Making Manufacturing Choices For X-Band BPFS

Bandpass filters help extract desired signals from a portion of frequency spectrum while rejecting unwanted signals. In addition, they can symmetrically or asymmetrically modify the amplitude and/or phase of a signal. They are essential to a host of commercial and military systems and have evolved over the years, so much so that a review of available filter design and materials technologies can aid engineers faced with developing high-performance bandpass filters for X-band applications.

In complex EM environments, where signals must be strictly confined to allotted bandwidths to minimize interference, optimized RF/microwave component performance is critical, with the filter being among the most important of these components. Of the numerous configurations of filters, the planar waveguide, transmission-line structure in microstrip is well suited for high-precision thin-film, manufacturing techniques, passive microwave component manufacturing, and module assembly techniques.

In this configuration, dielectric material is sandwiched between metallized conductor(s) with a patterned metallization circuit conductor on top and blanket ground-plane on the back. Although the resultant, electromotive (E) and magnetic (H) fields provide for quasi-transverse electromagnetic (TEM) propagation due to the dielectric being asymmetric with respect to the top circuit conductor (touching the conductor only on one side), microstrip geometry provides for good power handling, moderate radiative loss (with reasonable crosstalk), and reasonable dispersion over a wide characteristic impedance range (about 15 to 150 Ohms).

Filter performance figures-of-merit (FOM) are designated by S-param-eters, bandwidth, center frequency, ripple, rejection, group delay, and power-handling capability. With a specification set using these FOMs, calculation-intensive EM modeling and optimization can be performed. An appropriate transfer function, which closely approximates a desired filter response (Chebyshev, Bessel, elliptic, etc.), can be selected and an iterative FOM optimization process performed. The resulting filter structures can be a cascade of end-coupled, edge-coupled, interdigitated, half-wavelength-long, and open-circuited resonators. The offset parallel positioning of the resonators has a symmetry that allows for coupling along the length equal to one-quarter wavelength at the center frequency of the filter. Adjusting the offset symmetry, spacing, width, length, and thickness of the conductor resonators and the dielectric constant (e_r) and thickness of the "sandwiched" dielectric provide for optimizing the FOMs. Examples of completed edge-coupled and interdigitated high-frequency filter structures are shown in Fig. 1 and Fig. 2, respectively.

For manufacturing a filter, once an optimized design is achieved, the appropriate conductor artwork is generated and conventional thin-film processing is used to deposit the metallization patterns required. Filled viaholes were used in these filter layouts to allow access with ground-signal-ground (G-S-G) probes on the microstrip conductor top side. These filter designs also featured polyimide-supported air-bridges for conductor interconnects. Measurements of material e_r and loss tangent (tan Δ) were made with a commercial HP8510 vector network analyzer (VNA) from Agilent Technologies from 18 to 25 GHz using an open resonator with the E-field parallel to each of the two principal in-plane axes of the substrates.

Parallel-plate techniques were used to derive e_r and tan Δ from capacitance and dissipation factors, respectively. The 99.6-percent, 0.015 x 4.5 x 3.75-in. Al₂O₃ ceramic substrate was cleaned, sputter-metalized with TiW/Au (1000-Angstroms/2500-Angstroms), and electroplated with 3.75 m gold (Au). Substrates were diced to nominal dimension of 4.40 x 3.70 in. to provide isolated top and bottom electrodes. An inductance-capacitance-resistance (LCR) meter/ fixture was used to perform capacitance measurements with subsequent calculation of e_r using dielectric thickness, electrode area, and measured capacitance.

Surface roughness was measured with a contact profilometer in a similar 12-point array on each plate. On each of 50 Al₂O₃ substrates, thickness measurements were made in a 12-point array distributed over the 4.40 x 3.70 in. area, and two length and width measurements were made on the major and minor axes of the substrates. Table 1 shows e_r, tan Δ., and thickness distribution data for the 50 Al₂O₃ substrates with "within-substrate" data for the "best" and "worst" individual substrates. The surface roughness was superior to the material requirements data (MRD).

Table 1 shows the e_r and tan Δ results for a high-quality 99.6-percent Al₂O₃ ceramic dielectric. The MRD for e_r was 9.9 and tan Δ was 0.0001 at 1 kHz and 1 MHz (ASTM D150). The manufacturer-reported e_r (including tolerances) was slightly higher, but in reasonable agreement with the 20-kHz measurements for the (50) plate lot statistical data. However, the 20-kHz "best" to "worst" range of 9.78 to 10.2 (Δ = 0.42) for e_r exceeded the reported data of 9.8 to 10.0 (Δ = 0.2). The 17-to-25-GHz measurements trended similarly with respect to the MRD and had a "best" to "worst" range of dielectric constant with Δ = 0.25. The tan Δ for the (50) plate lot statistical data was approximately three to four times the MRD. The measured thickness range and mean thickness in the (50) plate lot was slightly higher than the nominal MRD of 0.015 in. (0.381 mm) and tolerance was 0.369 to 0.393 mm.

