Issue link: https://iconnect007.uberflip.com/i/258943
February 2014 • The PCB Design Magazine 21 are required. Procedure with GMS-parameters has minimal number of smooth complex func- tions to match during the identification pro- cess. Specifically, one S-parameter for single and two S-parameters for differential lines have to be matched. All reflection and modal transfor- mation parameters are exactly zeroes. Identified models are frequency-continuous and models (Eq. 2–4) are not restricted to the frequency band used in the identification process—they are naturally extendable above the upper fre- quencies and below the lower frequencies. Practical Example As an example of material parameters iden- tification up to 50 GHz (for 25–50 Gbps data channel) we use measured data provided by Wild River Technology for CMP-28 channel model platform validation board made with Isola FR-408 materials and regular copper. The board and stackup are shown in Figure 2. Five points for dielectric constant and loss tangent are available from the datasheet for the FR-408 material, but the points are measured with different methods and the maximal fre- quency is only 10 GHz. Frequency-continuous multi-pole Debye model cannot be accurately defined up to 50 GHz with those data. No data for the conductor roughness was available from the board manufacturer. To identify dielectric and conductor roughness model parameters, we can use 2-inch and 8-inch single-ended strip- line links in layer L03 (see stackup in Figure 2). S-parameters measured for two line seg- ments are shown in Figure. 2. Following the procedure outlined in the previous section, we first estimated measured S-parameters quality in Simbeor Touchstone Analyser tool 8 . As we can see from Figure 3, the quality of S-parameters for the stripline seg- ments is excellent (final quality metric is close to 100%). TDRs of both segments have been also evaluated and both links looked consistent (impedance and discontinuities are close). After measured S-parameters for test structures were pre-qualified, we convert the measured reflec- tive S-parameters into generalized or reflection- less S-parameters shown in Figure 4 (red and blue curves). Dielectric specifications (see insert in Figure 2) show that this dielectric has dielectric con- stant (Dk) 3.66 and loss tangent (LT) of 0.0117 at 1 GHz (the other points are also consis- tent with that value and WD model). We can use that point to define the wideband Debye model (Eq. 2). If we compute GMS-parame- ters for 6-inch segments with the electromag- netic analysis with wideband Debye model and Dk=3.66 and LT=0.0117 defined at 1 GHz (shown with blue curves in Figure 4), the dif- ference in the measured and computed group delay is relatively small, but the difference in GMS insertion loss is large, up to 25% as shown in Figure 4. PCB AND PACkAGING DESIGN UP To 50 GHz continues feature Figure 3: Measured S-parameters for two stripline links and quality evaluation.