диафрагмированные волноводные фильтры / 4435b183-3308-4f56-8231-f48188cdee04

Abstract—Two millimetre-wave filters (a Chebyshev 4-pole and a quasi-elliptical 6-pole 2 zeroes realized with dual-mode resonators) centred at 39 GHz are presented in this paper. They are both obtained by laser machining Alumina substrates and metallized with an electroless Copper plating technique. Laser etching is finally used again to etch the different patterns required for their input and output as well as other features. Despite the simplicity of this method, good agreements are obtained between full wave simulations and measurements, validating the proposed approach.

Index Terms— Q-band filter, laser micro-machining, millimetre wave, Alumina, ceramic.

I. INTRODUCTION

Facing the increasing demand of the incoming millimetrewave applications for 5G, Wigig or satellite communications, the possible technologies for filling their filtering specifications have been extensively studied. Among the possible solutions, we can find the well-established LTCC technology [1- 3], the SIW components [4-5], Silicon wafer micro-machining [6], metal micro-machining [7,8] E-plane metal septa in machined waveguides [9] and many others. These different techniques and their associated performances are therefore providing different balances between these millimetre-wave filters key requirements: size, insertion loss, price and performances other temperature for the most demanding ones.

This paper proposes here to evaluate the concept of micromachined via-less SIW dual-mode filters [10] for Q-band high order bandpass filters following 3 steps: laser cutting of low thickness Alumina substrate, electroless copper plating and finally laser etching of lines on the filter top face. This paper therefore aims at studying the credibility of this simple, cost effective and somehow generic technique of the fabrication of mm-wave filters.

II. DESIGN OF THE Q-BAND FILTERS

For the purpose of this manufacturing approach, the following specifications have been arbitrarily chosen as representative of Q-band bandpass filters: an operating frequency f0 = 39 GHz, a bandwidth of 1.5 GHz minimum with insertion loss as low as possible and an isolation better than 20 dB for f0 ± 2 GHz. The filter size should be as small as possible.

Two filters have been designed in order to satisfy these specifications: a 4-pole Chebyshev and a 6-pole quasi-elliptic filter.

Their synthesis follows the conventional guidelines of coupled resonators filters. Nevertheless a particular attention has

been paid to the choice of resonator which has to be a TEn0m mode since no resonance can be expected in the substrate thickness. TE102/201 (dual-mode) has been chosen due to its better behaviour toward fabrication dispersion: at 39 GHz, for a permittivity of 9.75 (the permittivity of the Alumina substrates used for this work), a TE101 resonator measures 1.74 mm x 1.74 mm, and a TE102/201 mode measures 2.751 mm x 2.751 mm. When considering a change of 10 μm (expected manufacturing tolerances from the laser cutting) in both lengths of each resonator, it results in a variation of nearly 220 MHz for the TE101 and 140 MHz for the TE102/201 mode, which gives a clear advantage to the second mode. Of course this shift could be lowered to 100 MHz for a TE103/301 mode but frequency isolation should become much more limited in that case. In addition a filter made with TE102/201 will be a bit larger and so, a bit less brittle than with smaller resonators. This mode is thus a good balance between size, mechanical behaviour during the laser machining, Q and spurious free range.

The 4-pole Chebyshev filter coupling matrix is given in Table I. The used substrates have a relative permittivity of 9.75 and a loss tangent of 0.0033 at 40 GHz. Since manufacturing accuracy is linked to substrate thickness, 0.25 mm thick substrates will be used. Thicker substrates would bring better Q but the available laser process produces a less precise cutting and would require more time to process.

The conductivity of the obtained metal plating is evaluated to be 10 S/μm at this frequency. Considering these technological data, quality factors are expected to be around 150 for the TE102/201 resonators mentioned above.

TABLE I

COUPLING TERMS OF THE 4-POLE FILTERING FUNCTION

TABLE II

COUPLING TERMS OF THE 6-POLE FILTERING FUNCTION

EM-simulations were done using ANSYS HFSS and responses of the final structure are given in Fig. 1. Overall dimensions are 2.770 mm x 6.084 mm x 0.25 mm.

Fig. 1 Scattering responses and top view of the simulated 4-pole filter.

Fig. 2 Scattering responses and top view of the simulated6-pole filter.

For the quasi-elliptical topology, a 6-pole filter with two transmission zeroes is used to fill the specifications. However, since it would be difficult to properly couple resonators 2 and 5 to create the zeroes (considering a folded topology), a coupling between 1 and 6 is chosen instead. Rigorously, 4 transmission zeroes can thus be created, but only 2 are transmission zeroes. As a consequence, the objective matrix has been obtained by optimization and its values are given in Table II.

It has to be noticed that two box sections could have been designed, but the resulting filter with additional couplings path (realized the same way as 1-6 in Fig. 2) between resonators 1- 4 and 3-6 would have led to something even more brittle with small and therefore fragile details.

