PDF of this tutorialMaterial losses amalgamate with copper losses, giving the designer several choices in distributing losses throughout the system. For cost motives, changes in dielectric materials may be one of the last changes made to the system. Line widths are often one of the first changes to be made. Intuitively, 20-mil lines give less conductive loss than 6-mil lines. What is the quantitative difference, however, between different line widths? Using single-ended lines and VNA through measurements (S21), different materials were analyzed over a spectrum of four different line widths (6, 8, 12, and 20 mil). As different materials experience peaks and notches at different frequencies, and some materials were more broadband than others, a mean loss value was found for each test case by sweeping from 50 MHz to 20 GHz (see Figure 4).
Figure 4. Comparison of Mean Loss Values for S21 Measurements, Single-Ended Lines (0.5 m, layer 11)
This type of result gives a more easily analyzable result than overlaid test cases. Actual waveforms are included in the Appendix.
Moving from a 6-mil to an 8-mil line yields 1.3-dB improvement, which is 4.4 percent. Likewise, going from an 8-mil line to a 12-mil trace width yields 1.3-dB improvement, or 4.4 percent. Because 8-mil lines seem to be about at the point of diminishing returns from an electrical, manufacturable, and routable standpoint, 8- and 10-mil lines were selected for system simulations. A 6-mil to a 20-mil increase in line width yields an average of 4.2-dB improvement over all materials (14 percent). This is roughly the same improvement (as averaged over line width) seen by moving from FR–4 to Megtron material (3.95 dB or 13.2 percent). So, at some point in time, a slight change in material may yield more significant results than major changes in line widths. The article entitled “On the Dielectric Material Properties for Thin Film Integrated RF and Microwave Applications” points out that, at high frequencies, broad line widths may contribute to additional parasitic effects such as radiation and capacitive coupling.2
Some criticism for not impedance-matching all test cases materialized with the above S21 graph (see Figure 4). The number of test boards was limited to control cost and lead-time. In addition, keeping the boards at a reasonable thickness limited the number of layers per board. One line width was chosen to be impedance-matched for each trace topology. The 8-mil differential lines were matched to 100 Ohms, and the 12-mil, single-ended lines were matched to 50 Ohms. To achieve all the desired measurements, multiple line widths needed to reside on the same layer, causing impedance mismatches in many of the line widths.
Additional data was analyzed to ascertain the impedance mismatch’s effect on measured loss. The analysis plotted risetime degradation against impedance mismatch. The impedance mismatch was calculated by subtracting 100-Ohms (all are differential cases) from the measured impedance of the line. Theoretically, if the impedance mismatch outweighed the line width effect on the TDT’s risetime, the trend lines would form a V around the “0” impedance mismatch point. The trend lines clearly increase in risetime with respect to line width increases—not impedance mismatch. To prove this point further, 8-mil lines with varying degrees of impedance mismatch were plotted on the same graph. The risetime (10 to 90 percent) for all of these 8-mil lines remains approximately the same (see Figure 5).
Figure 5. Risetime Plotted against Impedance Mismatch
(for FR–4; all layers; 0.5-m differential; 6-, 8-, 12-, 20-mil Line Widths)
Based on the impedance mismatch data, the demonstrated improvements on the S21 graph shown in Figure 5 are due primarily to material parameters. Moving from FR–4 to a Nelco 6000/Polyclad LD621 material represents a 20.5-percent improvement. Moving from FR–4 to Rogers 4000 series is a 26.8-percent improvement, and going from FR–4 to Arlon is a 34.1-percent improvement. Although electrical improvement makes obvious decisions for material selection, cost, manufacturability, and mechanical properties of the material enter into a decision matrix for material selection.
2Pieters, P., S. Brebels, G. Carchon, K. Vaesen, W. DeRaedt, E. Beyne, and R.P. Mertens. "On the Dielectric Material Properties for Thin Film Integrated RF and Microwave Applications." Advancing Microelectronics, volume 26, number 5, page 22.
