Power density in active devices is increasing according to the demands of transistor users. Applications in commercial wireless, avionics, broadcast, industrial, and medical systems are pushing the envelope for solid-state power, with growing requirements for higher output power levels from fewer output-stage devices. At Freescale Semiconductor, supplying high-performance radio frequency (RF) and microwave transistors for these applications is only part of the challenge, as the company backs its devices with unparalleled capabilities in characterization, packaging, and applications engineering.
Freescale Semiconductor enjoys a rich heritage in fabricating and selling both discrete and integrated RF semiconductors. Last year, the company introduced its seventh generation of silicon RF laterally diffused metaloxide-semiconductor (LDMOS) in the form of the HV7 process, with the output power and linearity performance through 3.8 GHz needed for WiMAX infrastructure applications. They have also announced a high-voltage version of this process, operating at 48 V, for industrial, scientific, and medical (ISM) applications. Freescale has also extended its high-power GaAs PHEMT device performance to 6 GHz, for WiMAX amplifier applications.
More recently, the company announced the first two-stage radio-frequency integrated circuits (RF ICs) capable of delivering 100 W output power. When driven by Freescale's cost-effective MMG3005N general-purpose amplifier (GPA), the MWE6IC9100N and MW7IC181 00N RF ICs form a complete 100-W power-amplifier solution for wireless base stations operating at 900 and 1800 MHz.
While the performance levels of these discrete and integrated RF power devices are outstanding, putting the devices into the hands of their customers is only the beginning. Each shipped device is supported by the company's "service-in-waiting" personnel with diversified expertise in testing, modeling, packaging, and applications support.
RF Power Characterization
Load-pull (LP) measurement techniques have been increasing in popularity in recent years. RF power amplifiers are generally characterized using such techniques to determine the parametric values of, for example, peak output power, gain and efficiency under various complex load conditions presented at the device reference plane. The use of multiple complex modulated signals in the same measurement environment is also becoming increasingly commonplace. For a manufacturer of high power RF semiconductors, the difficulty in characterizing the products accurately is compounded by the fact that the device development must be carried out on large periphery devices, typically 60 mm, presenting terminal impedances in the sub-0.5-Ω region and with quality factors (Qs) in the range of 8 to 10.
For the past several years, the RF division of Freescale has developed several accuracy-enhancement methodologies and a multitude of automated custom measurement techniques. The division has well equipped high reflection (high gamma) load-pull labs capable of covering frequencies from 250 MHz to 8 GHz and power levels as high as 100 W CW (500 W pulsed) which service the company's GaAs, GaN and LDMOS device, modeling, applications, and other functional groups (Fig. 1). Freescale's systems are capable of performing advanced measurements on devices with impedances of 0.5 Ω and less. To enable such advancements, the company has developed a series of specialized test fixtures with optimum impedance transformation ratios transitioning a 50-Ω system characteristic impedance to the low impedances required for load-pull measurements of high power transistors.
In addition to the fixture-based systems, Freescale also uses on-wafer load-pull systems based on commercial wafer-probe equipment which is used mainly for device research and development, as well as modeling. The onwafer load-pull system features a unique three dimensional anti-vibration mechanism to minimize the effects of tuner vibration, thereby minimizing probe-to-wafer contact damage.
The accuracy of the Freescale Semiconductor load-pull systems typically shows transducer gain differential, ΔGt, of less than 0.25 dB at maximum gamma (0.93 to 0.95 or the edge of the Smith Chart) and less than 0.1 dB inside the measurement region.1 This level of accuracy is in part achieved by the use of precision 7-mm coaxial connectors at all measurement reference planes. These connectors exhibit typical VSWR of less than 1.008:1 at 2 GHz. The center contact resistance of less than 0.1 mΩ and excellent calibration characteristics, with unit-to-unit impedance variation of less than 0.1 percent and phase variation of less than 0.21 deg. at 18 GHz also contribute to excellent measurement accuracy.
A thru-reflect-line (TRL) calibration is used with the vector network analyzer (VNA) in conjunction with the load-pull test system to achieve source match of better than 45 dB.2 In contrast to other VNA calibration approaches, such as the short-openload-thru (SOLT) method, a TRL calibration is not burdened by the parasitic circuit elements (inherent additional capacitances and inductances) of the calibration load standard at high frequencies.
