Electrical Performance of Microvia PCB Materials for WLAN and RF Module Applications Gregg S. Wildes, Ph.D. Are Bjorneklett Ph.D. W. L. Gore & Associates Ericsson Mobile Communications Elkton, MD, USA Lund, Sweden Abstract Conventional PCB processing of organic dielectrics has been proven to be a very economical interconnect solution for a variety of PCB, module, and packaging applications in the telecommunications and computing markets. Organic microvia dielectrics are well suited for mobile wireless as well as telecom and computing infrastructure products where the latest fabrication techniques, such as high-speed laser microvia drilling, are used to create the cost effective high density circuit boards found in these systems. For high speed/high frequency designs, epoxy resin coated copper foil often cannot meet electrical performance requirements while traditional high speed materials are often too expensive. However, substrates based on FR4 core constructions utilizing low loss microvia dielectrics can offer low cost, high performance solutions for these applications. This paper presents experimental results of electrical testing conducted on typical microstrip PCB and module constructions. The following quantitative, measured electrical data is presented. 1. Digital signal integrity of microstrip transmission lines 2. RF insertion loss of microstrip lines 3. RF performance of integrated passive, band-pass filter elements 4. Simulation of RF band-pass filter elements based on the measured data The effects of surface finish (Ni/Au - with and without), microvia dielectric (RCC, FR4 and low loss), and solder mask (with and without) are examined. Key words: HDI, microvia, high speed, high frequency, integrated passives, buried passives, RF filter, microstrip, eye pattern, insertion loss, dielectric, RF module, WLAN, IEEE802.11, Wi-Fi, Bluetooth, transceiver, controlled impedance Background High-density module substrates and PCB requirements are driving the need for high density interconnect (HDI) design capabilities using laser microvias (note that HDI, high density, and microvia are often used synonymously when describing these types of designs). The advantages of HDI designs include lower layer counts, smaller PCBs, lower costs, improved electrical performance, thinner constructions, and lighter weight end products 1-2. An additional benefit for PCBs using HDI outerlayers is the ability to incorporate high density, high performance area array packages to further increase the performance of your WLAN card, router, base station, server, or wireless PDA/phone. Many of these applications require both RF and digital circuitry on the same (HDI), multilayer circuit board or module. WLAN has been one of the true bright spots in the recent economic downturn. Global demand for WLAN shipments had been projected at 15M units in 2002 growing to 45M in 2005 3,4. However, recent numbers have shown that actual 2002 shipments were closer to 25M, more than doubling the number of units sold in 2001, and some are now projecting 2005 shipments to surpass 100M units 5. Clearly, in the current economic conditions, there are not many market segments that are showing sustained growth rates of 50-100% like WLAN. One of the greatest factors affecting the adoption rate of new technologies such as WLAN, Infiniband, Bluetooth and others is acceptable pricing. At the same time, higher bandwidth and greater operating range require improved electrical performance while needs for smaller product sizes present additional constraints upon circuit designers and substrate manufacturers. Electrical Performance Challenges Increased bandwidth and operating frequencies will continue to drive needs for increased signal integrity which in turn places added value on the electrical performance of PCB and substrate materials 3-6. Controlled impedance designs require greater
thickness uniformity, minimal variation of dielectric constant and loss over the operating frequency range, and thinner dielectrics due to reduced trace widths in high-density applications. A homogeneous dielectric (laminate or prepreg) will also provide a more uniform medium in which the electrical field will travel. Total signal losses are comprised of losses due to both the copper conductor and the dielectric. However, as frequencies increase above 1-2 GHz, such as in many routers, servers and RF applications like WLAN, dielectric losses begin to dominate. Figure 1 - Commercial WLAN card operating at 2.4 and/or 5.2 GHz RF frequencies Therefore, designing with a low loss dielectric can minimize the total loss of a given design. A low Dk material enables the use of wider traces for the same impedance, maximizing conductor as well as dielectric losses. Low loss and low DK materials provide superior signal integrity for digital signals over a given length of circuitry and minimize power losses for RF applications. High-Volume manufacturing Challenges Microvia PCBs and substrates are manufactured in large panel, usually 457 mm X 610 mm (18 X 24 ) or larger, format using laser processing, predominantly CO and YAG 7. Some of the critical characteristics of HDI dielectrics that facilitate low cost, high-volume production include laser via drilling speed, homogeneity of the