If someone asks about the necessary requirements for 5G, then I might talk about massive MIMO and many exotic parallel DSP processing.
I may need new intelligent methods in the infrastructure for link adaptation and intelligent network slicing. However, in the RF front-end, there is a device called filter, which is very important. The filter extracts the RF channel of interest from the surrounding radio noise, and ignores all other frequencies in 5G.
This level of filters looks completely different from conventional circuits (analog or digital). They act on a piezoelectric substrate; an electronic transducer (driven by an input radio signal) at one end stimulates mechanical action, which generates sound waves. It travels to the other end, where the sound wave triggers a second transducer, converting the sound wave signal back into an electrical signal. There may seem to be more work to do, but the magic is managing these sound waves. Just like miniature instruments, the filter (and a cavity below) has a narrow range of resonant frequencies.
Everything outside this frequency range is suppressed from non-existence. For musical instruments, the resonance range depends on the mechanical design of the device-size, thickness, material and cavity. It seems that there is not much work to be done, but the magic is how to deal with sound waves. Just like miniature instruments, the filter (and a cavity below) has a narrow range of resonant frequencies. All signals outside this frequency range are suppressed. The resonance range depends on the mechanical design of the device-size, thickness, material and cavity. 2G, 3G, and 4G front ends use surface acoustic wave (SAW) filters, where the sound waves travel along the surface of the device.
These are obviously very cost-effective, but limited to frequencies below ~ 2GHz, at which point the selectivity of the filter begins to decline. This is very good for 3G, it is on the edge for 4G, and it cannot be used when 5G is used. This has driven people to switch to higher frequency, more expensive bulk acoustic wave (BAW) filters. One of the reasons for the increased cost is the complexity of designing such filters. You will find that since they are electromechanical products, they are actually MEMS devices. Even if you can’t see any motion, sound waves are mechanical deformations in piezoelectric (PE) structures. A typical filter is a PE film between two electrodes on top of the cavity. We have previously talked about the challenges when designing MEMS. There are no predefined cells or well-defined PDKs that can be reliably modeled.
There is a second problem, the sound waves will reach where they want to go. Although it seems that square or rectangular structures are the logical way to build these things, sound waves can be reflected from the ends or propagate on the surface. Both effects can interfere with ideal overall performance. Therefore, construct the structure in an interesting shape (such as an irregular pentagon) to suppress the adverse effects. In addition, it is common to build a network of resonators, and each resonator can have a different geometry.
Now that you’ve seen the problem – electromechanical 3D modeling (because you are modeling bulk and surface acoustic waves), guide the model with fancy geometries and few reference data. I was told that some of the leading companies producing these filters are still using design-fab-analyze-correct loops to optimize their designs. There is no better way. But it’s still worth it because of the huge capacity of these devices, which are needed in all 5G edge devices including mobile phones. But now there is a better way, and that is to start with a customized PDK to virtualize prototypes of these devices.
For example, using Mentor/ Tanner-SoftMEMS-OnScale solution, devices can be designed layer by layer in Ledit (this is excellent at dealing with strange shapes such as irregular pentagons). By adding material definitions, by matrix’s piezoelectric properties, thickness, process data, mechanical properties and boundary conditions, we can transform it into a 3D model. Then use OnScale’s scalable finite element analysis in the cloud to model the whole thing or a part of it. They can even model a complete wafer and observe the behavior and yield of its edges. Better virtual modeling and better analysis, all the way to chip-level analysis.
Read more articles about filter.