The Al₂O₃ dielectric substrate properties that are desirable for superior microwave electrical performance are low loss tangent (low dielectric loss), high resistivity, isotropic/uniform dielectric constant, uniform/smooth surface and a uniform thickness. The characteristics of the dielectric should be similar in the as-received and processed conditions and stable over frequency and temperature. Isotropy of these properties are especially important in microstrip circuits to minimize fringing capacitance issues in fine line/gap precision coupled structures such as filters and Lange couplers.

The major contributors to dielectric losses are conduction and polarization effects. Conduction losses typically are attributable to conduction band-to-valence band electron transitions and can be attributed to impurities, secondary phases, and point defects that disturb the Al and O sublattice symmetries. Polarization losses are an indication of the efficiency of the dielectric at propagating the EM field. The dissipation factor (tan Δ) is the assessment of the dielectric loss. The efficiency of the alignment of the electric dipole (ions-electrons) in the Al₂O₃ dielectric as the time-variant E field propagates through the dielectric will affect loss characteristics. The polycrystalline Al₂O₃ may have preferred crystallographic orientations and grain size distributions that can promote/inhibit this alignment thereby affecting the dielectric constant and loss tangent. Non-uniform dielectric properties can lead to increased radiative loss, increased crosstalk, localized variations in characteristic impedance (mismatch effects), degradation in signal time delay and rise time and the appearance of intersymbol interference (ISI) effects.¹ These conduction and polarization losses increase as temperature and frequency increase.

For the microstrip planar waveguide geometry, the dielectric media in which the conductor exists is non-homogeneous; for both the signal and ground conductors, one surface is in contact with the Al₂O₃ dielectric (e_r of approximately 9.9) and the remaining surfaces are in contact with air (e_r = 1.0). This geometry supports quasi-TEM wave propagation with a significant portion of the EM-signal (field) propagating in the Al₂O₃ dielectric substrate. The critical RF-current-carrying regions are the metal/ceramic interface regions of the signal traces and the ground planes. In microwave applications, microstrip dielectrics are typically high-quality, low-loss materials, and conductor loss dominates the total loss performance. However, the contributions of dielectric losses to the total losses will increase with increasing frequency. Therefore, manufacturing processes that can modify the dielectric/conductor interface (plasma and/or chemical cleaning, annealing, conductor deposition parameters) through surface chemical, crystallographic defects, and interdiffusion modifications can have an effect on performance. Although acceptably uniform, the Al₂O₃ thickness was higher than the MRD, which could lead to unanticipated microstrip circuit characteristic impedance increases if not identified. For high quality 99.6-percent Al₂O₃ (R_rms of about 0.05 m) dielectric losses due to surface roughness effects are not influential (for Au or Cu conductors) until reaching 100 GHz and higher frequencies.

Continued on page 2

Page Title

Since all conductors are nonideal (with some resistance), EM energy is absorbed in a conductor, causing loss and dispersion effects. Conductor attenuation (α in dB/mm) was evaluated for DC magnetron-sputtered Au and electroplated Au (sputtered TiW 500-Angstrom adhesion layer) using a 20.0-mm microstrip ring resonator on polished 99.6-percent Al₂O₃ (0.015 in. thick). The ring-resonator geometry was used to eliminate end-effect corrections and to reduce surface wave and radiation parasitic losses. S-parameter measurements were performed on an HP8510C VNA using a model MTF26 microstrip test fixture from Cascade Microtech. A 12-term calibration was performed before and after each measurement and cable/fixture effects were de-embedded. Attenuation was determined from unloaded Q values obtained from ring resonance measurements where Q_UL = Q_L/(1-(10^-L/10)^1/2); L is the measured loss and QL = _3dB/_res. The total microstrip attenuation (in dB/mm) was then calculated from α = (0.693 _res(e_eff )^1/2 (cQ_L) and used for the determination of the conductor attenuation. Since the total microstrip attenuation is comprised of both the conductor and dielectric losses, the conductor attenuation (and dielectric attenuation, tan Δ (Table 2, note 2) was calculated from the microstrip geometry empirical equations.^2,3 The extracted conductor attenuation (in dB/mm) is shown in Fig. 3.