material and lamination cycle. Some of these material attributes, as well as electrical performance values, are shown for common PCB materials in the following table. Table 1 - Common microvia PCB materials, a comparison of electrical, physical, processing and economic properties Measure FR4 Epoxy RCC MICROLAM 630 GETEK PTFE Dk (10 GHz) 4.0 3.4 2.6 3.9 2.5 Df (10GHz) 0.025 0.028 0.004 0.015 0.002 Layer 1-2 (50 Ohm) thickness vs. FR4 1X 0.9X 0.7X 1X 0.7X Laser microvia drill throughput 1X 2.5X 3X 1X 1X Tg ( C) 150 150 >200 170 >200 Insertion loss vs. FR4 1X 1X 0.6X 0.9X 0.5X Cost Low -------------------------------> High Laser drilling speed is a function of the dielectric composite material composition. Organic materials (epoxy, BT, and PTFE resins for example) laser drill significantly faster than glass based materials. Faster laser drilling throughput can translate directly into cost savings by increasing the capacity of the current laser drills at a given fabricator and reducing operating expenses. The relative homogeneity (glass weave, ablative match of resin and reinforcement, etc.) of the material will also be important in establishing microvia hole quality and the variability of the laser drilling process. For example, five laser pulses may be enough to ablate through a resin rich area of FR4, yet may not ablate through a glass rich area (a knuckle), leaving an open. For all applications using HDI dielectrics, materialprocess interactions such as via taper angle, wicking, and dielectric cracking can all affect the density capability of a fabricator. Nan Ya engineers presented a useful, detailed analysis of these and other via quality issues, 7 Items of Laser Drilling Quality, during a recent Taiwan Printed Circuits Association (TPCA) meeting. Test Board Stack-up and Processing As seen previously in table 1, there are many material choices for these types of boards, but only the three most cost effective options were chosen for further evaluation. Four layer (1/2/1) and six layer (1/4/1) (notation meaning one microvia/four core/and one microvia layer from top to bottom) boards were manufactured in different versions. All versions used a FR-4 core with buried and laser blind vias. Resin coated copper foil (often called RCF or RCC ), FR4 prepreg or MICROLAM 630 dielectric layers were laminated on each side of the core as shown below.
Microvia layer Buried via FR4 Core Figure 2 - Typical 1/4/1 microvia PCB construction Additional boards were fabricated with and without soldermask as well as with and without electroless Nickel Gold (ENIG or Ni/Au). A comparison of the processing of the microvia dielectric materials detailed in this paper ais shown in table 2. Table 2 - PCB fabrication conditions for this study Process Parameters RCC Microlam 630 FR4 Microlam 630 Temperature (C) 185 185 185 185 Pressure (N/cm2) 180 200 180 200 Press time (min) 140 160 140 160 Vacuum Yes Yes Yes Yes Trace width (um) 125 125 225 200 Dielectric thickness for 50 75 57 125 86 Ohms (um) Laser microvia throughput Good Excellent Poor Good Laser via quality (YAG/CO2) Excellent Excellent Good Excellent Digital/RF Test Methodology and Transmission Line Measurements High frequency build-up constructions were examined with the above-mentioned material variables using test boards. The design used for the evaluation included microstrip transmission lines and a five-element L-C high-pass filter. The circuit block was manufactured in many different variants, which simplified design optimization. Test boards were fabricated to evaluate the relative electrical performance of the three dielectric materials (RCC, MICROLAM 630 and FR4) as well as solder mask and ENIG. Electrical testing was done using a 500 mv 2^7-1 PRBS test signal from an Advantest D3186. This was input to a single board of each of the different material types at both 5 and 10 Gb/s and the eye patterns were measured using a Tektronix CSA 8000. Measurements of insertion loss for RF applications were done using an Agilent 8720E network analyzer over a frequency range of 50 MHz to 10 GHz. Testing was done using edge launch PCB connectors. Microstrip lines of 25 and 200 mm in length were both tested, thus allowing connector effects to be quantified. The S21 data was obtained by using a Cascade Matrix test method developed at W. L. Gore. The method is similar to a network analyzer TRL calibration. The Cascade Matrix method assumes repeatable connections, requires two samples of different lengths for the same test condition, and more accurately accounts for connector losses than the commonly used stored trace divide method. The Cascade Matrix method has been found to be a useful tool for removing the effects of connector loss from the data. This method achieves good results based on one important assumption, that the connection to the two different samples is the same. Based on this assumption, the Cascade Matrix Method is performed as follows: 1. The uncorrected scattering parameters of a short and long sample of the same material are measured. 2. The S-parameter matrix is converted to a Cascade matrix. 3. Using linear algebra the propagation constant is extracted. 4. S21 for the long sample minus connector effects is charted. The digital eye patterns can be seen in figure 3.
