Sometimes social networks (in vias) are not a good thing
1/3/2011 5:59 PM EST
"I expect to see far end noise in microstrip lines, but there's not supposed to be any far end noise in stripline," Jeff Loyer lamented. Yet, he and his team saw a strong far end cross talk signature in their measured data from long uniform test lines. An example of the far end noise measured between pairs of stripline and microstrip traces is shown in Figure 1.
Figure 1. Measured far end for microstrip (in red) and stripline traces (in blue) for the same rise time and same coupling, and same length pairs of lines.
What could be causing this far end noise problem in stripline? Was it from the board fab, the materials, or the board design, he wondered?
As a signal integrity engineer with Intel's server division, one of Jeff's roles is to develop test methods to evaluate new designs, materials and board fab suppliers. A standard test structure he and his team use is a pair of long, uniform, coupled transmission lines from which they get loss, singe ended and differential impedance and line to line skew, for example. The differential pair structures are probed from their ends using microprobes.
In a recent batch of boards, many striplines showed this far end noise problem. An obvious possibility was non uniform materials distribution in the stack up. The stripline signal line had a layer of core below it and pre-preg above it. If the dielectric constants of these materials were radically different from each other, this could cause the far end noise. But there was so much noise. Was this an indication of a bad batch of core or pre-preg?
This is where the detective skills of a good engineer come in. With a four port measurement, whether measured in the time domain or the frequency domain, there are 72 different, unique S-parameter elements, spanning time domain, frequency, single ended, differential, as step response or as impulse response. With so much data, it's important to be selective and data mine the most important terms.