InGaP HBT vs CMOS for mobile handset power amplifiers

Designers of mobile handsets and wireless networking products know that the power amplifier (PA) is a make or break component in terms of performance, footprint, and battery life. Selecting the right type of PA is one of the key decisions that can help achieve a compelling design that delivers a competitive advantage.

Today, most PAs for mobile applications are manufactured using a gallium arsenide (GaAs)-based bipolar transistor (called InGaP HBT, for indium-gallium-phosphide heterojunction bipolar transistor) and a relatively small number are manufactured using silicon (Si) CMOS. CMOS offers the tantalizing possibility of much higher levels of integration and lower cost as compared to GaAs and therefore the question of if and when GaAs will be supplanted by Si is often raised.

To answer that question, the designer needs to develop a clear understanding of the competing technologies for wireless applications and make a comparison in terms of realistic integration levels, performance and cost. While CMOS may be the dominant general-purpose semiconductor technology, it is by no means the best choice for every consumer application. For instance, GaAs PAs dominate the market for wireless networking and handsets because of their superior ability to support high-frequency and high-power applications with good efficiency. CMOS, on the other hand, dominates in Bluetooth and ZigBee applications because they typically operate at lower power levels and have less stringent performance requirements.

RF integration
CMOS is the technology of choice today for the highly integrated baseband and transceiver die for mobile radios due to its ability to integrate an enormous number of transistors as well as its relatively low cost at high volume. While baseband and transceiver ICs used to be two different die a few years ago, the trend now is to merge the two on a single CMOS chip with lower overall cost. It is natural to wonder if the PA could also be integrated on the same die to achieve true single-chip radios. There are, however, some practical considerations that make this integration unlikely.

Highly complex and highly integrated baseband and transceiver RFICs require the most advanced, smallest gate-length CMOS process in order to integrate functions in the smallest possible die. CMOS PAs on the other hand, have relatively few transistors to integrate but they require specialized, area consuming, passive circuit elements (resistors, inductors and capacitors) that are not usually provided in the most advanced CMOS processes. Those processes are tailored in favor of digital circuitry.

CMOS PAs use inductors, capacitors and resistors that are fundamental to RF applications but are not normally part of a digital circuit. In addition, the Si substrate used in traditional CMOS is conductive, which increases RF loss and severely degrades the performance of these passive circuit elements unless special steps, not normally a part of the CMOS process flow, are taken. These steps include raising the metal transmission lines off the surface of the Si substrate by putting them on thick dielectric layers, etching a trench below the passive structures to reduce eddy currents and other, innovative though non standard, processes and structures. These specialized processes and structures increase the cost of manufacture. Also, the large size of these structures poses a problem with integration with other components of the wireless radio.

Components such as the highly integrated transceiver are manufactured in the latest CMOS node to minimize their size and therefore the cost. It does not make economic sense to use a very expensive process to minimize the size of the transceiver circuit only to give away the benefits by adding large structures that do not require the same high performance technology on the same die. CMOS PAs are best fabricated using a process tailored to their special needs.

Another point to consider is the suitability of transistors fabricated with the latest-generation CMOS for RF applications. As gate lengths shrink with each successive generation of CMOS technology, the transistors retain the property of behaving like switches (which is what is needed for digital applications) but they become increasingly non ideal for processing analog signals (such as those encountered in the PA). In fact, it is often preferred to use an older version of CMOS for RF applications because transistor characteristics are more ideal.

The models used for circuit design are very different for RF circuits as compared to digital circuits. RF circuit models tend to be much more complicated, especially for power applications, because all the non-idealities of the transistor need to be accounted for including the deleterious effects of the conducting substrate. Accurate large-signal RF models for CMOS transistors are still not widely available. While this is not a fundamental problem, it does increase product development cost and the number of design spins needed to converge to an acceptable performance level.

Packaging is another area where the needs of components such as transceivers and power amplifiers differ substantially. Power amplifiers are typically packaged in laminate-based or low-temperature co-fired ceramic (LTCC)-based modules whereas transceivers are packaged in much simpler plastic ball grid array (PBGA) type packages. PA packages have only a few I/O ports but the module needs to accommodate low-loss RF matching circuitry and a few surface-mount devices (SMDs) like capacitors and inductors. PBGA packages, on the other hand, are designed for large numbers (typically over 100) of I/O pins and have no place for external SMDs and RF circuitry. Although it might be possible to come up with a package that meets the requirements of both types of circuits, it is not a natural fit.

Instead of struggling to integrate the PA and transceiver, a more logical integration could occur in the RF section, bringing together the PA, voltage regulation, antenna switch, low-noise amplifier (LNA), and other components onto a single die to create a complete RF front end. This die could be packaged along with filters in a single, highly functional, module. This approach simplifies handset/WLAN designs and allows the overall design to be nicely partitioned for maximum performance and lowest overall cost.

Performance challenges
While CMOS amplifiers have steadily increased in frequency and power handling capabilities, they still face challenges in satisfying the stringent performance requirements of each successive generation of wireless standards (3G, 4G, etc.). Even in applications where both CMOS and InGaP amplifiers are available, for instance, the InGaP devices tend to offer better linearity, efficiency, and harmonic performance at higher power levels. For example, Table-1 compares the high-band performance of the Si4300, a dual-band GSM CMOS PA from Silicon Labs with the performance of the AWT6166, a quad-band InGaP GSM PA from Anadigics. It is clear that the InGaP PA offers higher gain, higher efficiency, better isolation, lower harmonics and lower receive band noise than the silicon counterpart.

Page 2: Linearity
Page 1 | 2