FR4-125 um FR4-125 um Microlam630-86 um Microlam 630-86 um 210 um traces RCC - 80 um Microlam630-57 um 125 um traces Figure 3 - Eye patterns for 200 um long, 50 Ohm traces comparing 210 um wide traces (top two patterns) and 125 um wide traces (bottom two patterns); results are for RCC or FR4 prepreg (left) and MICROLAM 630 dielectric (right), both on an FR4 core In figure 3, you can see that by looking left to right, the effect of microvia dielectric (layer 1-2) losses is apparent. The wider eyes on the right are due to lower dielectric losses, which directly impact the jitter and overall resulting signal integrity. The effect of conductor losses is evident when looking at the MICROLAM 630 eye patterns on the right side and comparing top and bottom. The wider trace on the top has a more open eye pattern because of reduced conductor losses. The larger cross-sectional area of the trace reduces copper losses due to the skin effect at gigahertz frequencies. Note that in order to increase the trace width, the thickness of the dielectric also had to increase in order to maintain an impedance of 50 Ohms. Although only a test coupon from the edge of a panel, a magnified cross section of an outerlayer construction showing MICROLAM 630 dielectric on an FR4 core is shown in figure 4. Microlam 630 FR4 core 75 um Figure 4 Cross-section showing MICROLAM 630 dielectric on an FR4 core (note the 75 um microvias, the ML630 dielectric is 57 um thick) The picture shows the woven glass FR4 core. The expanded PTFE (eptfe) micro reinforcement makes the MICROLAM 630 (layer 1-2) appear white while the dark areas are resin rich, where BT resin flowed during lamination. The same test boards that were used in obtaining the data seen in figure 3 were then
measured in the frequency domain and the results are shown in figure 5. The insertion loss was examined as a function of frequency from 50 MHz to 10 GHz and normalized by length. 0.0-0.2 Insertion Loss (db/inch) -0.4-0.6-0.8-1.0-1.2-1.4-1.6 0.24 db/in @ 2.4 GHz 0.45 db/in @ 5.2 GHz 200 um (8inch) microstrip traces Nominal impedance = 50ohms +/- 5%) ML 630 86 um ML 630 57 um FR4 125 um RCC 75 um -1.8 2 3 4 5 6 7 8 9 10 Frequency (GHz) Figure 5 - Insertion loss for 200 micron long, 50 Ohm microstrip traces (125 µm wide traces for 75µm RCC and 57 µm thick ML630, 210 µm wide traces for 125 µm FR4 and 86 µm thick ML630) As noted previously, the testing was done keeping the same trace width, which required thinner layers of the low Dk materials. However, if the same thickness of low Dk dielectric is used, the trace width can be increased thereby minimizing both copper conductor and dielectric losses. Subsequent experiments will examine these added benefits. RF Passive Filter Performance There is growing interest in integrating passive ("buried passives") functionality in the substrate or PCB 8-10. The primary initial reason for conducting this study was to evaluate the effect of board materials, solder mask, and metallization on the performance of PCB level passives, often used in filtering RF signals. The basic filter circuit described in this report is shown below. It consist of three series coupled capacitors and two inductors connected to ground. The component values in this low pass filter are suitable for manufacturing as integrated components in MCM-L technology using the microvia dielectric for parallel-plate capacitors.
