At bandwidths beyond 2.5 Gbps, it is critical to accurately characterize connector and BGA via field footprints as well as AC coupling capacitor launches. This is an example of a backplane connector via field which was modeled in Ansoft HFSS. In this case, four pairs were modeled and an .S16p file was created. These models are concatenated in a full serial channel which include table models of the differential trace geometries.

There is a lot of discussion today about how to ensure that backplane serial links will function at extremely low BER (bit error rates). The IEEE 802.3ap specification is a work in progress but it provides the most guidance in terms of frequency domain requirements and compliance. At these bandwidths, HSPICE simulations of eye diagrams in the time domain are relatively useless. This slide shows the insertion loss (SDD21) of a 6.25 Gbps backplane channel as well as the total crosstalk (FEXT & NEXT) generated in the via fields and connectors. You can see that the insertion loss (in blue) does not quite meet the IEEE mask requirement (in red). As such, a different material or different trace geometry could be recommended and then verified in simulation. The total noise is relatively low. Note that there is a 26.8 dB signal-to-noise ratio at 3.125 GHz (6.25 Gbps).

The deviation in the insertion loss is important in terms of the effectiveness of the receiver equalizer in cleaning up the signal. This slide shows that this particular channel insertion loss deviation is well within the high confidence area.

The ICR (Insertion Loss Crosstalk Ratio) is another way of looking at the signal-to-noise ratio as described in the IEEE 802.3ap spec. The diagram shows that at 3.125 GHz (6.25 Gbps) the signal to noise ratio is right about at the spec limit. This limit was recently changed in the spec and is significantly more stringent than it was even 1 year ago.

The return loss (SDD11) of the backplane channel is a measure of energy transfer through the link. This link is well within the IEEE return loss requirement up to just over 5 GHz.

This is a measured eye pattern at 6.25 Gbps overlaid with an HSPICE simulation. Normally, we do not recommend time domain simulation above 2.5 or 3 Gbps. A frequency domain analysis of the passive channel yields a much better picture of expected performance, however, with appropriate device and channel models good eye correlation can be achieved. Noise was not modeled in this case.

This slide shows the impulse response of a 5 Gbps channel. The impulse response provides a quantitative picture of how noise and energy are dissipated in the passive channel. Ideally, the pulse should die out smoothly -- as it does in this case. StatEye was used to generate this impulse response plot.