Enhancement-Mode HEMTs vs. HBTs HBTs have recently established themselves as the technology of choice for the PA function in handsets, largely due to excellent power performance and linearity for CDMA and new W-CDMA applications. And while new players such as Alpha, Anadigics, Raytheon and Celeritek are joining established manufacturers such as RF Micro Devices and Conexant in volume production of these devices, companies such as Motorola and Agilent have recently introduced enhancement-mode (E-mode) PHEMTs products that could challenge the HBT for sales of handset PAs. The Single Supply Voltage Issue The GaAs HBT has effectively replaced the MESFET for the power amplifier functions in wireless handsets. An important reason for its popularity with system designers is the single positive supply voltage. The HBT features a near zero leakage current in the off-state, which has allowed handset manufacturers to eliminate the negative voltage supply and the switch needed to turn off the drain current. This leads to battery power savings and increases the amount of usable board space, and reduces the overall cost of the phone. In contrast, MESFETs and PHEMTs require an additional negative voltage supply to isolate the appreciable leakage current (several mA) that would otherwise place a drain on the battery when the phone is in the standby mode. But with the E-HFETs and E-HEMTs recently introduced by Motorola and Agilent, the drain currents are now low enough (i.e. in the A range) to eliminate the negative supply voltage and the need for a MOS drain switch. And with the playing field now level, both technologies can compete for the PA function. Horses for Courses So could HBTs be in for more competition from E-mode PHEMTs in future? According to Earl Lum, market analyst for CIBC World Markets, E-PHEMT devices will become very important for control functions such as switches and modulators, but the jury is still out regarding the PA function. "In general, we are seeing more modules being employed in handsets, which favors HBTs for the PA," says Lum. "Modules are certainly the direction the industry will be taking in future. If the application is CDMA, the chances are that HBTs will dominate because of the better linearity. However, in the case of GSM, the market will be split between E-PHEMTs and HBTs. For the W-CDMA applications that are now coming on-line, where more stringent demands are placed on linearity for data transfer, HBTs will dominate. E-PHEMTs do however show some advantages in terms of improved efficiency, and traditional FET manufacturers will probably develop these devices in parallel with their HBT efforts." Lum adds that last year saw several companies undergoing a design and product introduction phase, and 2001 will likely see companies consolidating production orders of HBTs, with InGaP replacing AlGaAs as the emitter structure. "This year we will see a critical mass for production of HBTs, with companies such as Celeritek, Anadigics, Raytheon and Alpha ramping up their production," he says. Best of Both Worlds Motorola has recently qualified its E-HFET process, and is also working on a HBT process. The company s new E-HFET features a hetero-insulating gate that allows a high turn-on voltage [CS 6(9), p. 28]. "The insulating gate allows us to apply a bias of up to +1.8 V on the gate," says Motorola s Mark Wilson. "This means we can set the threshold voltage up as high as +0.5 V, which is enough to achieve the necessary off-state leakage characteristic to achieve a truly-off device. Hence, we can eliminate the Si MOS switch required to shut off the drain current in the same way as a HBT device." Comparing both technologies, Wilson believes that the issue of die sizes, which is often quoted as a reason to favor HBTs over PHEMTs, is more complicated than at first sight. "An E-PHEMT will occupy more area than a HBT on the wafer, and if you only take into account the active area of the transistor, HBTs are smaller by a significant amount. However, in an HBT, you still need to provide ballasting to prevent problems with thermal runaway. In addition, passives and bond pads take up roughly the same area in both technologies." Taking this into account, he says, the E-PHEMT ends up being around 2030% larger. "However, the cost of an HBT epiwafer is around 20% higher than for the E-PHEMT, and the processing involves 12 more mask layers, which tends to offset the gains made in die yield," he says. So if there is no big cost differentiation, what about the performance? Wilson says some applications are well suited to E-PHEMTs, such as PAs for GSM applications, which benefit from the excellent PAE performance (>60%). This is because the PA does not need to be backed off to improve the linearity, and can thus be driven hard. The E-mode device is also inherently more robust and reliable. "As you back off the current density in the HBT transistor to give you more reliability, you have to increase the die size. This means you start to lose some of the benefits you might have gained through the die size of an HBT." According to Sanjiv Shah, director of product management, Raytheon is moving ahead aggressively with InGaP HBT technology, in both existing processes and with the development of next generation processes, and the company expects significant product announcements to be made early this year. "We have evaluated E-mode PHEMTs, and concluded that these devices tend towards