4. A system engineer faces a multitude of options when
designing a high-speed backplane channel.
- Among the available options, different board materials and
backdrilling options for both daughter cards and backplane as well
as different connector types might result in different performance.
- On top of that, different crosstalk configurations and board
thicknesses might be pre-defined by mechanical limitations.
- Today, we will address how these variables impact the
performance of different channel lengths running different data
rate.
We will also present a methodology on how to evaluate
a large set of variables.
6. In total, 1536 scenarios were simulated.
- Each channel was obtained by cascading 9 different components:
• Trace on daughter card 1, via on the daughter card 1, connector 1, via 1 on
the backplane, trace on the backplane, via 2 on the backplane, connector 2,
via on daughter card 2, trace daughter card 2.
The via footprint and the connectors were simulated
using Ansys HFSS. The traces were simulated using
Ansys Q2D. An in-house Matlab-based software was
used to cascade the components.
Trace daughter
card 1
Connector 1
Via daughter
card 1
Via 1 on backplane Trace on backplane Via 2 on backplane
Trace daughter
card 2
Connector 2
Via daughter
card 2
7. Once the channel is obtained, channel operating
margin (COM) version 11/2014 was used as the metric.
For more on COM:
- http://www.ieee802.org/3/bj/
- R. Mellitz, et. al., “Channel Operating Margin (COM): Evolution of Channel Specifications for
25 Gbps and Beyond”, DesignCon 2013.
- X. Dong, et. al. “Relating COM to Familiar S-Parameter Parametric to Assist 25Gbps System
Design,” DesignCon 2014.
- D. Correia, et. al, “What Makes a Good Channel? COM vs. BER Metrics” DesignCon 2015.
- More this week…
8. For each channel, the COM number was evaluated with
two different configurations.
Parameter Setting Units Information
f_b 25.78125 GBd
f_min 0.05 GHz
Delta_f 0.01 GHz
C_d [2.5e-4 2.5e-4] nF [TX RX]
z_p select [1 2] [test cases to run]
z_p (TX) [12 30] mm [test cases]
z_p (NEXT) [12 12] mm [test cases]
z_p (FEXT) [12 30] mm [test cases]
z_p (RX) [12 30] mm [test cases]
C_p [1.8e-4 1.8e-4] nF [TX RX]
R_0 50 Ohm
R_d [55 55] Ohm [TX RX]
f_r 0.75 *fb
c(0) 0.62 min
c(-1) [-0.18:0.02:0] [min:step:max]
c(1) [-0.38:0.02:0] [min:step:max]
g_DC [-12:1:0] dB [min:step:max]
f_z 6.4453125 GHz
f_p1 6.4453125 GHz
f_p2 25.78125 GHz
A_v 0.4 V
A_fe 0.4 V
A_ne 0.6 V
L 2
M 32
N_b 14 UI
b_max(1) 1
b_max(2..N_b) 1
sigma_RJ 0.01 UI
A_DD 0.05 UI
eta_0 5.20E-08 V^2/GHz
SNR_TX 27 dB
R_LM 1
DER_0 1.00E-05
KR4 25Gbps
Parameter Setting Units Information
f_b 30.45 GBd
f_min 0.05 GHz
Delta_f 0.01 GHz
C_d [2.5e-4 2.5e-4] nF [TX RX]
z_p select [1 2] [test cases to run]
z_p (TX) [12 30] mm [test cases]
z_p (NEXT) [12 12] mm [test cases]
z_p (FEXT) [12 30] mm [test cases]
z_p (RX) [12 30] mm [test cases]
C_p [1.8e-4 1.8e-4] nF [TX RX]
R_0 50 Ohm
R_d [55 55] Ohm [TX RX]
f_r 0.75 *fb
c(0) 0.62 min
c(-1) [-0.18:0.02:0] [min:step:max]
c(1) [-0.38:0.02:0] [min:step:max]
g_DC [-12:1:0] dB [min:step:max]
f_z 7.6125 GHz
f_p1 7.6125 GHz
f_p2 30.45 GHz
A_v 0.4 V
A_fe 0.4 V
A_ne 0.6 V
L 4
M 32
N_b 16 UI
b_max(1) 1
b_max(2..N_b) 0.2
sigma_RJ 0.005 UI
A_DD 0.025 UI
eta_0 5.20E-08 V^2/GHz
SNR_TX 31 dB
R_LM 0.92
DER_0 3.00E-04
KP4 56Gbps
9. Three channel lengths: 0.3m, 0.5m and 0.7m
Backplane lengths: 0.2/0.3/0.3 m
Daughter card lengths:
0.05/0.1/0.2 m
Daughter card lengths:
0.05/0.1/0.2 m
10. For each channel length and COM configuration, a set
of 7 variables was simulated. Each variable had 2
possible values.
Daughter card
thickness
Daughter card
material
Daughter card
stub
Backplane thickness Backplane material
Backplane stub
Connector type
11. Each variable will have two values: “0” (premium) and
“1” (standard)
Variable Value 0 Value 1
DC material er=3.46, tand=0.005 er=3.9, tand=0.01
DC thickness 118 mils (3mm) 200 mils (5.1mm)
DC stub 12 mils (0.3mm) 18 mils (0.46mm)
BP material er=3.46, tand=0.005 er=3.9, tand=0.01
BP thickness 200 mils (5.1mm) 300 mils (7.6mm)
BP stub 12 mils (0.3mm) 24 mils (0.61mm)
Connector Molex connector “0” Molex connector “1”
12. For each channel and COM configuration, two crosstalk
configurations were considered: 8 FEXT and 3 NEXT
and 5 FEXT.
By including XT, we end up with a total of 8 variables. A
total 256 different combinations for each channel length
and COM configuration.
Victim
FEXT
NEXT
XT Config 1 (“0”) XT Config 2 (“1”)
14. Each variable will have two values: “0” (premium) and
“1” (standard)
Variable Value 0 Value 1
DC material er=3.46, tand=0.005 er=3.9, tand=0.01
DC thickness 118 mils (3mm) 200 mils (5.1mm)
DC stub 12 mils (0.3mm) 18 mils (0.46mm)
BP material er=3.46, tand=0.005 er=3.9, tand=0.01
BP thickness 200 mils (5.1mm) 300 mils (7.6mm)
BP stub 12 mils (0.3mm) 24 mils (0.61mm)
Connector Molex connector “0” Molex connector “1”
Crosstalk All FEXT 3 NEXT, 5FEXT
16. Suppose we have two variables (A,B) and each can
have a value “0” or “1”. Four combinations are possible.
- Assume each combination has a certain arbitrary metric value
(COM, cost). For example, (0,0)=4; (0,1)=-2; (1,0)=2; (1,1)=-4.
Notice the
values and
the gradient
17. Let’s add another 2 variables, C and D. Now we have a
total of 16 combinations.
- Once we see the real data, the matrix will be 8x8, resulting in 256
different COM numbers for each case.
Same as
before
-1 -4 -2 -6
0 -2 -2 -4
-2 -4 -2 0
4 2 -2 -1
Values:
26. The material used on the boards seems to be the most
important factor for the channel performance.
Connector choice defines an extra COM margin for a
given board material.
Crosstalk configuration did not have a large impact
since both connectors already had very low level of
crosstalk.
In the range used, stub did not have a significant
impact.
Thinner boards are better for shorter channels.
Similar analysis could be done for different variables.
