The Search For 5G mmWave Filters

Cellular telephone technology takes advantage of a large number of frequency bands to provide ever-increasing bandwidth for mobile use. Each of those bands needs a filter to keep its signals separate from other bands, but the filter technologies in current use for cellphones may not scale up to the full millimeter-wave (mmWave) range planned for 5G.

“MmWave will happen,” said Mike Eddy, vice president, corporate development at Resonant. “But the earth exploration satellite service is at 23.8 GHz, just below the mmWave bands that are being deployed for 5G, so you’re going to have to do some good filtering.”

So far, that isn’t happening. “The SAW devices or BAW devices don’t scale beyond 10 gigahertz,” said Anthony Lord, director, RF segment business development at FormFactor.

And therein lies a challenge. “None of those filters is working in millimeter range right there. Those things all topped out at like 6 or 8 gigahertz,” said Tim Cleary, senior director of marketing, RF product group at FormFactor. “The industry doesn’t have a lot of good answers.”

Surface acoustic wave (SAW) and bulk acoustic wave (BAW) filters predominate for current handsets. While they may be extended somewhat past the 6 GHz range with further improvements, it’s still a long way from the 28 to 70-GHz ranges where mmWave designs will operate. Solutions exist for implementations where space is less of a constraint, but they won’t work for handsets. That’s where new developments are required.

Band proliferation
With each new generation of cellphone technology, additional frequency bands have been opened up for use. The term “band” can mean different things, because broad bands are allocated and auctioned off, while individual channels represent bands that are subsets of those wide bands.

The number of those small bands is increasing dramatically. For channels utilizing frequency-domain duplexing (FDD), there are two adjacent sub-bands — one for transmit and one for receive — separated by a small gap to prevent interference. When time-domain duplexing (TDD) is used, then there is a single band for the full channel.

Each of these bands or sub-bands needs a band-pass filter. As the number of bands has proliferated, the number of filters needed has exploded. You can see quotes of more than 60 filters in today’s cell phones. 5G will raise that number, adding dramatically higher frequencies for mmWave bands.

A band-pass filter, in theory, will pass all signals within the band and reject all frequencies outside the band. One can think of it simplistically as multiplying a signal by 1 inside the band and 0 outside the band. Real-world filters, however, are not ideal and present more of a challenge.

Filter functionality
A practical filter doesn’t have a sharp cutoff at the edge of the band. Instead, it’s rounded, with the fall-off being sloped rather than vertical. Critical attributes are the center frequency and the upper and lower cutoffs. The cutoff is defined as the point where the ability to pass the signal drops by 3 dB (corresponding to the point where the signal power drops by half). The slope beyond the 3-dB roll-off is often referred to as the skirt, and it needs to drop as sharply as possible.

While it might be nice to engineer the three frequencies (center, upper, and lower) independently, in reality, the upper and lower cutoffs move together, making it possible to engineer the center frequency and the overall width, which will move with the center. The width is often specified as a percent of the center frequency.

Designing wider passbands can be more of a challenge, and some of the 5G bands have widths that can be as much as 20% of the center frequency. This presents a significant burden for filter design.

Fig. 1: Simplified band-pass filter showing the center frequency (f₀), the low end of the passband (f_L), and the high end of the passband (f_H). The width of the passband is B. Source: By Inductiveload — Own work, Public Domain

On the front end of a receiver, the signal needs to be filtered as early as possible to keep stray signals from entering the RF chain. That means filtering right after the antenna. With massive multiple-in/multiple-out (MIMO) technologies that allow beam-steering, arrays of antenna elements will be used. In that case, each element needs a filter.

“Element spacing now is based on the spacing of millimeter waves, which means your spacing on those piers there is about 5 mm,” said Eddy. “So you’ve got to fit in that spacing.” Today that’s not possible for mmWave, so any filtering ends up being done after the mixer.

Base stations have enough space to be more forgiving on filter size, but handsets present a demanding small-size requirement. The sweet spot for small filters is likely to be 28 GHz for the foreseeable future, because this is the mmWave frequency that’s likely to be used in handsets. The higher frequencies are more likely to be used for tower-to-tower communications, and those systems aren’t as space-constrained as a handset.

“For things like base stations, we’ll be relying on ceramic and cavity-based filters,” said David Vye, technical marketing director for AWR software at Cadence. “They just won’t ever fit the space requirements inside a mobile device.”

