This article studies the fiber weave effect on PCB in terms of mode conversion and differential insertion loss due to the disparity of propagation delay and characteristic impedance in single ended mode...
PCB dielectric substrate is composed of woven fiber glass that is strengthened by epoxy resin. The microscopic view of fiber weave on PCB and glass dimension for commonly used weave patterns are depicted in Fig. 1. The thick lines in light brown color are fiber glass, while the square columns in black color are the glass pitch filled with epoxy resin. Among the three weave patterns, 106 has the largest pitch or lowest glass density, while 3313 has the smallest pitch or highest glass density, in both the horizontal and vertical axis.
Fig. 1. Microscopic view of fiber weave on PCB and glass dimension for commonly used weave patterns (Isola)
Fiber glass material has dielectric properties that differ very much from the properties of epoxy resin. For instance, NE-glass has dielectric constant (εr) and loss tangent (Df) 4.5 and 0.001 respectively. Meanwhile, epoxy resin has εr 3, which is very much different versus fiber glass. When PCB substrate with lower glass density is used, traces of the differential signals could cross different regions of resin and fiber glass more frequently. As a result, the propagation delay of the signal in single ended mode changes frequently along the path from transmitting to receiving end. Moreover, the trace impedance in single ended mode also encounters disparity. The relationships are governed by Eq. (1) and (2) respectively.
This phenomenon poses a great challenge to the multi-Gigabit differential signal transmission. The trace of non-inverting signal (P-line) could be routed on fiber glass, while the trace of inverting signal (N-line) could cross many resin regions, i.e., non-homogeneous substrate condition. As a result, due to the consistently changing in propagation delay experienced by the N-line, the phase difference between P-line and N-line in common mode could be much less than 180o at the receiving end. The large extent of skew or misalignment between the rising and falling edges of the common mode signal waveform shrinks the eye diagram opening. Furthermore, the non-homogeneous substrate condition also results in the disparity of the trace impedance between P-line and N-line, which in turn intensifies the differential insertion loss of the transmission channel. Ultimately, high bit error is experienced at the receiving end.
The fiber weave effect of different weave patterns on characteristic impedance, differential insertion loss (Sdd21), eye diagram and differential to common mode conversion (Sdc21) is analysed with three-dimensional electromagnetic (3DEM) simulation using Keysight EMPro.
Analysis and Results
The differential microstrip with fiber weave pattern 106, 1080 and 3313 modelled using Keysight EMPro are depicted in Fig. 2. For each weave pattern, the P and N-line traces have 5 mil width, 10 inch length and 1.2 mil thickness, intra-pair spacing 10 mil, routed 3 mil above the ground plane. The fiber glass and resin in between the traces and ground plane are shown in green and white colour respectively. The glass and resin are modelled according to the dimension specified in Fig. 1. The port of simulation is set up at each end of the microstrip trace, followed by 4-port S-parameter extraction.
Fig. 2. Top view of differential microstrip with fiber weave pattern 106, 1080 and 3313 modelled using Keysight EMPro
With reference to the impedance plots depicted in Fig. 3, the microstrip with weave pattern 106 (i.e., largest pitch) encounters the longest cumulative non-homogeneous condition, hence largest disparity of trace impedance between P and N-line, i.e., 4 ohm. Meanwhile, the microstrip with weave pattern 1080 (i.e., moderate pitch) encounters shorter cumulative non-homogeneous condition versus pattern 106, hence smaller disparity of trace impedance between P and N-line, i.e., 1 ohm. On the other hand, the microstrip with weave pattern 3313 (i.e., smallest pitch) experiences minimal non-homogeneous condition, hence balanced trace impedance between P and N-line.
Fig. 3. Trace impedance for P and N-line with weave pattern 106, 1080 and 3313
The unbalanced impedance and propagation delay between P and N-line due to the disparity of εr in the non-homogeneous environment, intensifies the Sdd21, as shown in Fig. 4. Sdd21 experienced by the transmission line with the three weave patterns is similar up to 500 MHz. Beyond 500 MHz, microstrip with weave pattern 3313 experiences the lowest Sdd21, which is contributed by the attenuation loss of the fiber glass. Meanwhile, microstrip with pattern 106 and 1080 face the larger Sdd21 versus pattern 3313, due to the extra loss caused by the non-homogeneous environment of the fiber weave. The worst Sdd21 is encountered in pattern 106, due to its largest impedance disparity, contributed by its longest pitch and cumulative non-homogeneous condition.