Coupling between resonators 1 and 6 is done thanks to a line, however its resonance frequency has to be close to the filter passband. We then choose to make it resonate near the lowest transmission zero of the filter, resulting in a bump that can be seen around 36 GHz on S21 and in a sacrifice of selectivity in this area.

Corresponding responses are given in Fig. 2. The overall dimensions are 5.86 mm x 5.86 mm x 0.25 mm.

III. FABRICATION PROCESS

Filters are firstly cut in the selected Alumina substrates using a YAG laser with a 25μm laser spot. They are then copper plated thanks to an electroless process which easily plates all the filter surfaces (top, bottom and side surfaces) with a typical thickness between 2 and 3 μm. These surfaces are then gold covered by electrolysis, adding here an extra thickness of 0.2μm in order to avoid the oxidization of the copper layer. We expect here to have a metal coverage of the ceramic resonators with a thickness about three times higher than the skin depth at 39 GHz. A final step is then needed to laser etch the chosen coplanar accesses needed for the input and output of the two filters. This last etching step is also necessary for the quasi-elliptic filter for the patterning of the electrical line coupling the resonators 1 and 6. The same laser is used here for that purpose. No post-fabrication tuning has been applied.

That type of coplanar accesses, also thanks to the low thickness of the filter, could make that filters easily wire-bonded or even flip-chipped. These options would however need to be included in the initial design steps since these techniques would slightly changes the filters input/output impedances.

Photographs of the fabricated filters are given in Fig. 3.

Fig. 3 Photography of a fabricated 4-pole filter (left side) and of the fabricated 6-pole (right side).

Dimensions analysis, made on 3 copies of the 4-pole filter, has revealed an average manufacturing tolerance of ±10 μm on resonators lengths, with a maximum deviation of 27 μm. The patterning of the different lines on the filter top face (see Fig. 3) is very satisfactory with geometrical discrepancies around ±3 μm and with a maximum deviation of 15 μm. In order to be sure to clear the etched gaps of metal, over-etching along the Z axis (along the substrate thickness) has been applied, resulting in a 28 μm over-etching between the top of the metal layer and the bottom of the etched gaps.

IV. MEASUREMENTS

A. 4-pole filter

Measurements of the three copies of the 4-pole filter are shown in Fig. 4. They have been realized with GSG probes without being enclosed inside a conductive enclosure.

Fig. 4 Measurements of the fabricated 4-pole filters and comparison with EM-simulations.

Results are in good agreement with simulations despite a frequency shift of 300 MHz.

Insertion losses are around 3.5 dB confirming the EMsimulation losses level and a quality factor of 150.

The three filters have their return loss better than 15 dB and we can see in Fig. 4 that the filters responses are very close to each other’s, demonstrating a good repeatability of the process.

B. 6-pole filter

Only one copy of the 6-pole filter has been measured. The two others, even if successfully cut by laser have been broken during the metallization process. Measurements are given in Fig. 5.

Fig. 5 Measurements of the fabricated 6-pole filter and comparison with EM-simulations

Despite some visible defects in filter (lower left corner of the 6th order filter in Fig. 3) simulations and measurements

agree well except for the isolation around 37 GHz. The resonance of the electrical line coupling resonators 1 and 6 was experimentally higher than expected (+ 1 GHz).

V. DISCUSSION

Table III provides a comparison with other similar compact filters in Q-band, this table showing that LTCC and PCB technologies are typically used for this frequency band.

The proposed approach provides equivalent unloaded Q compared to regular PCB technology as seen in [4] and [13].

Air-filled SIW filters and filters made by LTCC technology provide better Q than our solution but need a multilayer approach and therefore a bit more complexity for their manufacturing.

In order to improve the unloaded quality factor of the proposed solution, a higher purity Alumina substrate and a better quality metallization must be used. For example, using a ceramic substrate with a loss tangent of 1.10-4 and a metal plating with a conductivity of 20 S/μm would lead to an unloaded Q of about 350 at 39 GHz. With these characteristics and a filter thickness of 0.5 mm, an unloaded quality factor of 600 could even be reached.

TABLE III

COMPARATIVE TABLE WITH TECHNOLOGIES FOR Q-BAND AP-

PLICATIONS

VI. CONCLUSION

Two filters in Q-band have been designed and manufactured. Their manufacturing is quite simple and can be decomposed in 3 steps: contour cutting by laser, metallization by electroless Copper plating and patterning of the top lines by laser etching. Despite this simplicity, good agreement between simulations and measurements has been obtained, showing the robustness and validity of the approach thanks to the typical accuracy of ±10 μm of the laser cutting technology.

Since the proposed approach is also generic, performances could be easily enhanced just by changing for another substrate with a better loss tangent, a higher thickness and/or by changing the metallization process to get a better conductivity. In addition, a true temperature stable material could also be used if no change in the operating frequency is required.