In the early days, the 28-GHz (or nearby) band may have relaxed filtering needs. “What we continue to hear is that, for the first few years, there probably won’t be any [mmWave] filters in the handset,” said Jeb Flemming, CTO of 3D Glass. “They’re going to use the antennas to do the filtering because they’re not going to break up the band yet.”

In that case, the antenna will be a sloppy filter that’s good enough. But at some point, we’ll need actual filters for the antenna elements. The question is, how will those mmWave filters be made?

Existing filter technologies
Most filters in today’s phones use acoustic-wave technology. This involves piezoelectric materials, which deform slightly under the influence of an electric field. Conversely, physical deformation will create an electric field. So electrical signals can be converted to mechanical vibrations and vice versa. These mechanical vibrations amount to acoustic waves within a crystal.

By setting up a structure that resonates acoustically, an input signal can be applied to one end of the resonator. That input signal will consist of many different frequency components. Some are signals intended for other bands, while others are ambient noise. The very first job of the filter is to eliminate anything that lies outside the passband.

Frequency components of the signal that are within the passband will cause acoustic resonance, which then can be detected and converted back to the electrical domain on the other end of the filter. Ideally, that output will consist of the input signal with all the undesired frequencies cleaned away.

These acoustic-wave filters have the benefits of very clean passbands, very small size, and a favorable cost structure. High-volume manufacturing has pushed those costs down.

At lower frequencies, surface acoustic wave (SAW) filters predominate. With these filters, the wave is excited along the surface of the material, and it’s coupled to an output terminal close by on the same surface.

Fig. 2: A simplified SAW filter. Source: By Matthias Buchmeier — Own work, Public Domain

For higher frequencies, bulk acoustic wave (BAW) filters take over. Instead of causing a wave on the surface, this leverages the bulk of the material, resonating from top to bottom, with the output electrode underneath. This takes more complex processing, so they tend to be more expensive than SAW filters.

Fig. 3: A simplified free-standing BAW (FBAR) filter. Source: By Khpsoi — Own work, CC BY-SA 4.0

There are two basic versions of a BAW filter, and the difference has to do with how the internal standing waves are set up. A reflection is needed from the bottom back to the top, and, with free-standing resonator BAW (FBAR) filters, an air cavity does the job.

Another version uses a series of layers that look like an acoustic mirror (analogous to a Bragg reflector for light). These are called solidly mounted-resonator (SMR) BAW filters.

Fig. 4: A simplified solidly mounted resonator (SMR) BAW filter. Source: By Khpsoi — Own work, CC BY-SA 4.0

Both SAW and BAW filters are built using MEMS machining techniques. But they appear to lose steam at higher frequencies. That suggests that the industry may need to find something new for mmWave bands.

mmWave options
mmWave radio signals aren’t new. For example, radar and microwave installations already use them. But those tend to be large installations dealing with only one or two frequencies. For 5G, many more bands must be narrowly filtered — and they have to be able to fit into a cellular handset.

While SAW and BAW aren’t considered to be options, Resonant has what it calls XBAR technology, which it claims can extend the usable range of acoustic technology. The company redesigned the BAW filter from scratch, using a different piezoelectric material — lithium niobate — and placing both contacts on the top surface, similarly to SAW.

A major difference from SAW, however, is that with XBAR the contacts don’t physically move. “With SAW, the metal fingers move physically, which means for power they failed for metal migration,” noted Eddy.

Fig. 5: An XBAR prototype shown at MWC in 2019. The filter itself is the small square in the middle. Source: Resonant

“When we modeled that structure, it gave the kind of Qs, bandwidths, and power handling you need for 5G — particularly when we were focused on 3 to 5 GHz,” he continued. “Now we’re looking at about 5 to 7.1 GHz WiFi. And then it’s 7- to 9-GHz ultra-wideband. Can you take that model and use it in millimeter wave? We think we can.”

While XBAR filters hold promise, they represent a new approach at this frequency range. Two alternative well-known mmWave filter technologies are waveguide and cavity filters. Unlike SAW and BAW filters, which use acoustic waves, these use electromagnetic waves for resonance. They both have a wide range of construction options and are commonly used for microwave applications.

The sizes of these resonators are set by the frequency of interest, with sizing or spacing in the quarter-wavelength range. The higher the frequency, the shorter the wavelength and the smaller the filter. For 5G frequencies, the sizes are coming down — but not enough to fit into a handset.

“There’s a medium called a ‘waveguide cavity,’ whose height and width dictate the energy that can propagate through it,” said Vye. “Below that frequency energy doesn’t propagate, and above a certain frequency it gets into a moding issue.”