Typically, 5000 to 6000 impedance points are characterized for each tuner to ensure a uniform distribution across the source and load impedance planes. A high density of points is required when evaluating large periphery unmatched devices, which are very sensitive to minimal impedance changes owing to their low terminal impedances. Such a high density may not be required in the assessment of the relatively high impedance production parts containing package matching elements. In this case, a sparse load-pull evaluation may be conducted.
A typical load-pull setup is shown in Fig. 2. Load-pull systems at Freescale are used to evaluate a device's peak pulse power compression, AM-to-AM conversion, AM-to-PM conversion, frequency response and large-signal device input impedance. The systems can also be used for measurements of complex signals to ascertain the average and peak power, adjacent-channel power (ACP), for two-tone and multi-tone testing of intermodulation distortion (IMD), and to assess the device behavior under different loading conditions with EDGE signals. Freescale also conducts complementary-cumulative-density-function (CCDF) analysis of device signal power. The CCDF testing is common to second-generation (2G) and third-generation (3G) wireless measurements. The requirement to perform measurements of CW, pulsed, and modulated signals comes from the fact that these signals exert different thermal loading on the device and, consequently, the optimum load impedance for each modulation format is also different, as shown in Fig. 3.3 In addition to this extensive measurement capability, Freescale has developed valuable data import and post processing tools to enable the user to analyze rapidly the behavior of the device under test (DUT) in two-dimensional or three-dimensional planes (Fig. 4).
A pulsed VNA load-pull technique is used for measuring a wide range of Freescale's power transistors, including the 170-W WCDMA device MRF7S21170H. The loadpull power contours for the device show a pulsed output power at 1-dB compression of better than +53 dBm (200 W) and gain of 19.94 dB at 2.14 GHz (Fig. 5). Armed with this knowledge, the final matching network design of the MRF7S21170H becomes a straightforward exercise: a choice in load and source impedances is made in order to optimize power density, gain, efficiency, and a synthesizable matching network simultaneously.
Modeling Power Devices
The design of RF power amplifiers (PAs) for modern communications and broadcast systems, industrial, scientific, and medical applications, and the avionics and radar markets is a significant challenge. Designers are required to meet goals for improved energy efficiency from the RF PA, while simultaneously meeting stringent regulatory requirements (e.g., for linearity), and demands for lower-cost amplifiers.
Traditional amplifier designs based on the Class AB mode of operation are being supplanted by design approaches to achieve higher efficiency, using architectures such as Doherty and Envelope Tracking, and perhaps by more inherently nonlinear modes of operation, such as amplifiers with Class D, E, F, and other operating modes. These conflicting demands for higher efficiency, higher linearity, and lower cost mean that the designer is faced with making a multidimensional compromise. This is an extremely difficult task to accomplish using empirically based or "cut-and-try" approaches. The designer must turn to computer-aided-design (CAD) techniques and circuit simulation to optimize the design. This increased use of CAD methods for RF power-amplifier design places a greater reliance on the availability of accurate transistor models for simulation. More and more companies are relying on CAD methodologies to reduce time-to-market dramatically, and to increase the robustness of the design in the face of process and manufacturing variations. For semiconductor vendors, the ability to provide accurate, nonlinear, electro-thermal models in a timely fashion has become an important differentiator between alternate suppliers.
Power-amplifier designers choosing from among Freescale's extensive lineup of highpower RF transistors are supported by the comprehensive experience of Freescale's RF Modeling team, and an impressive selection of nonlinear electro-thermal transistor models. The models, which are available online from the company's RF High Power Model Library at www.freescale.com/rf/models support a wide range of CAD software tools. For maximum flexibility, Freescale works with a large number of software suppliers (known as "Model Partners") to ensure the portability of their device models with popular CAD tools. Supported CAD tools include Agilent EEsof's ADS™, Microwave Office™ from Advanced Wave Research (AWR), Analog Design Tools™ from APLAC (now an AWR company), Ansoft Designer™ from Ansoft, and Genesys™ from Eagleware-Elanix tools (now part of Agilent Technologies).