Figure 7 - Top view of an integrated PCB bandpass filter (an integrated passive PCB component) Figure 6 - Schematic of a passive RF filter Terminal 1 Terminal 2 This type of filter might be used for protecting WLAN or Bluetooth receivers from blocking by cellular transmitters at the 1.7-1.9 GHz bands or below. Different versions of a filter with capacitors in the range of less than 1 pf and square inductors with varied numbers of turns providing a few nh were manufactured. Using capacitors and inductors in series or parallel facilitates tailoring of the RF filter properties. The circuit shown in figure 6 is basically a high pass filter, but it is never possible to manufacture the passive components so that they conform to ideal capacitors and inductors. So-called parasitic components will be associated with the components and especially at high frequencies they can dominate the overall behavior of the filter circuit. Integrated inductors will for instance have some capacitance between the spiral structure and the ground planes. Such effects may be used intentionally, and in this particular filter, a band pass characteristic is obtained rather than a high pass characteristic 11. This is actually an advantage because some attenuation is obtained even at frequencies higher than the frequencies of primary interest, the IEEE802.11b/g and Bluetooth bands at 2.4-2.5 GHz. The actual appearance of this type of filter on the surface of a circuit board would look like the printed features (square capacitors and spiral inductors) shown in Figure 7. The transfer function from one of these filters is shown in figure 8. [db] 0-10 -20-30 -40-50 -60-70 Connected to groundplane Transmission S21 Reflection S11 0 2 4 6 Frequency [GHz] Figure 8 - Measured transmission (S21) and reflection (S11) coefficients of the filter, without solder mask. The transmission coefficient at 1.9 GHz and below is 36 db. The minimum transmission coefficient within the 2.4-2.5 GHz band is 2.3 db. Additional filter designs were also tested and one was selected to examine the effects of dielectric, solder mask and ENIG. These variables are shown in table 3. Table 3 Band pass filter variables examined in this study Microvia Dielectric Core Ni/Au Solder Mask Microlam 630 A FR4 No No Microlam 630 B FR4 No Yes Microlam 630 C FR4 Yes Yes RCC A FR4 No No RCC B FR4 Yes Yes
The experimental structures for these conditions were tested and the transmission, or insertion loss, of each is shown in figure 9. 0 Insertion Loss, S21 [db] -2-4 -6-8 2.4E+09 2.6E+09 2.8E+09 3.0E+09 Frequency [Hz] Microlam 630 A Microlam 630 B Microlam 630 C RCC A RCC B Figure 9 - Insertion loss, S21, of the different conditions tested in this experiment (see table 4) While the RF filter would need to be tuned to the frequency band of interest, the transmission of the filters made with the different conditions shows a span from 0.9 db to 3.2 db. The graphical results are summarized in table 4. Table 4 - Band-pass filter performance results Microvia Solder Insertion Core Ni/Au Dielectric Mask Loss Microlam 630 A FR4 No No 0.9 Microlam 630 B FR4 No Yes 1.1 Microlam 630 C FR4 Yes Yes 2.2 RCC A FR4 No No 1.6 RCC B FR4 Yes Yes 3.2 The most important factors affecting the insertion loss were nickel/gold metallization, an effect of about 1.1 db, and microvia dielectric, 0.6 1.0 db. The insertion loss of filters with Ni/Au metallization was approximately twice as high as those without Ni/Au. Solder mask increased the insertion loss by 0.20-0.25 db. The use of a low loss dielectric also improved the insertion loss from 1.6 to 0.9 as compared with standard board without Ni/Au or solder mask. The difference was greater (1.0 db) for PCBs with Ni/Au and solder mask. For this filter, MICROLAM 630 dielectric reduced 0-10 -20-30 -40-50 RCC Microlam 630 1 1.5 2 2.5 3 Frequency [GHz] the band-pass loss by 0.5 db or approximately 25%, which is significant, while the transmission coefficient for both materials at 1.9 GHz and below was 36 db. MICROLAM 630 dielectric also provides improved thickness control of the dielectric, which can be advantageous for impedance controlled transmission lines as well as inductors and integrated capacitors. Some of the be relative dielectric constants of the materials used in these simulations were adjusted downward to provide better correlation with measured values. In order to illustrate the importance of dielectric loss in the filter, two additional simulations were made on hypothetical materials: 1. The dielectric loss factors of the core material and the solder mask were set equal to that of MICROLAM 630 dielectric. MICROLAM 630 dielectric All in the figure. 2. The dielectric loss factors of all dielectric materials were set to zero. Zero loss material in the figure. These simulations are shown in Figure 11, and the improvement in transmission loss is significant compared to previously measure values, but not as big as that obtained by replacing the RCC with MICROLAM 630 dielectric. Simulations of RF Band-Pass Filters Another high pass LC filter was simulated for both MICROLAM 630 dielectric and RCC materials. The simulation results are shown in figure 10. Figure 10 - Filter performance comparison for two of the dielectric materials examined
