Innovative RF Filter Technologies:
Guardrails for the Wireless Data Highway
By Robert Aigner, PhD, Director, R&D Acoustic Technologies, TriQuint Semiconductor

Introduction
Wireless broadband communication has gained tremendous popularity. However, allocated frequency spectrum is limited and the most favorable frequencies are occupied by cell phone bands, by governmental agencies or by unlicensed bands with restricted transmission range. New applications are forced to deploy relying on less favorable frequency bands that have noisy or “oversensitive” neighbors — sometimes both. The commercial viability of these new bands depends at least in part on equipment makers’ ability to solve neighborhood problems. RF filters play a key role in minimizing interference between systems operating in different bands. The selectivity of the RF filter determines how big a portion of the total bandwidth will be used — “wasted” in a real sense — for guard bands. The selectivity — respectively, the steepness of the filter skirts — is closely related to inherent losses in the reactance elements of an RF filter. Practical RF filters also show a shift of center frequency as a function of temperature, which complicates the design process.

RF filters traditionally used for cell phone applications are based on Surface Acoustic Wave (SAW) technology. The selectivity of SAW filters is good for a band at 1 GHz but degrades when the band is located closer to the upper limit of 2.5 GHz. Temperature drift is also a concern. For broadband communication systems above 2 GHz, the only solutions available are dielectric filters, waveguide filters and LC filters. Dielectric and waveguide filters are based on electromagnetic waves and/or integrated inductor and capacitor combinations. In “pure” electrical LC filters, the major losses are related to Ohmic resistance, skin effect and eddy currents in the metal conductors. Another issue related to overall usefulness is the quality of inductors, which is generally below 50 in the frequency range above 2 GHz, making it impossible to provide the required selectivity. Filters based on wave phenomena show significantly less losses than filters based on lumped LC elements.

“The Rediscovery of Slowness”
In a fast-paced industry, it is counterintuitive to settle for something less than maximum speed. However, for RF filters based on wave phenomena, it is a huge advantage to use waves with a slow velocity. Wavelength is proportional to velocity and inversely proportional to frequency. The only alternative to filters employing electromagnetic waves are those using acoustic waves in solid materials — the foundation of acoustic filter design.

The volume of a material comprising the physical structure of a SAW or BAW filter required to confine an acoustic wave is approximately a factor of 105 smaller than for an electromagnetic wave. Acoustic waves store and carry energy very efficiently and with extremely low losses. As a result, acoustic filters can be tiny in size and exhibit very low losses. Once an acoustic wave is excited in a solid material, it is easy to trap and guide the wave as needed to shape the transmission characteristics as a function of frequency. An electrical field can be converted to mechanical stress (and vice versa) by means of the piezoelectric effect, which is present in certain classes of crystalline materials. The electro-mechanical coupling between electrical energy and acoustic energy is extensively used in both directions and exhibits extremely low conversion loss.

Surface Acoustic Wave (SAW) and Bulk Acoustic Wave (BAW) Filters
SAW filters are the dominant technology for RF front-end filters and duplexers in almost all wireless phone frequencies. In terms of insertion loss, skirt steepness and relative bandwidth, SAW filters exactly match the requirements of traditional cell phone systems. This is not a coincidence; it is the consequence of defining the cell phone standards such that commercially viable filter technologies would be available to fulfill anticipated requirements. In SAW filters, a surface acoustic wave propagates in the lateral direction on a tiny chip made from a mono-crystalline piezo-material (Figure 2). The transducers which generate and re-convert the wave consist of micro-structured metal lines on top of the substrate surface. The pitch, line width and thickness of the transducer structures are the main factors determining the center frequency of the filter and the shape of the passband.

In a BAW device, the acoustic wave propagates in a vertical direction. A quartz crystal is the substrate used for the simplest example for a BAW device (Figure 3). Metal patches on the top and bottom side of a slab of piezoelectric material excite the acoustic waves. The wave bounces back from the top and bottom surface and forms a standing acoustic wave. The frequency at which the resonance occurs is determined by the thickness of the slab and mass of the electrodes. The upper frequency limit for a typical quartz crystal is around 50 MHz, which is defined by manufacturing issues when the quartz plates become too thin to manipulate without excessive breakage.

In order to utilize the same principle in the GHz frequency range, the thickness of the piezo-layer must be in the order of micrometers. As a consequence, the resonator structure must be built utilizing thin film deposition and micromachining methods. The processing is done on a carrier substrate. In order to prevent acoustic waves escaping into the substrate, a cavity is etched underneath the active structures so that suspended membranes are created. The resulting device structure is referred to as “thin film resonator” or a film bulk acoustic resonator (FBAR). An alternative embodiment design approach that also avoids acoustic leakage involves using acoustic Bragg reflectors instead of an etched cavity. An acoustic reflector can be built using a stack of thin layers with alternating stiffness and mass density. The resulting device structure is called a “solidly mounted resonator” (BAW-SMR).