A typical discrete RF transistor with in-package matching networks is illustrated in Fig. 6. The matching networks are included to improve the ease-of-use of the product, and its performance, by transforming the low input and output impedances of the transistor die to more practical levels. These matching networks are constructed with small diameter bonding wires and metal–oxide–semiconductor (MOS) capacitors; in the largest RF/microwave power transistors there are between 100 to 200 bondwires and several MOS capacitors all densely packed into the package cavity. For high-power RF IC products, the matching networks are constructed using on-chip spiral inductors, capacitors, and transmission lines.
The matching networks introduce very high-Q resonances that provide the necessary impedance transformation. Slight changes to the bond-wire arrays can result in frequency shifts of these resonances that may adversely alter the characteristics of the matching network. In many applications, bond-wires are considered to be parasitic elements as they only serve as a means to provide a conductive interconnection between the leads of the package and the semiconductor devices contained within it. However, within RF power transistors they are not parasitic elements, they are an integral part of the design and they must be modeled accurately.
High-power RF and microwave semiconductor transistors are generally enclosed in air-cavity or overmolded plastic packages. These packages protect the internal circuitry from the external environment, and they aid in the removal of heat generated in the active area of the transistor. In addition, these packages also serve as components of the low-loss matching network. Transistors used for wireless infrastructure applications generate some of the largest heat fluxes amongst all semiconductor devices and it is important that the effects of this self-heating are incorporated into the nonlinear transistor model.
The development of nonlinear electro-thermal models for these packaged transistors taxes the most sophisticated measurement and simulation techniques available.4 The number of issues that must be addressed in a successful realization of a model include the electromagnetic (EM) interactions between the elements of the matching networks, and between the bondwire arrays in the package; thermal management; the self-consistent integration of the thermal model with the electrical model of the device; and the construction of the nonlinear model of the transistor itself.
Freescale has adopted a segmented approach to model development,5, 6 in which a packaged transistor is considered as a system that can be broken down into smaller components, as shown in Fig. 7. Each of the components is modeled separately, and then the separate model contributions are integrated into a single model representing the packaged device. This approach eases the computational load and reduces the complexity of the modeling task, and features characterization of inter-component coupling, which is included in the final model for improved accuracy.
At the heart of the packaged transistor model is the nonlinear model of the intrinsic transistor. This is extracted from bias-dependent S-parameter measurements that are made under pulsed conditions to create an isothermal environment. Sophisticated deem-bedding techniques are used to describe and remove the manifold and extrinsic components, enabling the nonlinear model to be extracted. Freescale uses both the Root model and the MET (‘Motorola ElectroThermal') model for the nonlinear model description.6 The thermal component for the MET model is determined from measurements made over a range of die temperatures, and it is coupled self-consistently with the nonlinear electrical model. The MET model is the de facto standard nonlinear model for RF power transistors in the common CAD tools.
The models for the passive components in the in-package matching networks are determined from linear Sparameter measurements and electromagnetic simulations.5 The thermal model for the package and heatsink is found from measurements using high precision infrared (IR) microscopy measurements,7 in which the temperature of the transistor die can be determined while under RF drive.
After a model has been generated, the final process of validating the model begins, by comparing the model predictions against an independent set of measurements that have not been used in the model generation. Freescale has developed a proven method for validating its large-signal models, based on its load-pull measurement capabilities. Essentially, CAD tools are used to simulate the environment seen by a high-power device during load-pull testing. Measured loadpull S-parameter data for a DUT at fundamental and harmonic frequencies are synchronized to present load impedances during a nonlinear harmonic balance simulation. The coordination of measurement and modeling capabilities helps optimize the model to match the measured results.