Fig. 4. Sdd21 plots for 10-inch microstrip with weave pattern 106, 1080 and 3313
Subsequently, the S-parameter model of the 10-inch microstrip with each weave pattern is imported to the channel simulation topology using Keysight ADS, depicted in Fig. 5 to analyse eye diagram in differential mode and propagation delay in common mode at 1 Gbps (i.e., lower speed grade) and 5 Gbps (i.e., higher speed grade) respectively. At both speed grades, a non-return to zero (NRZ) differential square wave signal with 80 ps rise/fall time and 1 Vpp amplitude is injected at the transmitting end of the S-parameter model, while the eye diagram and propagation delay are analysed at the receiving end of the model. In this simulation with 1000 data bit being transmitted, the pre/de-emphasis and equalization functions are disabled.
Fig. 5. Channel simulation topology to analyse eye diagram and propagation delay
With reference to the time domain analysis results at lower speed grade of 1 Gbps (i.e., Nyquist frequency 500 MHz) shown in Fig. 6a, 6b and 6c respectively, pattern 106 faces the largest unbalanced propagation delay between P (i.e., red waveform) and N-line (i.e., blue waveform), indicated by condition whereby the rising and falling edges of the waveforms cross at +/- 0.15 V, i.e., the level further away from 0 V. On the contrary, in pattern 3313, the rising and falling edges of the waveforms cross at 0 V, indicating a balanced propagation delay between P and N-line. However, the eye diagram for each weave pattern has the similar height, i.e., ~ 0.92 Vpp, as the Sdd21 at 500 MHz experienced by each pattern is the same, i.e., -1.1 dB, as shown in Fig. 4.
Fig. 6a. Eye diagram (left) and propagation delay (right) at 1 Gbps with pattern 106
Fig. 6b. Eye diagram (left) and propagation delay (right) at 1 Gbps with pattern 1080
Fig. 6c. Eye diagram (left) and propagation delay (right) at 1 Gbps with pattern 3313
The fiber weave effect becomes more significant beyond 500 MHz, indicated by the eye diagram in Fig. 7a, 7b and 7c respectively, at higher speed grade of 5 Gbps (i.e., Nyquist frequency 2.5 GHz). The mid-point eye diagram height for weave pattern 3313 is the largest, i.e., 0.75 Vpp, versus the smallest height for pattern 106, i.e., 0.5 Vpp. This phenomenon is caused by the extra 3 dB attenuation at 2.5 GHz experienced in pattern 106, as shown in Fig. 4.
Fig. 7a. Eye diagram (left) and propagation delay (right) at 5 Gbps with pattern 106
Fig. 7b. Eye diagram (left) and propagation delay (right) at 5 Gbps with pattern 1080
Fig. 7c. Eye diagram (left) and propagation delay (right) at 5 Gbps with pattern 3313
Subsequently, Sdc21 for each weave pattern is analysed, plotted in Fig. 8. A smaller Sdc21 magnitude indicates a tougher differential to common mode conversion, i.e., better immunity of a differential signal pair against common mode crosstalk. The Sdc21 across the wideband experienced in pattern 3313 is at least 20 dB lower versus the other two patterns, hence less prone to common mode crosstalk.
Fig. 8. Sdc21 plots for 1-inch microstrip with weave pattern 106, 1080 and 3313
In this article, the fiber weave effect with three commonly used patterns in the PCB industry is compared. For PCB transmission channel design with high speed grade at 5 Gbps and beyond, weave pattern 3313 with minimal glass pitch shall be applied in the substrate to minimize the attenuation and unbalanced propagation delay, which in turn strengthens the signal integrity.
 E. Bogatin, “Signal Integrity – Simplified (1st Edition)”
 Isola – Understanding Glass Fabric, https://www.isola-group.com/wp-content/uploads/Understanding-Glass-Fabric.pdf
 EverythingRF – Impedance Calculator, https://www.everythingrf.com/rf-calculators/microstrip-impedance-calculator
— Chang Fei Yee is Technical Lead in Hardware & SI/PI at Keysight Technologies