Issue link: https://iconnect007.uberflip.com/i/1004549
JULY 2018 I DESIGN007 MAGAZINE 69 is already substantial at 5 GHz (about 20%) and may be totally un-acceptable at 30 GHz (about 50%). So, why do we see such discrepancies between the measurements and models? For the losses, it is quite obvious: We do not have any data at the stackup exploration stage to specify conductor roughness models. If such models are used to compute compliance met- rics, it can lead to complete design failure! Note that this is not just a problem related to losses; it also affects such metrics as the inser- tion loss to power sum crosstalk metric (ICR). Considering differences in the impedances and delays, it is either differences in the geometry of the cross-sections or in the parameters of the dielectrics or both. We will find out soon. To make it more complicated, note that the conductor roughness increases the inductance of the traces and can substantially change both the impedance and phase delay in addition to the losses [3] . All that must to be included in the transmission line model. Also, simple adjustment of the model parameters (geom- etry or dielectric properties) will not work; too many parameters to play with and a systematic approach is needed. Fortunately, with Simbeor software the parameters of the dielectric and conductor roughness models can be separately identified in the systematic way. It can be done with the S-parameters mea- sured for two-line segments and converted into either GMS-parameters, or into complex propagation constants (Gamma). Both GMS- Figure 2: TDR of 10 cm differential links measured and computed (left graphs) and GMS insertion loss (IL) and phase delay for 5 cm differential line segments measured and computed (right graphs) with the stackup structure and material parameters from the PCB manufacturer for layer INNER 1 (top graphs) and for layer BOTTOM (bottom graphs).