The use of resonators — often implemented as posts — helps to reduce unwanted modes. “A waveguide cavity filter has these posts inside it,” Vye said. “It acts in the same way as the ceramic filter, and it has characteristics that either stop or pass energy at a particular frequency based on the dimensions of that post. The physical dimensions between resonators will impact bandwidth, and the number of resonators will impact the roll-off. The more resonators you have, the sharper the roll-off. But then you’re adding more length to the filter, and you’re adding more material costs to the filter.”

Fig. 6: A simplified waveguide filter using posts as resonators. Source: Wikipedia user SpinningSpark

For base stations, adaptations of this technology may be appropriate because larger sizes can be accommodated. For handsets, however, these are still too large.

Microstrip filters are another option for frequencies up to about 30 GHz. With this design, microstrip lines are created on a printed circuit board (PCB) to support electromagnetic resonance. Manufacturing variations, however, are a concern, and in general the PCB material is not considered to be high quality.

“The thickness variations of the PCB will change the frequency of the passband,” said Eddy. “The changes in the permittivity of that material will change the passband frequencies. The line width variation, when you print it, will change the passband frequencies. And it varies with temperature.”

There are other considerations, as well. “Your material properties are really what’s going to drive your performance, and there are really just a handful of materials on the market,” said Flemming. “You have these very high-Q resonating ceramics, and they’re exotic and typically come with a higher price. MLCCs (multi-layer ceramic caps) are a reasonable material historically, but they start to lose a lot of steam at around 25 GHz.”

Substrate-integrated waveguides
Because mmWave frequencies have shorter wavelengths, it’s becoming possible to create waveguides in silicon or other materials. “It’s almost like MEMS in that you’re creating these channels that the microwave signals can travel through by etching out areas and then metallizing right on the silicon wafer,” explained Vye.

3D Glass creates structures within glass rather than silicon, with a photolithographic process that selectively converts amorphous glass into a crystal through exposure to UV light. The crystalline glass — effectively a ceramic — etches preferably, allowing through-hole features to be created.

“The ceramic will etch 60 times faster than the glass phase in acid,” said Flemming. “We can do cavities, but that’s a timed etch because this ceramic formation goes all the way through the glass.”

While a number of structures like inductors can be made in this manner, this approach also can be used to create cavities with resonators for mmWave filtering. “A metal line is my resonator, but I etch away almost all the glass,” said Flemming. “And so my resonator is floating in air for the most part. Since the limiting factor in 5G mmWave is the material, if I can remove the material and go to air and make it robust, that’s a win. This suspended strip line can go up to about 40 to 50 gigahertz. We’re showing something along the lines of 10% to 15% bandwidth, which is pretty broad.”

These air-filled cavities can extend up to the higher backhaul frequencies. “We are doing quite a bit of customer development in the 70 to 150 gigahertz range,” he noted. “Some people call this 5G, some people call it 6G.”

Filter design used to involve multiple manufacturing passes to optimize the performance. But there are enough variables, and the requirements are difficult enough, that simulation tools are now used so that the construction can be optimized before the filter is built.

That helps with the details, because the details matter. “How you package it and how you make connections to the rest of the circuit are very important,” said Vye. “People have really abandoned the empirical testing of designs and rely very strongly on EM [electromagnetic simulation] technology to do the design.”

Cadence worked with 3D Glass using Microwave Office for design and simulation, so it is familiar with 3D Glass’s work. “You have metal resonators inside a very low loss structure, which is air suspended by little glass pedestals, to construct very tiny filters, although not nearly as small as the acoustic wave stuff,” said Vye.

Conclusion
The economics of the glass process are enticing. Given a demand for volume, panels can be used instead of wafers. One can fit a lot of filters into a single 9’ x 9’ panel, so, while today’s work is on 6” and 8” wafers — and there’s a desire by some customers to move to 12” wafers, they see a clear pathway to lower cost.

While some exciting possibilities lie over the horizon, none of them is ready for commercial production. There is not yet a declared winner in terms of filter technology. MmWave implementations in 5G handsets are not yet happening, so there’s some time. But the pressure will be on for the industry to settle on a solid plan and roadmap, rather than some interesting ideas that will hopefully work.

Related
5G Chips Add Test Challenges
Commercial solutions will take time to ramp, but progress is being made.
Huawei: 5G Is About Capacity, Not Speed
One-on-one with CTO Paul Scanlan.
Challenges Grow For 5G Packages And Modules
5G wireless networks drive need for new IC packages and modules.
What’s After 5G
The path to 6G will require some radical changes to both infrastructure and use models.