SAW technology serves classic cell phone applications (all four GSM bands and all CDMA bands except the US-PCS band) very well. The manufactured volume of SAW filters exceeded 5 billion units in 2006. SAW technology is very mature and every aspect of the manufacturing process is optimized to achieve aggressive cost targets. The US-PCS band, with its narrow transition range of 20 MHz between transmit (Tx) and receive (Rx) bands, provides challenges which are difficult to overcome with conventional SAW technology. The two flavors of BAW (BAW-SMR and FBAR) have successfully filled this void in recent years, granting them a place in the wireless phone market.

Comparison of SAW and BAW
Deciding which filter technology is right for a certain application is usually a balancing act between performance, size and cost. In terms of performance, there are several disciplines in which technologies compete:

• Maximum achievable filter bandwidth as a percentage of the center frequency
  (relative bandwidth)
• Insertion loss in the passband (in particular at the edges of the passband) and steepness
  of the filter skirts
• Temperature dependency of the filter characteristics: temperature coefficient of
  frequency (TCF)
• Flexibility in port impedance and port configuration; for example: single-ended input,
  differential output
• Power handling capability and ESD robustness

In the categories of relative bandwidth and flexibility to accommodate different port configurations, SAW technology is clearly the winner at frequencies up to 2 GHz. Different bandwidth requirements are accommodated in SAW designs by choosing a suitable piezo-material with a certain crystal cut angle in the raw wafers. Choices range from very low bandwidth materials, for example, quartz and langasite, to medium bandwidth materials such as lithium tantalate. A typical high bandwidth material is lithium niobate. As a general rule, the higher bandwidth materials show larger temperature dependency and higher losses. SAWs have an inherent advantage when it comes to impedance conversion and arbitrary port configuration because these are determined by the transducer mask layout and do not require more complex processing. SAWs also have the ability to include a “balun” function, which can be used to create a differential output signal from a single-ended filter input; a widely used and practical advantage. For SAW it is possible to integrate filters and duplexers for different bands on one chip with little or no additional processing effort.

SAW technology approaches practical limits at 2.5 GHz because the requirements for line width and gap dimensions in the transducers call for less than 0.25 micrometer lithography resolution. Manufacturing such a structure requires efforts and investments that are commercially difficult to justify. SAWs with a relative bandwidth of larger than 0.5% show a significant temperature dependency. For example: the most widely used SAW substrate material is lithium tantalate, which will exhibit a TCF on the order of -45 ppm/C. The resulting frequency shift at -30C and at +85C must be accounted for by adding temperature margins to the filter characteristics. Filters in the transmit path are challenged by power handling requirements. The current densities in the tiny metal fingers are significant and coincide with mechanical stress. This gives rise to metal migration effects in the fingers which will destroy the device over time. A carefully designed SAW filter for 2 GHz will have a mean time to failure (MTF) of >10000 hours for continues 1W (30 dBm) input power at +55C ambient temperature. Higher levels of power or higher operating temperatures are very difficult to accommodate.

The BAW principle has inherent advantages with regard to losses. Acoustic energy density is very high in BAW designs and the waves are very well trapped. The Quality Factors (Q-value) that can be achieved with BAW resonators are superior to any other technology suitable for the GHz range. Q-values of 2000 at 2GHz represent the state of the art for FBARs and SMR-BAWs. As a result of the high Q-values, the filter skirts will be very steep while the insertion loss remains low even at the edges of the passband. This is a key advantage for duplexers in the US-PCS band and the main reason FBAR and BAW were able to conquer a large market share in this particular application. There are no tiny electrode fingers in a BAW resonator and therefore, the limit for power handling is defined by exceeding a temperature limit rather than by electro-migration effects. The long term power durability can be pushed up to 4W (36 dBm) at 2 GHz with moderate effort. With regard to ESD robustness, a BAW device is by far superior. BAW-SMRs also have significantly less temperature dependency and exhibit particularly favorable TCF compared to SAW, typically -20 ppm/C. All this having been established and proven in the marketplace, the most important advantage of BAW-SMR is the fact that frequencies up to 6 GHz can be addressed without running into practical manufacturing limits. The thickness of the layers to be deposited scale with 1/f, while the size of a BAW resonator scales with 1/f2 . Both parameters make it favorable to use BAW at high frequencies, but conversely, make it hard to compete at low frequencies with SAW products.

BAW-SMR and FBAR require a complex manufacturing process with a factor of 10 more processing steps than SAW. Even though material costs for both filter types are about the same, and even though BAW-SMR / FBAR are manufactured on larger wafer sizes (SAW on 100mm, BAW on 150mm or 200mm), the inherent cost per filter is much higher than for a SAW. As of today, the only thin-film piezo- material with proven manufacturability is aluminum nitride (AlN). The piezoelectric effect in AlN is relatively weak and as a consequence, the relative bandwidth of FBAR and BAW-SMR is limited to about 4%. This is just enough to handle most of the cell phone applications well, but it is too little for certain broadband wireless applications such as WLAN or WiMAX, where the passband can have up to 15% relative bandwidth. The other significant limitation of the current generation of FBAR and BAW-SMR is the lack of flexibility to transform impedance or to provide a built in “balun” function. Another disadvantage is that, while in theory it is possible to build monolithic BAW devices (which would cover more than one frequency band on a single chip), such an implementation is prohibited by practical and commercial reasons. While the active area of a SAW filter is slightly larger than a BAW between 1 and 2 GHz, a SAW solution for multi-band applications typically recovers the lost space by using monolithic integration.

