Miniaturized Satellite Switch at 2.5 GHz

Transcription

1 Miniaturized Satellite Switch at 2.5 GHz Dr. Etienne Hirt - Art of Technology AG, Gloriastr. 35, 8092 Zurich, SWITZERLAND Ph: , Fx: hirt@art-of-technology.ch Michael Scheffler, Gerhard Tröster - Electronics Lab, ETH Zurich, SWITZERLAND Wolfgang Wendel, Ralf Epple - Hirschmann Rheinmetall Elektronik, Neckartenzlingen, GERMANY Abstract In this paper we present the MCM implementation of a 9:4 satellite switch in MCM-C/D technology, operating at frequencies up to 2.5GHz. The MCM contains two ASIC switches, four DiSEqC controllers and an inverter. RF interconnections are realized as coplanar lines in order to obtain sufficient shielding and impedance control. With this implementation, it was possible to connect the ASIC outputs without a matching network due to the short distances. Also the DiSEqC controllers, formerly not included, could be moved into the module achieving a more complete partition and enhancing functionality towards digital switching. Measurements revealed that the fabricated module is fulfilling its specs, although a redesign is justified to fully exploit the technology s performance. With this switch MCM it is possible to double the functionality of a switch daughter card on the same form factor. 1 Introduction Satellite TV today has a certain share of TV supply, next to conventional cable TV, and is gaining more and more interest. Especially the new markets in middle and eastern Europe obtain easy access to a vast choice of channels without building an expensive infrastructure network. This increasing number of customers also drives the need for integration of underlying subsystems. Market forecasts are ranging from 0.5 to one million units per year for Europe. Fig. 1 shows a standard PCB containing a DBS 5:4 multi switch operating at 2.4GHz. DBS 5:4 stands for Direct Broadcast Switch, making four satellite plus one terrestrial signals available to four subscribers at the same time. The five inputs are located on the bottom, the outputs on the right and the left. The ASIC switch is placed in the center, surrounded by coplanar lines fanning out. Extending the subsystem to a 9:4 format would be desirable (sockets are already present on top edge), but impossible due to the large area consumption of the coplanar lines for two ASICs. Moreover, just connecting the ASIC outputs leads to performance degradation; a matching network on the other hand requires again too much area. This paper presents the realization of such a 9:4 switch module using MCM-C/D technology. 1.1 A 9:4 switch Next to the two switch ASICs, also four DiSEqC (Digital Satellite Equipment Control) controllers plus one inverter were allocated on the MCM. The DiSEqC system is a communication bus between satellite receivers and peripheral equipment aimed to replace all conventional analog switching [1]. Currently, there are some satellites as e.g. ASTRA and Eutelsat providing digital as well as analog signals. With this joint digital and analog switching capability, the 9:4 switch is fully upward and downward compatible. The entire functional schematic can be found in Fig. 2. On top the 8 incoming satellite signals in A to in H plus the terrestrial signal in TERR are directed to the switching ASICs, which in turn provide the four users U, V, X, and Y. These users can interact with the switch control via the control signals det UtodetY. The det signals are modulated and fed to the switch via the normal receiver coax cable. Demodulation and splitter are located outside the MCM. Additionally, 12 pull-up resistors and 8 capacitors for DC decoupling had to be positioned as close as possible to the ICs. The specifications of the 9:4 switch can be found in Tab. 1. Due to a targeted quantity of units per year and the cost sensitivity of the product, volume capabilities and low cost substrates are mandatory. Therefore, the 9:4 switch was chosen as a project demonstrator for the EU project LAP (Low-cost Large Area Panel Processing for MCM-D Substrates and Packages) [2]. This project has the aim to transfer the thin film substrate production from the actual scale of 4 to 6 diameter wafer lines to lines capable to process up to 24x24in 2. Instead of Silicon wafers, less expensive base materials as ceramics and laminates had to be evaluated. With these materials and due to the economics of scale, the production cost is

