Recently Cavendish Kinetics, a MEMS (Micro-Electro-Mechanical-Systems) design house, issued a press release announcing a partnership with TowerJazz to produce tunable MEMS capacitors for use in mobile wireless devices. According to the release, "The process technology combines the Cavendish NanoMech MEMS technology with the TowerJazz Power CMOS process and custom RF interconnect in a single chip solution." Since I have commented in this forum on a few occasions that I think the RF signal path is a great place for analog integration, and wrote recently about analog integration and RF PAs, I spoke with Larry Morrell, Vice President of Marketing & Business Development for Cavendish, to learn more about what they are offering.
The core product that is the subject of the press release is a tunable capacitor made using MEMS technology, having 5 bits of tuning resolution in a 2x2mm package suitable for SMT (Surface Mount Technology) integration into a mobile device board. Figure 1 shows the part with a coin as a size reference.
Cavendish Kinetics tunable capacitor next to a dime. Photo from Cavendish Kinetics.
Morrell indicated that the first application they are targeting is to use the capacitor with a mobile phone antenna to allow tuning the resonant frequency of the antenna. Having been in the handset antenna industry at a time when controllable or tunable antennas were first discussed, this piqued my interest. "People have talked about this solution for a long time, but there have been issues" preventing integration into consumer products, according to Morrell. Figure 2 shows a highly simplified schematic of the application.
Integration of a tunable capacitor into a mobile device antenna. The Cavendish Kinetics device has 32 states and can vary from less than 0.5pF to about 5pF. Diagram adapted from materials provided by Cavendish Kinetics.
The concept is to add the variable capacitance into the antenna structure (for instance, in series with the shorting pin of an inverted F antenna) and by varying the capacitance value, the resonant frequency of the antenna is varied (the point of best impedance match to the RF front end). I discussed the use-models at length with Morrell, which fall into a few instances: (a) changing channels within a band; (b) changing bands; and (c) changing the conditions affecting the antenna resonance (such as moving the phone near the head). Figure 3 shows how efficiency might be optimally adapted during use across bands or channels.
Data for a two-band passive antenna compared with a two-band active antenna. Chart developed from materials provided by Cavendish Kinetics; not intended to reflect actual performance in a given device, but to illustrate the application concept.
A passive antenna designed for two bands might have efficiency as indicated by the lower black curves. Compromises in achieving the bandwidth required result in lower than desired efficiencies. The series of upper curves indicate performance of an actively tuned antenna in some of the 32 states possible with a tuning capacitor. The groups within a band are "channels," which are subdivisions of the available frequency band to allow multiple connections simultaneously. By tuning for each band as well as the channels, a frequency tunable antenna scheme can achieve higher efficiency resulting in lower power consumption and higher possible data rates.
It turns out that channel hopping is expected to be the "worst case" use model, with hops every couple of seconds. Band changes (such as driving out of a 4G cell into a 3G cell) might occur only every few minutes, and changes in the external environment might fall somewhere in between. Over a three-year phone life, this leads to around 100 million cycles of state changes for the capacitor. Morrell showed me data indicating the parts are already qualified to over 2 billion cycles.
During my interview with Morrell, we talked about other potential uses of the part, the most obvious being adaptable matching networks (L-C matching circuits in the antenna feed path). This led to a lengthy discussion about the complexities of sensing the match (such as sensing reflected power, which is doable with some PAs today) and trying to adapt at the phone while the base station is also trying to assert control (asking the phone to power up or down via the control channel). Such designs can lead to instabilities due to having two independent control loops.
For these and other reasons, Morrell thinks the initial approach will be what he termed "quasi-open loop," wherein essentially a state table is used to set the levels of the capacitor. States might take into account not only what channel and band the phone is trying to use, but also use sensors in the phone to determine when it is near a head, held in a hand, or laying on a surface. These tables could be optimized in the final phases of the phone design process, and because they are deterministic, would not lead to any stability issues.
In order for such a device to be effective in a handset, power consumption and RF system losses are key concerns. Morrell told me that the effective series resistance of the MEMS capacitor is extremely low, resulting in very low insertion loss. In addition, he stated that a 100μs response time is sufficient, and the Cavendish part can change states in 50μs, including the time to process the command and settling time to a stable value. Power consumption is 18μW holding a given state, and 180μW during state changes. The part is designed to run on 1.8V. Assuming the worst case of a state change every two seconds, the total consumption during an eight-hour day is about 0.1mA⋅h. If the efficiency gains can be achieved in actual designs, there should be a significant net gain in power budget and performance. Perhaps this will be the next case of Analog Integration Is Saving Power.