An example of the model validation is shown in Figs. 8 and 9 for a silicon LDMOS transistor similar to that in Fig. 6.4 This device has been designed for N-CDMA, GSM, and GSMEDGE base-station applications in the 860- and 960-MHz bands, and in a typical GSM application with 28 V supply and a quiescent drain current of 1200 mA, this transistor is capable of delivering 160 W CW power at 1 dB compression. The transistor comprises three active die, with a total gate periphery of about 270 mm. An input matching network is included in the package: this is a T-network using 78 bond-wires and MOS capacitors. Models of the constituent components are derived as described earlier, and a complete model is then constructed from these parts. The validation consists of large-signal one- and two-tone simulations, performed under pulsed conditions to provide an isothermal environment. Load-pull measurements using the same thermal conditions and input and output loads are performed, and compared against the simulations. Measurements and simulations of the output power, third-order intermodulation distortion (IM3), power-added efficiency (PAE), and transducer gain are shown in Figs. 8 and 9. These power sweep or driveup measurements were performed with the packaged transistor tuned at the output for maximum power-added efficiency, and maximum output power, respectively. The measured and simulated results are in good agreement over range of the test conditions.
Packed In Plastic
High-power RF and microwave semiconductor transistors are generally enclosed within over-molded plastic (OMP) or air-cavity packages. Transistors used for high-power RF and microwave applications dissipate substantial amounts of power and consequently operate at high junction temperatures. During the design of a package, stringent thermal-mechanical design practices are followed to ensure that the package can dissipate the substantial heat-flux generated by the transistor while not degrading its electrical performance. In addition, the package must be rugged and have high mechanical strength to operate reliably within, for example, cellular base stations and broadcast systems.
Photographs of typical air-cavity and OMP packages are shown in Fig. 10. The internal components of the transistor within the plastic package are over-molded with a low-loss plastic material. The majority of high power transistor packages generally have two or four leads, although new multi-stage high power RF ICs that incorporate higher functionality have more leads. The packages are designed for the leads to rest on top of the microstrip transmission lines on a printed-circuit board (PCB). The back side of the flange contacts the heatsink of the power amplifier forming a conductive electrical connection to the bottom conductor of the microstrip and a conductive thermal connection to the heatsink, which enables heat to flow away from the packaged transistor.
The air-cavity package is the most expensive component of an assembled power transistor, attributable to the materials used in its construction. Since the power transistor is one of the most expensive components in a RF power amplifier, these air-cavity packages are often a target for cost reduction through design and material developments.
Over the past six years, Freescale has systematically re-engineered aircavity package designs with new materials to improve performance and decrease the package cost. In 2004, Freescale made a change to the heatsink materials in their packages, resulting in 15-to-35-percent improvement in thermal performance. With this improvement, the industry quickly adopted this improved package design.8
Freescale has driven further cost reductions in the package with the innovative development of plastic-packaging techniques for high-power RF transistors. Using an OMP package solution, Freescale is able to offer RF transistors with power levels of over 130 W at 2.1 GHz, with performance rivaling that of a metalceramic air-cavity product. To date, over 30 million RF power transistors have been delivered in over-molded packages. Further, Freescale offers more than a dozen different package outlines and lead configurations in the OMP package technology, enabling a variety of power RF IC products.
These OMP transistors are fully compatible with traditional highpower RF applications. The fundamental design of the package, its materials, and the manufacturing process derive from Freescale's legacy of innovations in high-power automotive and industrial packages designed for the most demanding environmental conditions. These are packages designed to achieve mean time between failures (MTBF) in excess of 1900 years. The OMP housings feature tighter mechanical tolerances with significant tolerance improvements (as much as 50 percent) over traditional air-cavity packages. The tight dimensional tolerances and excellent moisture-sensitivity-level (MSL) rating make these packages ideally suited for automated PCB manufacturing at the amplifier subassembly level.
As with the air-cavity housings, the OMP packages can operate reliably at device junction temperatures above +200°C. An integrated copper heat sink provides excellent thermal resistance and heat dissipation, and the packages support lead-free (RoHS) interconnect processing, with an MSL rating of 3 or better in a +260°C reflow solder process. For those concerned with standards compliance, the OMP packages are registered with JEDEC.
Freescale evaluates the performance of its different package types in a specially configured thermal-analysis lab. The thermal performance of a packaged transistor is a major factor in the system-level cooling required for a complete wireless base transceiver station (BTS). The ability of the package to dissipate heat is determined by its thermal resistance—the temperature difference between two points due to the power that is being dissipated.