Conductor losses are typically introduced through the series resistance of the multiple conducting materials. The losses have material (resistivity), geometry (line width/space, roughness, signal-ground spacing), and frequency dependencies. For the conductor attenuation in Fig. 3, the attenuation trend was the same but of lower magnitude, electroplated Au, sputtered Au, and bulk (model) Au. The loss slopes (in dB/mm) per GHz were 0.00052, 0.00043, and 0.00031 for electroplated Au, sputtered Au, and bulk (model) Au resonators, respectively. Both chemical purity and microstructural uniformity/symmetry affect the electrical properties of thin films. For noble metals, it is typically recognized that sputtered films can be more chemically and structurally uniform than electroplated films, hence the data for the sputtered Au resonator structures being superior to the electroplated Au structures.

Since metals are not perfect conductors, the E-field penetrates a finite distance into the conductor. When the RF current flows in the conductor, the current and E-field interact to reduce the current density from the conductor surfaces toward the conductor interior. Table 3 shows the increasing RF resistance and decreasing skin depth (Σ) with increasing frequency. The 1Σ and 3Σ depths are shown since the accepted (practical) skin-depth threshold is 3 to 5 to minimize current-crowding effects. With skin depth directly proportional to resistivity, and inversely proportional to frequency, thinner conductors are functional at higher frequencies if high-conductivity conductors are used. The greatest current magnitude is localized in the conductor at the contact interfaces between the metal and the dielectric, therefore it is important to consider the electrical quality and spatial uniformity of the conductor that is used as the signal line and the ground plane.

Although it has been traditionally accepted that conductor loss scales as f^0.5, it has been shown that there are transition regions where loss deviates from the f^0.5 model and is affected by the geometry, resistance, and inductive reactance of the conductor.⁴ The distribution of current flow in the microstrip conductor is frequency dependent and has been shown to be 4 percent greater current flowing on the metal/ceramic dielectric interface that on the metal/air dielectric surface at 10 GHz (X-band).

The entire metallization scheme should be considered in the electrical performance to achieve low attenuation. Typically, the adhesion metals are of high resistivity and high loss so the thickness should be limited to provide appropriate adhesion but not excessive increase loss. If the thickness is ranged around 500 Angstroms, low loss can be realized up into W-band frequencies (about 70 GHz). The TiW adhesion layer used in this work is 90-weight-percent W and 10-weight-percent Ti and is of superior electrical resistivity to Ti and Cr. The reactive nature of the adhesion metals is desirable for promoting adhesion; however, interdiffusion effects into the upper conductor(s) and lower dielectric must be controlled. In addition, the reactivity must be considered during processing to minimize metal reactivity with the processing gas ambient, which will increase resistivity and attenuation. Using characterization data from the Al₂O₃ dielectric and TiW/Au conductor loss evaluations, multiple design geometries for a 10-GHz, broadband, low-loss BPF were performed. The desired BPF specs are shown in Table 2, selected for applications at X-band.

Preliminary topologies of broadside edge-coupled filters were evaluated using quasi-static graphical analyses. The quasi-static analysis involved the following general steps:

1. Specify filter sections (coupled resonators)as admittance inverters with a length L of λ/4, of separation distance S, of line width W, supported on a dielectric of thickness t.
2. Use specified/calculated attenuation characteristics to determine the number of coupled sections; the calculated attenuation and VSWR results for the selected design minimum VSWR (1.07:1) and attenuation (1.8 dB) occurred between 10100 and 10200 MHz.
3. Specify filter response: multipole Chebychev, equiripple, etc.
4. Calculate the g-factor matrix.
5. Calculate the admittance inverter values (J) for a normalized element prototype lowpass filter (LPF).
6. Calculate odd-mode (Z_0o) and even-mode (Z_0e) coupled line characteristic impedances for all sections.
7. Select appropriate Z_0o and Z_0e on quasi-static plots and determine S, W, for specific t for microstrip line characteristics.
8. Transform the lowpass filter (LPF) to a bandpass filter (BPF) using a LPF-to-BPF transformation.
9. Calculate the coupled resonator section length with fringing field end compensation.
10. Use the quasi-static analysis results in the simulation model analysis.