Both SAW and BAW have specific strengths and weaknesses. For the most part, they complement each other. The number of applications in which they compete against one another is very limited. It appears that any controversy regarding whether SAW or BAW will dominate the filter market has ceased since major SAW players have acquired BAW capabilities. It is relatively simple to map out the application space for SAW and BAW for near term opportunities (Figure 5): BAW will expand the ability to serve high frequency and power applications through its ability to satisfy the requirements of high performance filters.

Innovations in the Works
There is a gap between the performance conventional SAWs can deliver and the boundary at which BAW becomes commercially reasonable. Some of the recently added WCDMA bands fall into this gap, which is best filled by offering an upgraded SAW process that, in particular, improves the temperature dependency of SAWs. Temperature compensated (TC) SAWs reduce the temperature margins applied during filter design by 50% which, in turn, enables the technology to meet more challenging specifications. Creating a TCF in the range of -20 to -25 ppm/C is a realistic goal for TC-SAW. The cost to manufacture a TC-SAW filter is higher than for a conventional SAW, but is still less than for a BAW filter. The frequency limitation for SAW remains the same.

In the 40-year history of SAW, approximately 100 different combinations of piezo- materials and cut angles have been characterized, but the top 10 cover over 95% of the manufacturing volume. Despite the long history of SAW and the fact it is considered a “mature” technology, some promising new combinations have recently been discovered. Such discoveries were primarily enabled by innovations in numerical modeling of acoustic propagation, including the piezo effect. Virtual prototyping allows one to judge whether a certain combination of material and cut angle will yield the desired results. One of the recent discoveries is that lithium niobate at a cut angle of 19°Y can deliver SAW filters with a relative bandwidth up to 20% at frequencies below 2 GHz.

The quest for the next generation of BAW is to stay ahead in performance and explore directions no other technology can pursue. Most importantly, the Q-values must continue to rise. One of the upgrades of a BAW-SMR is referred to as BAW-CRF (coupled resonator filter). In this approach, two BAW resonators are stacked on top of each other such that there is a degree of acoustic energy exchange between the lower and the upper resonator. The implementation of such a structure is extremely challenging, but the benefits are significant for future applications. A BAW-CRF can deliver filters with a relative bandwidth anywhere between 1% and 15% at frequencies up to 6 GHz. A BAW-CRF features built-in “balun” functionality and extremely high rejection in the stopbands.

Another application in which BAW excels is in the area of narrowband filters with extremely steep skirts and essentially zero temperature drift. Filters of such kind are the last resort for engineers who need to fix interoperability problems between adjacent bands. The number of reported problems in this category is increasing dramatically in recent years. Examples include the WCS band at 2.3 GHz and the planned re-assignment of analog TV bands at 700MHz. In BAW-SMR there are known methods to obtain temperature compensation in a straightforward way. Temperature coefficients |TCF| < 1 ppm/C are feasible with little additional effort if 2% filter bandwidth is sufficient. The long term goal is to maximize the bandwidth that can be achieved in a BAW-SMR with complete temperature compensation. A filter with zero TCF, 4% relative bandwidth and low losses will find widespread use.

The ‘Holy Grail’ of Filter Design...
Some products exist only in a designer’s fantasy, based on a belief that certain performance is achievable, or so desirable that the complexities of designing such a product will one day be overcome. Such a product, a “Holy Grail” of RF filters, is a low-loss filter with a small size that has an electrically tunable center frequency. Tunable electromagnetic-cavity filters for bench top experiments exist, but they are the size of a shoebox. Filters based on tunable capacitors of various kinds suffer from excessive losses, if not in the capacitors, then definitely in the inductors needed in the resonance elements. Unfortunately, there is no simple way to change the frequency of acoustic filters on the fly. The tuning range achievable with existing methods and materials is ridiculously small. The piezoelectric effect is not strong enough to create static deformation or stress big enough to change the frequency of a SAW or BAW substantially. Even if a DC bias voltage of 50V is applied across a resonator, the frequency change is typically less than 0.5%. However, new compounds belonging to the class of electro-active materials show a much more pronounced effect and may in the future provide a solution to build electrically tunable acoustic filters. A tuning range of 5% would draw attention, but 20% would be truly disruptive; this would have a major impact on the architecture of future RF systems. But, in the meanwhile, the combination of SAW filters for lower frequency applications and BAW for high frequency needs continues to meet the demands of present and next-generation communications designs. TriQuint is committed to offer a complete portfolio of leading edge SAW and BAW filter technologies along with active RF components.
For more information, click below

TRIQUINT SEMICONDUCTOR
www.triquint.com
TXTLINX.COM78
Email this article to a friend!