To obtain the thermal resistance of a packaged transistor, Freescale has developed a rigorous methodology,9 using an infrared (IR) microscopy is used to measure the temperature on the transistor die while it is operating under realistic termination impedances and signal excitation. With this microscope, the temperature distribution across the die can be viewed as a function of power level, bias level, matching condition, frequency, and even by the selected modulation scheme, such as WCDMA or IS–95. A photograph of an IR image of a transistor dissipating 60 W power is shown in Fig. 11.
Using IR microscopy, the maximum die surface temperature in the measurement field can be located. During the thermal measurements the temperature at the bottom of the packaged transistor can be determined and easily monitored by a thermocouple located directly beneath the transistor's active cell, or heat-generating area, as illustrated in Fig. 12. A hole is drilled in the ceramic lid of the transistor, or the lid is removed to allow an unimpeded view of the surface of the die. For OMP packages, the mold compound can be etched away until the surface of the die is exposed.
Applications Engineering
Freescale's four-tiered support structure for its high-power RF devices includes inventive and experienced applications engineers who assist customers with circuit design and troubleshooting for a wide range of requirements in commercial industrial, medical, avionics, broadcast, and cellular infrastructure applications. This role has become a necessity as the system-level complexity of modern RF power amplifiers has increased while the typical design cycle times have decreased dramatically. Freescale's RF applications team focuses closely on these complexities and how devices are used in customer applications to facilitate a rapid and seamless integration of Freescale's transistors into customer designs.
To facilitate faster customer design cycles, Freescale's applications team has recently begun developing demonstration circuits of optimized RF device line-ups for specific, high-volume applications, such as GSM, CDMA, WCDMA, TD-SCDMA, and WiMAX. These system-level circuits are designed to demonstrate the performance of the entire RF line-up using realistic space constraints, common and economical RF components, and typical assembly procedures such as solder reflow, surface-mount, or device-clamping methods. Additionally, circuit-level efficiency and linearity enhancement techniques such as Doherty-combining and analog predistortion (APD) approaches are incorporated when possible for customers to understand the benefits of different design methods. The goal is to create RF line-up demonstrators for common applications that can be easily incorporated into customer systems.
One example of a line-up demonstration circuit is shown in Fig. 13. This circuit was created to demonstrate a complete 1800-MHz GSM lineup consisting of the MMG3005 GPA driving the MW7IC18100N high-power RF IC. The MMG3005 is a Class A biased InGaP HBT with +30 dBm-rated output power at 1 dB compression. The MW7IC18100N is an HV7 LDMOS two-stage RF IC with 100 W (+50 dBm) rated output power at 1 dB compression. Together, these two devices create a high-performance 1800 MHz GSM line-up with nearly 50 dB of gain, 37 percent line-up efficiency, and 1.5 percent EVM performance at +46 dBm output power, as shown in Figs. 14 and 15. The low-cost plastic packaging, compact circuit layout, and minimal use of RF components make this applications line-up an ideal solution for the cost-sensitive GSM market.
Figure 16 shows the typical performance of an RF line-up demonstration circuit that was created for the emerging TD-SCDMA market in The People's Republic of China. This circuit consists of the MW6IC2215N RF IC driving the MRF6S21100H discrete transistor. The MW6 IC2215N is an HV6 LDMOS two-stage RF IC and the MRF6S21100H is an HV6 LDMOS discrete ceramic transistor. These devices have rated output power at 1 dB compression of 15 W and 100 W, respectively. While neither of these devices was targeted specifically for the TD-SCDMA market, when evaluated together they demonstrate excellent six-carrier TD-SCDMA performance. At +38 dBm output power, the line-up gain is 43 dB, the uncorrected adjacent channel power is –51.4 dBc, and the uncorrected alternate-channel power is –52.3 dBc. These performance levels are achieved with an industry-leading line-up efficiency of nearly 15 percent.
In addition to line-up demonstration circuits, circuit-design assistance, and system troubleshooting, Freescale's applications engineering group works closely with the modeling and measurement teams to fully characterize the large-signal behavior of Freescale's RF power products. With these four groups in RF characterization, modeling, packaging and applications engineering, Freescale is much more than a device supplier— it's a company that offers a comprehensive set of tools to aid in the success of its customers.
REFERENCES
