CARTS Europe 2008
20-23 October
Helsinki, Finland
Evaluation of Broadband Filters in an LTCC Process for 24GHz Applications
Veljko Napijalo, Denver Humphrey, and Billy Verner TDK Electronics Ireland, 3022 Lake Drive, City West Business Campus, Dublin D24, Ireland, Phone +353 1 4133 235 Email:
[email protected] Introduction Multilayer technologies such as Low Temperature Co-fired Ceramics (LTCC) offer the advantage of reducing the size of microwave circuits in a low cost, high volume production environment. This technology combines cheap and well established screen printing technique to print conductor patterns on low-cost ceramic sheets. Any connections required between the layers are usually formed by laser punching of holes in the ceramic sheets and filling the holes with conducting paste. To form the final multilayer circuits, the layers are stacked together and fired in the oven. Fabricated multilayer substrate can integrate passive elements such as capacitors, inductors and various types of transmission lines, while large variety of passive and active components or chips could be mounted on the top of the substrate. LTCC technology was successfully applied in numerous modules for mobile phones, Bluetooth, WLAN and similar systems. All these applications are allocated to frequencies that are less than 6 GHz and could be referred to as low frequency applications. At recent years number of application that are requiring high volume and cheap manufacturing emerge at frequencies 20-26 GHz or even higher. Before moving to higher frequencies, a particular technology process has to be tested against stringent requirements of high frequency designs. This paper presents the continuation of the previous work [1] on testing of suitability of standard TDK LTCC technology for realization of 24 GHz circuits for wideband automotive radar applications. The other goal of this work is to explore the capabilities of CAD tools available, especially use of electromagnetic (EM) simulators, and suggest the design flow outline. Conclusions of this work are also important for addressing other 24 GHz LTCC circuit designs e.g. amplifiers. Strategy of Characterization of an LTCC Process Trough Filter Designs The first step towards the design of high frequency circuit was precise characterization of LTCC electric properties (dielectric constant, dielectric loss tangent and conductor loss) up to 40 GHz, and addressing process parameters of interest (line and gap width tolerances, minimum printable line-gap-line pattern resolution, surface roughness, conductor definition etc.). This work has been described in [1] with some filter responses shown there to illustrate the accuracy of measured electrical parameters. More details concerning filters are described here.
For 24 GHz applications, designs are mainly based on distributed circuit concept. The electrical parameters of distributed circuits are critically dependent on dimensions of the transmission lines used for designs, thus the tolerance and resolution of the conductor printing process are very important. Therefore, a procedure of design, manufacturing and measurements of an adequate distributed circuit should be carried out in order to draw the final conclusions on suitability of a particular LTCC process for successful 24 GHz circuit designs. Certain types of microwave transmission line filters can be used as representative distributed circuit to carry out the technology characterization procedure. Besides being the most used circuits in microwave systems, the filters were selected because their synthesis is preformed through precisely defined design process [2]. Furthermore, well designed filters should resemble the ideal performance described by mathematical equations, and reasons for any discrepancy can be easily tracked [3]. Two types of bandpass filter were designed. The first one was the filter with edge, parallel coupled lines. The motivation to choose such a filter for the purpose of LTCC technology characterization at 24 GHz was to test the impact of the tolerance of an LTCC conductor printing process to filter performance. As this filter type utilizes narrow microstrip lines for the first and last coupled line sections, this influence can be easily tested by comparison of EM simulated and measured responses. Furthermore, this filter type represents good test circuit to benchmark different EM simulator types and conclude which one to use in subsequent 24 GHz LTCC circuit designs. In general, 2.5D EM solvers are considerably faster then full 3D EM solvers, but the simulation times become similar if conductor thickness is included in the 2.5D EM analysis. The properties of parallel edge coupled lines are dependent on the conductor thickness and in some cases neglecting the conductor thickness can cause large differences between simulation and measurements. Therefore, the parallel edge coupled lines filter designed using 2.5 D EM simulator can be used as a test vehicle to study the effect of neglecting the finite conductor thickness in a particular LTCC environment at 24 GHz. The second type of filter was the shunt open stub type that is derived from first higher order passband of the lowpass filter prototype [2]. This type of filter has well defined multiband characteristics in both low and high frequencies, so its response could be used to verify the value of measured dielectric constant of LTCC material in both frequency regions. Also, the out-of-band attenuation for this filter was specified as relatively high to test the possibility of realizing highly selective structures for 24 GHz applications with available LTCC technology. Details on Filter Design Flow Modern microwave circuit design programs, such is Advanced Design System (ADS) from Agilent Technologies [4], include a filter synthesis tool that significantly simplifies the design flow. A circuit schematic with required transmission line elements can be obtained automatically after filling in the desired filter parameters in a program filter synthesis dialog window. For the parallel edge coupled lines filter, the bandwidth was specified to be 22-26 GHz and a value of 30 dB for out-of-band attenuation was chosen at 20 GHz. For the case of shunt stub filter, the bandwidth was specified to be 21.5-26.5 GHz and the out-of-band attenuation at 20 GHz was specified to be 40 dB. Chebishev types were selected for both filters and a value for maximum return loss of the filters was 20 dB.