better linearity and PAE," says Shah. "However, working against this is the die cost per square mm, and most importantly, the leakage current, which is an issue for most manufacturers." Shah says that in terms of performance, HBTs are making improvements all the time and that the technology gap between the two types of devices is now insignificant. "At the end of the day, people have made their bets with the choice of technology, and are moving ahead with these [choices]," he says. "That said, Raytheon is developing an E-PHEMT process, although not for handset applications." Infineon is also forging ahead with its new GaAs HBT process, which is designed for volume production of PAs for Bluetooth applications and handset PA modules. "HBTs currently dominate the handset business because of the smaller die size and the superior linearity, which is important for highly linear PAs and modules used in next generation W-CDMA phones," says Markus Behet, product manager for GaAs projects and wireless projects at Infineon. "Nevertheless, studies have shown that E-PHEMTs can offer higher PAE, and are well suited to saturated-mode PAs for GSM applications. The E-PHEMT technology also exhibits good thermal stability, and is not subjected to reliability issues known for HBTs," he says. RF Micro Devices is the world s leading supplier of GaAs HBTs. In a recent panel discussion (GaAs IC Symposium, Seattle October 2000), a spokesman for the company said that HBTs were the preferred technology as a result of their very low leakage current in the off-state, and because most HEMTs needed a drain switch as a result of an appreciable residual drain current. In addition, in a HBT the control of the turn-on voltage is set by the material bandgap, not by the recess etch of layer thickness. In a HEMT, the turn-on voltage variation can be as much as a 630 mV for a epilayer thickness variation of several monolayers. Agilent recently unveiled its own single supply voltage E-PHEMT device, which it touts as demonstrating superior PAE performance to both current and newer generation InGaP HBTs for CDMA, DCS and GSM applications [see CS 6(8), p. 19]. The company claims its new device should also increase the CDMA phone battery life by up to 15%. "Direct comparisons [between HBTs and E-PHEMTs] are difficult," says Dave Wu of Agilent s Wireless Semiconductor Group. "When you design a PA for a system there are many other design constraints, and you have to meet all the specifications. This includes system and internal circuit specifications, but also the ability to supply product in high volume. That said, we do see that the E-PHEMT device has a raw performance advantage over the HBT, and our initial evaluations for CDMA and W-CDMA PAs have revealed very good linearity from these devices." Alpha also subscribes to the ideal model of having both technologies on board. "In my opinion, it is beneficial to have both HBT and E-PHEMT technologies as part of our available process portfolio," says engineering manager Gene Tkachenko. "While the HBT is preferred in many wireless phones as a result of the single supply operation and smaller die size, E-PHEMT devices have higher PAE, and the performance gap over HBT widens as the supply voltage is reduced from 3.2 to 2.4 V, and below. Both technologies have pluses and minuses that have to be carefully weighed when making a decision as to which technology platform to use for a specific PA design. In the case of HBTs, the device performance is traded-off to achieve a thermally robust design and ensure device stability, reliability and ruggedness. In the E-PHEMT device, the main trade-off is between electrical performance and process reproducibility. "Ultimately, it comes down to providing a solution at the lowest possible cost," Tkachenko concludes. "The majority of today s cellular phone PA solutions are designed to be in the form of a module and the PA chip cost is only a fraction of the overall module cost. Thus, design and price considerations dealing with other parts of the module such as the Si controller chip are becoming increasingly important." A clear Winner? "It is difficult to judge which of these technologies will win the race," concludes Infineon s Markus Behet. "Clearly, right now the HBT is the dominant technology, mainly because of its maturity. But my feeling is that the E-PHEMT device will capture a share of the market once the technology is stable and more companies jump on the train. In future, there will be a market for both technologies, taking advantage of their respective strengths." Mark Wilson says that Motorola s strategy is to have both technologies. "If you are pushing the AlGaAs HBT hard there could be some questions about reliability, especially for base station applications, but it s also important to remember the handset has a relatively short life. However, the newer InGaP HBTs are more reliable and manufacturable than the original AlGaAs HBTs, and that s certainly the approach we have taken with our HBT development. We look forward to evaluating the manufacturing economics of these two technologies over the next year," he says. CIBC s Earl Lum believes that PA modules may also be a deciding factor in future. "This year it will be very important which company comes out with the best PA module. In general, we are seeing that handset makers are favoring modules, and the company that supplies these will have an edge in terms of design wins."