This broadside edge-coupled BPF topology with the GSG "coplanar" source and load ports and the via ground-cage perimeter is shown in Fig. 4. A drawback to these coupled-line structures is the requirement for small gap dimensions to achieve strong coupling. Precise control of conductor geometries and electrical properties are critical to superior performance. The elements are comprised of conductor lines and streets whose dimensional precision and uniformity determine the desired impedance matching and degree of coupling, respectively. Control of these conductor dimensions are performed using highly conformal resists applied at specific thicknesses and dimensionally compensated artwork to accommodate the vertical/ lateral growth aspect ratio of electroplated conductors.

Maximum resonator coupling for a filter occurs for quarter-wavelength elements; at 10 GHz, the wavelength is about 3 cm (1.18 in.), so maximum coupling would occur for structures on the order of 7 to 8 mm. To satisfy the desired coupling performance, the microstrip resonator geometry was varied from 5.00 to 6.00 mm long, 0.030 to 0.035 mm wide, and the coupling gap was varied from 0.130 to 0.160 mm to optimize (minimize) the value of S₂₁. Multiple calculation iterations were performed to achieve a reasonable convergence on performance specifications.

Using these calculated values, full-wave, 3D EM simulations were performed using tetrahedral mesh- ing. Model-order-reduction techniques were used for S-parameter calculation and EM-field evaluation was performed using modal analysis (eigenmode) considering losses. The analyses were performed for the discrete filter case ("open") and the in-cavity case ("closed"). Although not addressed in detail here, performance modifications due to packaging must be considered.

Measurements on fabricated BPFs were performed with a HP8510C VNA following a full two-port short-open-load-thru (SOLT) calibration. The S-parameter measurements were performed on edge-coupled, 10.1-GHz BPF structures. Figure 5 shows S₂₁ and S11 VNA traces from 9.0 to 11.0 GHz, with the full set of measured BPF characteristics shown in Table 3. The simulated filter characteristics (first iteration) of Fig. 5 showed a slightly lower insertion loss and narrower bandwidth values than the actual measured filter. For the simulations, selection of the physical dimensions (L, W, t, S) and the electrical/dielectric (ρ, e_r, tan ρ) properties for the conductors and dielectric can be different than those for the actual fabricated filter even using empirical data from specific measurements of these properties. Typically, adjustments in the properties used for the simulation in subsequent iterations will produce improved convergence of simulated and actual measured properties.

Dielectric and conductor properties, dimensions and geometries were considered in precision microstrip bandpass filter design and fabrication. Parallel-plate capacitance and resonant cavity dielectric characterization techniques were used to measure e_r and loss tangent of the 99.6-percent, 0.015-in.-thick, polished Al₂O₃. The dielectric-constant measurements using a resonator technique showed an average e_r of 9.796 and a standard deviation = 0.005 and parallel-plate technique showed an average e_r of 9.815 and a standard deviation = 0.060. These measured values were in reasonable agreement with manufacturer- reported values, however, the range of the e_r measurements was 9.7 to 10.2 and the measured loss tan (tan λ) was three to four times greater than the reported value. Conductor characterization was performed using microstrip ring-resonator structures (20.0 mm diameter). For TiW/Au (5 m) metallization schemes over 0.5 to 20 GHz, attenuation was 0.0021 to 0.0094 dB/mm for sputtered Au resonators and 0.0029 to 0.0144 dB/mm for electroplated Au resonators. Using Al₂O₃ dielectric substrates and TiW/Au metallization schemes with these nominal properties, X-band microstrip bandpass filters were EM-simulated, thin-film fabricated, and VNA tested. The measured filter characteristics were a center frequency of 10.15 GHz, S₂₁ = -0.975 dB, VSWR of 1.0:1, 1-, 3-, and 10-dB bandwidths 610 MHz, 1140 MHz, and 1590 MHz, respectively, with a shape factor of 0.034 dB/MHz (Fig. 6).

References
1. S. Haykin, Communications Systems, 3rd. ed., Chaps. 4.4-4.5, Wiley Interscience, New York, 1994.
2. E. O. Hammerstadt and F. Bekkadal, "A Microstrip Handbook"
ELAB Report, STF44 A74169, University of Tronheim, Tronheim, Norway, 1975.
3. T. C. Edwards, Foundations for Microstrip Circuit Design, Wiley, New York, 1981.
4. James C. Rautio and Veysel Demir, IEEE Microwave Theory & Techniques, Vol. 51, No. 3, March 2003.
5. Y. H. Shih and T. Itoh, Electronics Letters, Vol. 18, No. 12, December 1982, p. 1450.
6. G. Ghione and C. Naldi, IEEE Transactions on Microwave Theory & Techniques, MTT-S, 1987, p. 260.