Who said you can’t put a cell-phone power amplifier on a CMOS chip?
According to market research firm In-Stat, worldwide sales of mobile phones will rise from 935 million units in 2006 to more than double that in 2011 (“The Big Trends For Cell Phones, 2006-2011,” May 2006). With cell-phone manufacturers now producing more than 1 billion units annually, this has become one of the most competitive markets to be addressed by the semiconductor industry.
Because cost is king, manufacturers are under increasing pressure to produce chipsets that are smaller and more cost-effective while offering broader feature sets. The big problem is determining how you integrate a power amplifier (PA) on the same chip with all the other cell-phone circuitry.
The CMOS PA Challenge
Some experts say that specialty processes such as gallium arsenide (GaAs), laterally diffused MOS (LDMOS), or silicon-germanium (SiGe) bipolar CMOS (biCMOS) with less precise geometries may offer the short-term cost advantage and linear modulation demanded by manufacturers and designers.
However, the economies of scale inherent in silicon CMOS have driven the semiconductor industry to make significant investments in this process technology, creating capabilities and capacity that have outlasted and will continue to outlast any niche process offerings.
For example, transceiver blocks previously required specialty biCMOS processes from vendors such as Infineon, NXP, and Skyworks. But they have long since been implemented in CMOS and, in some cases, integrated with the handset’s main processor inside a system-on-a-chip (SoC). Designers have repeatedly found that implementing an analog block in standard CMOS has paid off in the long run, despite the hurdles of implementing challenging circuit blocks in a less forgiving process.
Despite these advances, though, the CMOS process has not been able to successfully penetrate the PA block, which continues to remain a key element within the cell phone. It is still considered the most difficult block to implement both as a component and as part of the final application.
Up until now, the PA block has been developed using a specialty GaAs or LDMOS process coupled with a hybrid module packaging technology—in total an expensive manufacturing flow, which has made it a substantial part of the cell-phone bill of materials. The specialty semiconductor process is required to provide a high-gain, high-frequency transistor element with a high breakdown voltage. The hybrid packaging technology provides high-Q passive components to generate the 50-Ω matching circuit.
Implementing in standard CMOS means the designer has to live without enhanced transistors and high-Q passives. This makes the development of a fully integrated PA extremely challenging. A smaller challenge is simply transferring the active core from a specialty process to CMOS, but the cost savings simply aren’t sufficient.
In current GaAs implementations, a high percentage of the cost is incurred in the manufacturing of the multichip module, with its expensive substrates and requirement for the sourcing and assembly of surface-mount (SMT) components. Integrating the 50-Ω match on die, therefore, was necessary to eliminate the need for the module.
Analysis of the Problem
A first step in taking on the challenge of implementing a high-power non-module PA in CMOS, specifically a quad-band GSM/GPRS device, is to examine the conventional techniques employed to realize existing PAs.
In the cell-phone environment, operation must be achieved from a battery that produces a relatively low voltage. Nominally, nickel-metal-hydride (NiMH) or lithium-ion (Li-ion) cells produce a nominal voltage of around 3.6 V. Printed-circuit board (PCB) trace resistance, coupled with the order of magnitude of current drawn, means that the available voltage at the PA is around 3.5 V. Without the use of a load impedance transformation, the maximum power that can be generated into a 50-Ω load is provided by the equation:
Given that the GSM standard requires output power levels of around +35 dBm or greater than 3 W, some kind of impedance transformation is required. The normal implementation of an impedance transformation places a resonant match at the output of the main power stage (Fig. 1).
Resonant matching structures require high-Q elements to maintain a reasonable passive efficiency, making it difficult to implement them on a typical CMOS process where Qs in the range of only 5 to 15 are achievable.
Another approach is to use a transformer type match. A transformer has the benefit that inductively stored energy is low compared to a resonant match, meaning that a transformer structure can have lower Q and still perform the impedance transformation to the same degree of satisfaction (Fig. 2). Simply using a conventional transformer structure does not solve the problem, as the primary and secondary windings require inductance values that aren’t easy to implement.
The Ultimate Solution
The transformer idea can lead to a solution with some modification. This technique is known as a distributed active transformer (DAT) technology. It employs a unique geometry that enables a relatively low-Q semiconductor metal to be used to provide a transformer-based matching circuit.
First it was observed that the Q of a metal slab considered as an inductor was significantly higher than a multiturn spiral—in fact, 20 to 30 times higher. The challenge presented, then, was determining how to connect an active circuit to it so the increase in Q wasn’t instantly lost due to the parasitic effects of the connections themselves.
The solution lies in distributing the PA core into several blocks and combining the power using the transformer structure (Fig. 3). Each power core is differential in nature, which, among other advantages, allows one terminal of a slab inductor (as described above) to be connected to it and the opposite phase of a neighboring amplifier core.
A tuning capacitor may be connected in parallel with the output of the differential pair, providing a resonant structure that may be tuned for frequency response. Placing a single-turn secondary loop in close proximity to the primary slabs creates a transformer structure with high passive efficiency. The power from each core is combined, and the impedance is transferred relative to the ratio of number of distributed primaries to the single secondary.