Although execution of the synthesis routine is not influenced by a user, the actual operation flow inside the routine is easy to understand as it must be in accordance with a well known manual routine. The required order of a canonical filter prototype comprising ideal transmission line elements is calculated first based on the filter specifications according to the user input. Next, the routine calculates electrical parameters of the ideal transmission lines i.e. required values of the characteristic impedances and electrical lengths at the specified central frequency of operation. Using calculated electrical parameters of the ideal transmission lines, the values of physical dimensions of microstrip lines required for a filter circuit are calculated next. The execution of a synthesis routine ends with the automatic creation of microstrip filter schematics. The schematics of the filters specified in the previous paragraph are presented in Fig. 1 and Fig. 2. It has been determined by a synthesis routine that, in order to meet the specifications given above, the required order for the parallel edge coupled lines filter must be five, while for the case of shunt open stub filter, the required order is seven. For both filters dimensions of the microstrip lines were calculated for LTCC substrate with relative dielectric constant of εr=7.5 and thickness of h=320 µm.
Fig. 1 ADS schematic of the 5th order Chebishev filter with parallel edge coupled lines (microstrip models).
When calculating the dimensions of microstrip lines, synthesis routine uses substantially simplified equations to relate electrical characteristics of microstrip lines, as required by an ideal transmission line prototype, with their physical dimensions. The synthesis routine does not automatically correct these dimensions to account for the influence of various microstrip discontinuities such as step in width, open end and line junctions. As the impact of such discontinuities is very strong at 24 GHz, the responses of a microstrip filter circuits from Fig. 1 and Fig. 2 are significantly different from the responses of corresponding prototypes. Therefore, dimensions of microstrip lines need to be optimized for each of the filters, and the ones calculated by synthesis routine should only be taken as the initial values. According to the dimensions for microstrip elements as obtained directly from a synthesis routine, a value of a microstrip line width for the first and the last resonators of the filter with parallel coupled lines was required to be less than 100 µm, which is the value of recommended minimum [1]. It was decided to change the width of the resonators to be at the recommended minimum value and to re-optimize dimensions of all of the resonators to recover the specified passband and stopband characteristics. The electrical impact of the changes is that the filter is effectively scaled to impedance different from 50 Ω, which in
turn results in the increase of the filter return loss. The values for the return loss around 10 dB obtained from optimization were found acceptable for this evaluation work. The synthesized circuit of shunt open stubs did not required any adjustment related to technology rules, therefore the optimization of microstrip dimensions could be performed immediately after the synthesis.
Fig. 2 ADS schematic of the 7th order Chebishev filter with shunt open stubs (microstrip models). To accurately predict the characteristics of the filters when fabricated, the microstrip circuit of the filters must be analyzed using an EM simulator. Planar, 2.5D EM simulator Momentum, available as an integral part of the ADS package, was conveniently used for the EM simulations. To significantly reduce the simulation time as explained earlier, the conductors were modeled as zero thickness. Losses in metal and dielectric material have been taken into account according to the values determined in [1]. For both of the filters, initial EM simulated response was different from the one optimized using microstrip circuit simulations. The typical differences between microstrip circuit response and the EM calculated filter response for the case of microstrip transmission line filters are downwards shift in center frequency of a filter - by amount ∆f, and narrower bandwidth by amount ∆B. Direct optimization of a design using EM simulator can be rather time consuming and the process might not end successfully as all of the parameters (line lengths, widths and gaps) of all resonators should be optimized simultaneously. A good engineering practice for such microwave filters is to re-optimize a microstrip circuit using circuit simulator, but with adjusted optimization goals. Optimization goals should be adjusted according to observed differences between the responses of the filter circuit as optimized using circuit simulator and obtained from EM simulations. Re-optimization goals should be defined to “overshoot” the initial ones, i.e. re-optimization goals should require a filter center frequency to be increased by ∆f and the bandwidth to be increased by ∆B. After such an optimization is executed, the filter circuit should be again analyzed using EM simulations, and any discrepancy remain can be removed by performing additional iterations of the described re-optimization procedure. Following the procedure, both filters were optimized to be close to the specifications. However, to fully meet all of
the specifications, the values for fine adjustments required for the dimensions of the microstrip resonators were found smaller then the resolution of the conductor printing process, therefore the optimization was stopped. The optimized microstrip layouts of the filters are presented in Fig. 3 and Fig. 4 while the filter responses obtained using 2.5 D EM simulations are shown with dashed lines in Fig. 5 and Fig. 7.
Fig. 3 EM optimized microstrip layout of the 5th order Chebishev filter with parallel edge coupled lines.
Fig. 4 EM optimized microstrip layout of the 7th order Chebishev filter with shunt open stubs.
