10mil stripline -> cleanSMA and found:
1. S21 of the 4mil trace was very similar in shape and loss
points(-22dB@5GHz) to the backplane's S21, but the S11 of the 4mil trace was
curved across the freq range (-30dB@50MHz to -0.5dB@15GHz)instead of flat
like the backplane, and the step response showed minor (4ohm) impedance
change due to DC loss. This showed that the small discontinuity allows you
to see the frequency dependant nature of the clean channel, but the S21 is
still not optimal since it is slightly degraded by small DC loss, and skin
effect.
2. S21 of the 10mil trace was also very similar in shape with lower loss
points(-15dB@5GHz), the S11 was similar (-35dB@50MHz to -0.5dB@15GHz) but
better than the 4mil trace, and the step response showed almost no impedance
change, indicating almost no DC loss across the 32 inches. This showed that
the smaller discontinuity still allows you to see the frequency dependant
effects, and the DC loss is minimized, so the impedance is better matched
across the whole 32in trace.
From these findings I can't propose a new mask limit for SDD21 or SDD11, but
(discounting crosstalk optimization, which must be addressed) I will offer
that interconnect improvement and increased trace width are significant
variables to get lower loss channels than what we've seen so far.
I've conversed with John D'Ambrosia about this, and he will be presenting
his SDD11 findings in Ottawa.
-Shannon
Shannon Sawyer
Agilent Technologies Inc.
ASIC Products Division shannon_sawyer@agilent.com
4380 Ziegler Road, MS72 (phone) 970-288-7855
Fort Collins, CO 80525 (fax) 970-288-6580
Shannon Sawyer
Agilent Technologies Inc.
ASIC Products Division shannon_sawyer@agilent.com
4380 Ziegler Road, MS72 (phone) 970-288-7855
Fort Collins, CO 80525 (fax) 970-288-6580