The primary is between the positive supply and the drain of the transistor, as the positive supply is an ac ground. Therefore, each physical slab is two primaries. The voltage transformation ratio is then 8:1. As power is conserved in the transformer, assuming no loss, power in the primary is equal to power in the secondary. Therefore, as P = V2/R, the impedance ratio is the square of the voltage transformation ratio, making it 1:64.
This technique addresses several other major implementation problems. First, the use of a differential amplifier means that a low-impedance ground isn’t required. Of course, this isn’t the case in single-ended implementations. There is also a virtual ground at the supply connection. In traditional single-ended implementations, a high-value supply choke is required.
Furthermore, as both supply and ground are then ac grounds, sensitivity to wire bonding variation is significantly reduced. As the amplifier cores are distributed about the die, thermal hot spots are more spread out, making heat dissipation less of an issue. And, as the output connection had to be singled-ended, simply grounding one of the outputs of the secondary provides a neat differential to single-ended conversion.
Low current drain is always a key design requirement for mobile communications components, and the PA is probably the most critical due to its high power output and hence power-consumption requirements. It was therefore a design goal to use a highly saturated switching amplifier to maximize achieved efficiency. A new amplifier class, class E/F, was created, so the highest efficiency was obtained while also maintaining good control over harmonic generation.
As the battery may generate very high supply voltages while it is either charging or recently fully charged, designers need a deep understanding of the breakdown mechanisms experienced when using CMOS transistors. A fully charged battery may produce voltages up to 5.5 V, which the PA must handle reliably.
A 0.13-μm CMOS process from TSMC was selected. Before design began, the transistor primitives were characterized for several failure modes, including gate oxide breakdown and hot carrier degradation. In one implementation, several innovative circuit techniques were used that allow the device to withstand supply voltages as high as the specified 5.5 V while still producing full output power.
Unlike GaAs equivalents, with the power core now on CMOS, the small signal control circuitry required to bias and regulate the power of a GPRS-type power amplifier may now be integrated in the same die as the main power stages, further reducing the cost of the whole subsystem (Fig. 4).
A commercial implementation of this design is Axiom Microdevices’ first product, the AX502. This quad-band GSM/GPRS device integrates a closed-loop-type power controller on the same die as the power core. The power controller measures the output power by using two on-chip sensors to calculate power, compares the measurement to a known reference, and sets the power via the previously mentioned actuation point. To operate under changing load conditions while still complying with GSM specifications such as transient spectrum and output power variation, the AX502 uses its power control to estimate the load impedance and compensate for variation.
Although integration of small signal circuitry on the same die as the PA is now possible, there are still challenges to overcome. At its maximum operating point, the PA produces RF voltages in excess of 20 V, which couple to almost every point on the die. Careful design and layout were required so the small signal circuitry may operate reliably in this environment. It was like trying to jump on a trampoline at the same time as a 600-pound gorilla.
All voltage regulation to provide stable supplies for the power control is also integrated on the same die to allow the device to operate all circuitry directly from the battery supply. Low-power biasing circuitry had to be created to provide references to the voltage regulators generated from the battery supply, which could still withstand high battery voltages without breaking down.
In the realization of this device, Axiom found that traditional simulation techniques weren’t sufficient to predict the large signal RF performance of complex passive and active structures. Computationally intense simulation tools to predict passive performance, which had been traditionally deployed for non-IC applications, had to be integrated into a more conventional IC design flow.
With requirements for slim form factors becoming more important, Axiom selected a low-profile micro lead frame (MLF) package, which is only 0.9 mm in height. This package also is very low cost by virtue of its simple lead-frame chip carrier. It has excellent RF and thermal properties as well, as it was originally designed with a high-power millimeter-wave monolithic IC (MMIC) application in mind.
The DAT-based CMOS PA technology is now proven within a standalone GPRS product, through the large-volume production shipments of Axiom’s AX502 device. Axiom has now shipped tens of millions of this device to handset customers.
An added bonus of designing in CMOS as compared to a GaAs module, and having strictly followed the foundry design rules, is the device reliability achieved. Axiom’s design team carried out extensive reliability testing and found zero failures. This reliability has also carried forward into production shipment, with customers so far also reporting zero failures.
Having the CMOS power amplifier proven as a standalone product provides the potential for future integration, which will continue to help the semiconductor industry continue to address and fulfill the demands of manufacturers and designers alike for chipsets that are smaller and more cost-effective while offering broader feature sets.
Donald McClymont is the vice president of marketing with Axiom Microdevices Inc., Irvine, Calif. He can be reached at firstname.lastname@example.org.
- Power Amplifiers Crank Up Mobile Device Performance
- Fifth-Generation PAs Extend Talk Time
- UHF Power Amplifier Boosts 290 To 320 MHz
- High Power Amplifier Is Broadband
- New WiMAX Power Amplifier Modules
- Report Indicates Increase In CMOS PA Suppliers