Technical Insight
Building better RF front-ends with UltraCMOS Technology
When it comes to efficient delivery of power to the antenna, UltraCMOS technology is now outperforming GaAs
Peregrine's UltraCMOS Global 1 is the industry's first reconfigurable RF front-end system.
BY DUNCAN PILGRIM FROM PEREGRINE SEMICONDUCTOR
In today's typical LTE-capable mobile handset, the incredibly complex RF front-end is long overdue for rationalisation. To blame is the multi-die, multiple-technology approach that can be "“ and demonstrably, has been "“ made to work to an acceptable level of performance. What's needed is a new approach that can trim the costs associated with such a diverse bill-of-materials, and with the complexities of manufacturing and testing the air-interface.
The way to address this, as is invariably the case with semiconductor technology, is to move to further integration. Currently, a typical handset is packed with ICs from a range of different vendors, each sporting different technologies. One of the key elements "“ the stand-out item, it might be said "“ that lies in the way of fully integrating the RF front-end is the status-quo of the GaAs-based PA.
This amplifier has been refined as mobile wireless standards have evolved. There has been a rapid growth of spectrum allocations over time, and that has brought us to the present state-of-the-art loosely termed "˜4G'. Now there are over 40 distinct frequency bands throughout the world, and that is just one aspect of a multi-dimensional array of variables. There are also differences in modulation schemes, operating modes for the PA, antenna tuning settings, and accommodation of down-link carrier aggregation. The latter, like almost everything else in the mix, has been brought into play to serve the ever-increasing demand for data download capacity, while retaining backward capability with prior-generation air interfaces.
Multiplying all these factors together in a simple-arithmetic manner suggests that there has been a 5000-fold increase in possible operating states that the PA/RF front-end might occupy. Naturally, among those many states there will be groupings that will be quite similar; but equally, there is great diversity in operating conditions.
What is clear, therefore, is the need for a reconfigurable front-end. At Peregrine Semiconductor we are meeting this need with our UltraCMOS technology, which is a patented, advanced form of silicon-on-insulator (SOI).
Peregrine's UltraCMOS 10 technology addresses the unique growth requirements for mobile applications and is the foundation of Peregrine's next-generation RF switches, tuners and power amplifiers, including UltraCMOS Global 1.
Peregrine's state-of-the-art processes ensure uniformity and quality of high-end RF solutions.
The benefits of this technology have already been seen in the RF domain. In RF switches "“ to cite just one key function "“ CMOS was initially thought of as incompatible with RF signal paths. However, that's not the case now: devices fabricated in SOI have progressed rapidly from an initial demonstration to providing the required low losses, high linearity and high isolation, and this has made them the default choice of switch in a complex reconfigurable design. Switching "“ of passive component values "“ also underpins integration of functions such as antenna tuning.
Progress of silicon over the past decades is also evident from the continuing incursion of silicon CMOS into successive areas where it was initially thought of as fundamentally unsuitable. Looked at in this light, the objective of converting the power amplifier in a mobile terminal to CMOS, and then integrating it along with the other RF front-end functions on a monolithic device, can be viewed as obvious and attractive "“ although this move draws scepticism from analysts and handset makers.
CMOS credentials
Ignore, for a moment, the task of building the PA in CMOS. This leaves the challenge of integrating a reconfigurable RF front-end in a process such as our UltraCMOS 10 technology as the main challenge, and that is less formidable. After all, CMOS technology has already demonstrated its capability to build elements such as RF switches "“ in the complex multi-way, multi-throw configurations that 3G and 4G terminals require "“ and antenna tuners, which rely on the same switching technology.
The capability of this switch is evident in the widely used figure-of-merit for RF switches: The product of the on-resistance of the switch in its "˜on' or conducting state, and the capacitance it presents in its "˜off' state. The on-resistance determines the signal loss through the switched channel, so the smaller it is, the better. Meanwhile, the off-state capacitance sets the degree of isolation that the switch can provide; lower capacitance means lower signal leakage, so again smaller is better.
With our UltraCMOS 10 technology "“ the latest generation of a succession of technology steps that have employed diminishing feature sizes and constant refinements in performance "“ the key figure-of-merit formed from the product of on-resistance and capacitance in the off-state outperforms that of the closest competitive technologies by at least 30 percent (see Figure 1).
Figure 1: Peregrine's UltraCMOS 10 technology platform delivers significant performance enhancements, measured by the RonCoff figure of merit in fS.
The objective of antenna tuning is to improve the impedance match between the PA output (or receiver input) and the antenna element itself. In smartphones, there are a variety of compact antenna forms, which are difficult to drive and often have to be used with a relatively high VSWR. These antenna also suffer from external loading effects.
One solution that can greatly improve the RF efficiency of the air interface and thereby, the battery life of the terminal, is to alter feed points (by switching) and impedances (by switching, for example, shunt capacitances, into or out of the feed). Our digitally tuneable capacitors can realise this by supplying the necessary degree of control, offering 5-bit resolution, and delivering tuning ratios of as much as 7:1 with capacitances in the range of several pF. These properties, coupled with the high linearity UltraCMOS technology, enable the progression to a complete, reconfigurable RF front-end. On a conventional silicon substrate, a variety of parasitic effects manifest themselves. Parasitic capacitances exist between various parts of the active (and passive) devices, and in addition are voltage-dependent "“ doping profiles cause them to act as variable-capacitance junctions. Such effects are highly undesirable because they are inherently detrimental to linearity.