2 E 0 E / E. E - E, E + E * E ) + K J 7 K J J K J J K J 9 K J J! K J J" K J 8 K J :!! " " " " A J A J A J A J : E M A H /, Figure 1: PCB containing a 5:4 switch Figure 2: Functional schematic for the 9:4 switch expected to decrease down to 1US$/in 2 (0.15EUR/cm 2 ). For the switch demonstrator, in a first step a ceramics base material has been chosen which has to be mounted onto a intermediate BGA laminate carrier. The purpose was to evaluate the performance of the MCM-D buildup with the option to abandon the laminate carrier during final production (see Fig. 3). A cross section and technology specifications can be found in Tab. 2. The adopted thin film process exhibited also the possibility to integrate the resistors and capacitors into the substrate. For the resistors, a NiCr deposition has been used, whereas for the capacitors the normal ISO layer served as dielectric material. 2 Design ofthe MCM Special attention had to be paid to the signal lines requiring a matched impedance of 75Ω. Due to the thin dielectric layer, micro strip waveguides cannot be realized, thus driving the need for coplanar waveguide design. Based on investigations in [3], the graphs in Fig. 5 were developed showing the ratio of line width w to total opening d versus the characteristic impedance with the ratio w to structure width b as parameter (see Tab. 2 and Fig. 4 for illustration). The effective dielectric property value ffl r;ef f has been derived using SONNET simulations, considering M2, ISO layer, and base substrate. Given a line width of 60μm and a calculated ffl r;ef f of 4.38, a w/d ratio of 0.27 and a w/b ratio of 0.1 could be derived for a 75Ω coplanar line. In order to minimize the influence of process tolerances a curve with low slope in the target region is inevitable. These ratios result in a minimum line space d of 80μm and a minimum GND width of 200μm. Another critical point was the connection of the AS- 1 line width/line space/via land; all measures in microns ICs output lines before leaving the MCM on out Uto out X (see Fig. 2). To obtain a controlled impedance and to suffer not from any phase shift distortion, usually a so-called Wilkinson distributor using =4 interconnects would be required [4]. But due to the small ASIC pitch offered by the MCM technology, a phase shift of zero can be assumed, if the two ASIC outputs are connected with equal stubs shorter than =16 (=: 2.4GHz). Whereas a Wilkinson distributor would have required a length of 14mm (roughly 10mm 2 ), the stub connection was possible after 3mm (PCB minimum after 10mm). The suitability of these assumption has been verified by a SPICE simulation (see Fig. 7). Whereas the 3mm MCM stubs exhibit only -1dB attenuation, the 10mm PCB stubs already have reached 2.4GHz. Moreover, the receiver outputs require a AC coupling to avoid a DC mismatch. In order to avoid the SMD mounting onto the substrate, 10pF coupling capacitors have been integrated into the substrate. They are realized as metal-insulator-metal structure. 3 Results Fig. 6 shows the implementation of the 9:4 switch module, top view onto the populated substrate, bonded to the laminate carrier. Due to the design rules of 40/60/60 1 for the digital part only two interconnect layers were needed to achieve a final substrate size of 17x17mm 2. The substrate is then wire bonded onto a laminate BGA carrier (25 x 25mm 2, 143 I/Os). The RF section is located in the upper part, including the two switch ASICs, coplanar lines, and eight coupling capacitors. The lower section hosts the digital part with four DiSEqC controllersandaninverter.

3 b d M2 M1 (n/a) s w Cu BCB t_m2 t_iso Alumina t_base Figure 3: Intermediate (top) and final (bottom) packaging solution Figure 4: Cross section of the substrate Table 2: Substrate specifications Table 1: Specs 9:4 switch Freq range (incoming) Signal type Line impedance Internal gain Isolation between any terminal MHz analog FM digital QPSK 75Ω -6to0dB > 20 db Base substrate material Alumina Al 2 O 3 thickness t base 500μm dielectric ffl r 9.8 ISO layer material BCB 4024 thickness t ISO 5μm ±0.5 dielectric ffl r 2.65 Metal layer material copper M2 thickness t M2 5μm M1 thickness t M1 2.5μm min signal width w 40μm min line space s 60μm min distance GND - GND b (coplanar) min outer width d (coplanar) After fabrication, the MCM has been mounted onto a DUT board. Examples of the thrupass and cross talk measurements are shown above. Fig. 9 (left) details the signal ratio SAT5 (in E) to receiver U. The thrupass attenuation ranges from -6dB to -9dB, being slightly too high, but the ASIC itself contributed also -4dB. This requires some redesign effort. In Fig. 9 (right) the cross talk from inputs SAT4 & 5 (in DandinE) is displayed while switching SAT1 to receiver U. Here, both signals fulfill the specifications. The thrupass attenuation at higher frequencies is caused by parasitic capacitance of the coupling capacitance as described in the subsection below. 3.1 Integrated Capacitors Integrated capacitors require a large area if no special dielectric material is used as shown in Fig. 6. The selected asymetric configuration (Fig. 10) allows to place the capacitors on the substrate border, thus avoiding a long interconnection distance from ASIC to ASIC. However, this configuration is susceptible to parasitic capacitances to GND. The thrupass attenuation from IC1 is increased from the parasitic load at C1 shown in figure 11. This mismatch can be corrected by reducing the stub length L 3 and enlarging L 1. This increases the attenuation for the path from IC2 for higher frequencies than 3GHz while reducing the attenuation for IC1 and thus outbalances the two paths at no performance loss. 4 Cost Issues Finally, we want to review the cost situation. The scenarios to be compared are on the one hand a 9:4 switch similar to the PBC in Fig. 1 with dual side assembly (full SMD) and on the other hand a 9:4 switch realized with the switch MCM. (From a performance point of view, only the MCM solution has proven its capability to fulfill the specs. Dual side SMD assembly is expected to pose serious cross talk problems.) Cost figures can be found in Tab. 8. Capsuling a number of I/Os in the MCM eliminates the need for a 4-layer board and dual side assembly. Moreover, some components are included in the MCM, thus reducing assembly and part cost of the PCB. Also, the form factor of the outer casing could be improved yielding again in lower cost. These reductions are cur-

