Issue link: https://iconnect007.uberflip.com/i/1004549
68 DESIGN007 MAGAZINE I JULY 2018 According to the manufacturer, the expected impedance variations should be within 8%. That is too large to expect excellent correla- tion up to 30 GHz for 28 Gbps NRZ links, but it may be acceptable. The board manufacturer provided stackup geometry as shown in Figure 1 on the left side, and corresponding stackup entered for the pre-layout analysis into Sim- beor software is shown on the right side. Meg- tron6 specs provide dielectric constant and loss tangent at multiple frequencies—just one frequency data can be used to define causal wideband Debye model. The values for Dk in the Figure 1 are slightly different from the Megtron6 specs and are provided by the PCB manufacturer based upon their experience with this material. The major problem here is with the conduc- tor roughness models—all we know that the copper foil roughness is specified as H-VLP and no other data. PCB manufacturer also rough- ens the shiny side of the copper foil during the board manufacturing, without any parameters for the electrical modeling. Even if we had data for the mate side of the copper foil from the copper foil manufacturer, the PCB manu- facturer treatment of the shiny side makes it practically useless. Thus, we start the stackup exploration without the conductor roughness model and with the trace adjustments provided by the PCB manufacturer. The rest of the EvR-1 validation board design is covered in detail in this paper [1] . To validate the preliminary data, we will use 10 cm differential links in strip layer INNER1 and microstrip layer BOTTOM (Figure 1). The results of the first experiment are shown in Figure 2. Left graphs show mea- sured TDRs—the response is computed with S-parameters measured up to 50 GHz. TDRs for 10 cm segments of the transmission line model computed in Simbeor are also shown on the same graphs for comparison. We can see acceptable TDR correlation for the strip line, but computed impedance of the microstrip line is substantially lower. The models for these preliminary comparisons do not have the launches. Right graphs in Fig - ure 2 show correlation between the general- ized modal S-parameters (GMS-parameters) measur ed and computed for 5 cm segment of the differential lines. GMS-parameters are reflectionless transmission parameters; the reflection losses are completely removed in Simbeor software. That makes this type of S-parameters ideal for precise material param - eters identification and model quality evalua- tion. From GMS-parameters we can observe that the model delays are off by less than 2 ps/inch. The most important is the obvi - ous difference in the losses—the difference Figure 1: The EvR-1 validation board stackup from the PCB manufacturer (left) and the initial material models, and exactly the same stackup in Simbeor software defined for pre-manufacturing analysis (right).