Building on an insulating substrate removes many of these effects at a stroke. In addition, as demonstrated by the antenna-tuning components alluded to above, as the technology is CMOS, adding control logic on the same die as the RF path is straightforward. For example, the stand-alone antenna tuning parts in our range offer standard SPI and MIPI RF front-end control interfaces.
Many of the same properties of UltraCMOS technology lend themselves to further integration. With freedom from parasitic capacitances and from stray conduction paths via the substrate, it is possible to realise a high degree of isolation between functional blocks "“ linear RF signal paths, and control logic "“ and between those blocks and interconnection paths. The remaining critical question, therefore, is this: Can a CMOS power amplifier be designed into an RF front-end and surpass the performance obtainable with a GaAs device?
Trumping GaAs
With our development of the Global 1, that goal has been achieved. In terms of PA performance, no concessions have been made to a GaAs-based circuit. The incumbent technology, the GaAs-based PA, has dominated advanced handset air interfaces for some very good reasons: It can be designed to deliver the necessary power; it is broadband, with an individual power output stage serving a broad groupings of bands within the complete frequency range that the terminal covers; and it is relatively rugged. However, they cannot be optimised, relative to operating frequency, on a per-band basis.
Global 1 not only implements an all-CMOS RF front-end, it embodies a design approach that makes the optimal use of the characteristics of UltraCMOS technology. It takes advantage of the intelligence that can be integrated on a CMOS IC, to make use of the device characteristics possible in RF SOI. Once that intelligence is added, it becomes possible to not only recognise the need for the 5000-fold increase in possible operating states of the RF front-end, but to optimise operation for every one of them. Control is implemented, via the on-chip logic, through a MIPI interface, to deliver a performance that is markedly in advance of standard CMOS (see Figure 2). Handling a WCDMA signal, UltraCMOS Global 1 has better performance than GaAs technology, while bulk CMOS lags by an uncompetitive margin.
Figure 2: Peregrine's UltraCMOS Global 1 PA surpasses the leading GaAs PAs and exceeds the performance of other CMOS PAs by 13.5 percentage points. These figures are for the benchmark for PA performance of PAE (power-added efficiency) using a WCDMA (voice) waveform at an ACLR (adjacent channel leakage ratio) of -38 dBc.
In an LTE-era smartphone, a typical PA stage will have three paths "“ in effect, three distinct power-amplifiers "“ that each cover part of the overall frequency range. They are selected for operation according to the band in use, and other operating conditions (see Figure 3). Operation of a GaAs-based PA can be optimised, at best, for each of those three frequency segments, and one set-up has to serve across the whole of each segment.
Figure 3: Peregrine's UltraCMOS Global 1 PA enables performance optimisation through tuneable matching networks; band-specific tuning provides additional rejection to other frequency bands, which helps mitigate some of the difficult interoperability cases.
With the architecture of Global 1 it is a very different story. In this case, there is a much more focussed optimisation operating at the level of the actual band in use; this may only be 10 MHz for the narrowest bands as opposed to over 200 MHz for the entire path. For each operating state, it is possible to set up antenna tuning, filtering and stage-to-stage matching. To efficiently achieve high delivered-power levels from CMOS devices, it is necessary to use a stacked configuration of several FETs, connected in series "“ this overcomes the relatively low breakdown voltage of an individual FET. This structure functions as a single device that can handle the higher supply voltage needed to reproduce the LTE waveform profile, but presents challenges in setting the bias level of each transistor in the stack. However, this is not an issue with UltraCMOS technology, which can yield consistent devices that meet this requirement. This allows several devices to operate effectively as a single high-voltage FET.
With Global 1, the PA is not limited to just competitive WCDMA performance in terms of efficiency, output power and linearity. Instead, GaAs-equivalent performance is maintained for LTE waveform allocations up to 100 Resource Blocks (RB) (see Figure 4). In mid and high bands of operation, similar competitive results have been demonstrated. The CW measurements show efficiency when the PA runs into saturation and is a good indication of the approximate efficiency that can be achieved with the addition of envelope tracking.
Figure 4: With W-CDMA, LTE5, LTE10 and LTE20 waveforms, Peregrine's UltraCMOS Global 1 PA shows little performance roll-off across the low band.
There is a clear precedent from many other semiconductor-device domains; as soon as CMOS with high-integration-levels can match the performance of a prior technology, then it quickly becomes the preferred solution. In this case, not only is the performance being matched, it is matched in a single device that is capable of being fabricated on a standard silicon CMOS line. Moreover, by covering all the operating states required for all worldwide frequency allocations and signal configurations, it will equip handset makers with what they have been wanting for years: a single stock-keeping unit that will allow them to build one PCB for all markets.