0 0
S parameters (dB)
-10
-10
-20
-20
-30
20
22
-30
24 26 Frequency (GHz)
28
-40
-50 0
5
10
15 Frequency (GHz)
20
25
30
Fig. 5 Comparison of measured and 2.5D EM simulated S parameters of 5th order microstrip coupled lines filter. Solid lines – measurement, dashed lines – simulation; x – S11, ○ – S21. Measured Results and Comments Designed filters were fabricated and measured using ground-signal-ground (G-S-G) coplanar waveguide (CPW) probes. The results of the measurements are presented with solid line in Fig. 5 - Fig. 7. The passband response is expanded as an inset at each figure.
0 0
-10 S paramters (dB)
-10
-20
-20
-30 20
-30
22
24 26 Frequency (GHz)
28
-40
-50 0
5
10
15 Frequency (GHz)
20
25
30
Fig. 6 Comparison of measured and 3D EM simulated S parameters of 5th order microstrip coupled lines filter. Solid lines – measurement, dashed lines – simulation; x – S11, ○ – S21.
0
S parameters (dB)
-10
-20
0
-10
-30
-20
-40
-30
20
22
24 26 Frequency (GHz)
28
-50 0
5
10
15 Frequency (GHz)
20
25
30
Fig. 7 Comparison of measured and 2.5D EM simulated S parameters of 7th order microstrip shunt open stub filter. Solid lines – measurement, dashed lines – simulation; x – S11, ○ – S21.
Measured S parameters for the coupled lines filter are shown in Fig. 5. A shift with respect to frequency between measured and EM simulated values can be observed. The cause for discrepancy was found to be the assumed zero conductor thickness. The same filter circuit was re-simulated using 3D EM simulator [5]. For this simulations, the conductors were modeled to have the thickness of 16 µm and with half of the cross section embedded in dielectric, as shown in Fig. 8. Such a conductor modeling better approximate the real LTCC conductors [1] and takes into account the influence of the finite conductor thickness to properties of the coupled lines. Better agreement between the measurements and 3D simulations is illustrated in Fig. 6. The relative difference between corresponding frequencies of return loss dips and the passband shift is less than 1% which can be considered as very good. For shunt open stub filter measured results are presented in Fig. 7. The agreement between the measurements and 2.5D EM simulations is much better than for the case of coupled lines filter. All of the measured return loss dips and the passband location agree with simulations better than 1%. The agreement at low frequency region is even better as the measured and simulated responses are almost exactly matched up to 10 GHz. The slight disagreement at higher frequencies can be attributed to manufacturing tolerances and zero conductor thickness used for EM simulations. However, the impact of those parameters to the filter response is much less pronounced than in the case of coupled lines filter (results from Fig. 5).
LTCC
LTCC
(a)
(b)
Fig. 8 Conductor cross section modeling (a) in 2.5D EM simulator; (b) in 3D EM simulator
Conclusions
Two LTCC broadband filters designs for 24 GHz applications have been described. The motivation for choosing two filter structures to test suitability of standard LTCC technology for 24 GHz circuit designs was explained. The outline of filter design flow has been presented. Measured results of the fabricated filters are in very good agreement with the simulations. Discrepancies were explained and their minimization illustrated. Shunt stub structures seem to be very robust to the standard screen printing process limitations and can be used in 24 GHz designs e.g. as elements of an amplifier matching network. This type of structure can be accurately analyzed using 2.5D EM simulator. The coupled line structures require full 3D EM simulator for accurate analysis which will result in increased simulation times. Furthermore, a realistic conductor cross section shape must be taken into account in the simulations. The use of this structure may require the trade-offs between the desired performance and technology limitations regarding minimum conductor thickness. However, the standard conductor printing process tolerances are acceptable for the 24 GHz designs as the error between 3 D simulations and measurements is less than 1%. Measured insertion losses of both filters are on average 2.5 dB higher than simulated. The error comes from inability to include realistic conductor shape e.g. edge definition and surface roughness in EM simulations. This value can be scaled with respect to the structure length and used to correct EM simulation results of other 24 GHz LTCC circuits e.g. for amplifiers when estimating achievable amplifier gain to determine required number of stages in the system. References
[1] D. Humphrey, B. Verner, V. Cojocaru, B. Clarke, T. Young and V. Napijalo, "RF Benchmarking of LTCC Circuit up to 40 GHz", International Conference on Electronic Packaging, ICEP 2006, Tokyo, Japan, 2006, Paper TA3-3. [2] G. L. Matthaei, L. Young, and E. M. T. Jones,”Microwave Filters, Impedance Matching and Coupling Structures”, Artech House, 1980. [3]
Dj. Budimir, “Generalized Filter Design by Computer Optimization”, Artech House, 1998.
[4]
ADS 2008A Manual, Agilent Technologies, Paolo Alto, CA, USA, 2008.
[5]
CST Microwave Studio v2006, CST GmbH, Darmstadt, Germany, 2006.