4 Characteristic Impedance [Ohms] Characteristic Impedance of Coplanar Line for M2 and Air Ereff=4.38 w/b=0.01 w/b=0.1 w/b=0.2 w/b=0.3 w/b=0.4 w/b=0.5 5 M EJ? D ) = F =? EJ H I 1 L A HJA H w/d, E5 - G + Figure 5: Coplanar w/d ratio vs. characteristic impedance Figure 6: Layout of the 9:4 switch ccoupling Figure 8: Cost data for intermediate packaging -12 S12 (lin) x 1g 1.5g 2g 2.5g 3g 3.5g 4g 4.5g 5g 5.5g Frequency (lin) (HERTZ) Design Type File Wave Symbol D0: /usr/ife/mcm/schmid/aot/projekte/hirschmann/spice/csym AC csym.ac2 D0:A2:s12(db) D0: /usr/ife/mcm/schmid/aot/projekte/hirschmann/spice/csym AC csym.ac0 D0:A0:s12(db) full SMD MCM System Board 15 8 PCB parts production/test MCM substrate/ n/a 76 assembly Outer case, pwr supply Total Figure 7: Transmission characteristics of MCM (triangle) and PCB (circle) stubs rently offset by the high MCM substrate and assembly cost. The substrate cost was expected to decrease when switching to large panels on laminate. This would also eliminate the intermediate carrier cost. However, when preparing the series production, it was immanent that the thinfilm substrate supplier could not deliver at the foreseen cost. Thus, an immediate backup solution was necessary. Therefore, as a commercial implementation wire bond onto an SBU substrate using discrete passives was chosen as described in the following section. 5 Commercial Implementation Laminate sequential build up substrates (SBU) feature microvias that enhance the routing density compared to a standard PCB substrate. Furthermore, a laminate substrate is easy to package because it allows to mount balls on the backside thus providing directly the package of the module. The resulting module measures 22.5*20 mm which is smaller than the packaged thinfilm substrate. But the switch ASICs can not be placed as close as on the thinfilm substrate caused by the larger design rules. However, this does not lead to a performance degradation due to the smaller E r and the advantage that the ASICs can be connected below them on the bottom side without adding much parasitic capacity versus the die backside. The dies are wire bonded and the passive component are mounted as SMT components. The passives are discretes because it is more difficult to integrate the passives in a laminate substrate than in thinfilm. On the other hand passive mounting on a laminate is more cost effective as not only the substrate production but also the assembly can be done in rather large panels. They are glued together with the dies and after wirebonding the whole module is covered with glob top. 6 Conclusion In this paper we have presented the implementation of a 9:4 DBS switch on MCM-C/D technology. The MCM

5 Figure 9: Measurement plots: left thrupass attenuation SAT5 to rec U, right SAT1 switched to rec U, cross talk from SAT4 and SAT5 " G * + * H A % A J = A J = $ " F. # 0 )! AH ' & F. # 0 A J = /, 5 K > I J H = J A * =? I E@ A Figure 10: Asymetric DC Decoupling for the Switch ASIC Outputs Figure 11: Measured Parasitic Capacitance from C1 to GND on Substrate Backside includes 7 ICs plus integrated passives and is currently packaged on an intermediate BGA carrier. Measurements showed that the MCM is mostly fulfilling the specs up to the desired frequency of 2.4GHz. The next step will be to migrate the design to a laminate SBU implementation. This enlargers the module size but does not degrade the performance and results in a cost reduction. To improve the thrupass attenuation, parasitic capacitances have to be eliminated. Cross talk can further be improved by adding capacitors to the digital control lines. In the final implementation, the 9:4 MCM switch can be foreseen to fully meet the specs at no or only a moderate cost penalty. References [1] EUTELSAT (European Telecommunications Satellite Organization), 70, rue Balard PARIS Cedex 15, France, Digital Satellite Equipment Control Bus Functional Specification, 4.1 ed., [2] Esprit project 26261: Low cost large area panel processing of MCM-D substrates and packages [3] A. Thiel, C. Habiger, and G. Tröster, Investigations on Novel Coaxial Transmission Line Structures on MCM-L, in Proc. IEEE Multi-Chip Module Conference (MCMC 97), 38-43, Santa Cruz CA, USA, Feb. 4-5, [4] D.M.Pozar,Microwave Engineering, ch. 7.3 The Wilkinson Power Divider, p. 363ff. John Wiley & Sons, Inc., 2 ed., 1998.