0-0.5-1 -1.5-2 -2.5-3 Zero loss materials Microlam 630 All Microlam 630 RCC 2.2 2.3 2.4 2.5 2.6 2.7 Frequency [GHz] Figure 11 - Simulated filter performance These results show that the microvia outer layer dielectric (RCC or MICROLAM 630 dielectric in this experiment) is the most important dielectric material with regard to loss. It has a stronger effect on electrical performance than the core material or solder mask when used in a microstrip construction. This is easily explained by the fact that the highest electric fields in the filter design are seen between layers one and two in the stackup, thus contributing most to dielectric loss. The inductors can have significant dielectric and conductor losses due to their long trace length. Conclusions A substrate material with superior electrical properties at gigahertz frequencies can significantly improve digital and RF signal transmission as well as RF filter performance. PCB metallization also has a strong effect on electrical performance and to a lesser extent, solder mask. The dielectric loss factors specified by the suppliers of resin coated epoxy foil, FR4 and MICROLAM 630 dielectrics seem to fit well with the measurements made. At gigahertz frequencies, both measurements and simulations show that transmission losses are strongly effected by both copper and dielectric losses. Using a low Dk dielectric enables the use of wider traces while maintaining a thin layer 1-2 thickness that helps to keep an acceptable microvia aspect ratio (thickness/width < 1). Wider traces along with low loss dielectrics will minimize both conductor and dielectric losses for improved electrical performance. The use of low loss microvia dielectrics may allow RF and digital circuit designers greater design freedom due to reduced total signal losses. This in turn enables the designer to maximize system performance (greater wireless range and/or data rate) or use some of their loss budget elsewhere in the system design. These losses can benefit both transmit and receive portions of a design. Lower losses will reduce power requirements, providing: 1. Longer battery life 2. Reduced thermal requirements 3. Reduced EMI Lower PCB losses can provide the ability to have longer distances between components on a substrate or PCB, thus allowing greater freedom during layout. Other options that reduced PCB losses may provide include using cheaper, lower performance components or using PCB traces as the RF antenna. These options can reduce the number and/or cost of components and minimize the total system bill of materials (BOM). Manufacturing tolerances, thickness variation of dielectric and alignment accuracy between layers during fabrication will determine to a high degree how aggressively circuits can be designed with these or other conditions. Future Work Future publications and presentations will provide additional details of the RF filter designs optimized after these initial experiments as well as greater details regarding RF and digital microstrip transmission line performance of these board constructions at gigahertz frequencies and Gb/s data rates. Additional studies will also examine the effects of thickness variation on filter performance for the materials under consideration. While most of this work has focused on high density RF designs, longer trace lengths will also be investigated for server/router types of high speed PCB designs and digital SI metrics such as velocity of propagation, jitter, and cross-talk will be reported. Acknowledgements This was a collaborative effort between Ericsson, Multek, and W. L. Gore & Associates. RF filter testing and simulation was performed by Ericsson. Digital signal integrity and RF transmission line testing was done by Gore and the boards were fabricated by Multek. Special thanks to the Ericsson teams in Kumla and Lund, Multek Kumla and globally, and the Gore signal integrity team. We are also indebted to the many others who contributed their insights or fabricated, tested or simulated as part of these evaluations.
References 1. Holden, Happy, Happy Thoughts HDI s Improvement of Signal Integrity, Circuitree, January 2002. 2. Holden, Happy, Happy Thoughts HDI s Improvement of Signal Integrity, Circuitree, April 2002. 3. Schneider, M. V., Dielectric Loss in Integrated Microwave Circuits, Bell System Technical Journal, Vol 48, pp. 2325-2332, Sept. 1969. 4. Rao, B. Ramo, Effect of Loss and Frequency Dispersion on the Performance of Microstrip Directional Couplers and Coupled Line Filters, IEEE Trans. Microwave Theory Tech., Vol. MTT-22, pp. 747-750, July 1974. 5. Bogatin, Eric, Signal Integrity, The Board Authority, December 2000. 6. Ritchey, Lee W., What Makes a Circuit High Speed?, The Board Authority, December 2000. 7. Raman, Sudhakar, Laser Microvia Productivity: Dual Head Laser Drilling Systems, Circuitree, July 2001. 8. W. J. Borland, S. Ferguson, "Embedded Passive Components in Printed Wiring Boards: A Technology Review." Circuitree, March 2001. 9. D. MacGregor, "Standards Development Efforts for Embedded Passive Materials." IPC Annual Meeting (Orlando, FL), October 2001. 10. P. Sandborn, B. Etienne, D. Becker. "Analysis of the Cost of Embedded Passives in Printed Circuit Boards." IPC Annual Meeting (Orlando, FL), October 2001. 11. Bjorneklet, Are, Conversations regarding RF simulations, 2001 and 2002.