Technical Insight
Next-generation LDMOS pushes limits of silicon base-station PAs
Silicon LDMOS will continue to dominate the base-station power-amplifier market until an alternative technology can deliver higher performance at a lower price, writes Hamish Johnston.
The cellular base-station power-transistor market is dominated by silicon, with about 90% of devices being LDMOS. While on paper LDMOS may not be the best technology for W-CDMA power amplifiers (PAs), generational improvements in the technology are ensuring that it will dominate in the foreseeable future. This is despite the theoretical benefits offered by wide-bandgap materials such as GaN.
A shining example of LDMOS evolution is the fifth-generation LDMOS power transistor recently unveiled by Philips Semiconductor, headquartered in the Netherlands. Based on a 0.4 µm process, Philips claims that these transistors offer 30% efficiency for 100 W W-CDMA applications and 17 dB gain for 2.5G PCS/DCS systems.
Mark Murphy, who is the program manager for RF power base stations at Philips, told Compound Semiconductor that the 30% efficiency rating for W-CDMA is a "world record". He also says that increasing the gain to 17 dB from the industry-standard 14-15 dB was an important achievement. Higher gain allows PA designers to reduce the number of amplifier stages by pushing more of the amplification into the final stages of the PA. This reduces the overall cost of implementation.
According to Murphy, Philips has recently moved its process to a "leading-edge CMOS fab that can go down to 0.14 µm". He says that this has allowed the company to achieve significant improvements in the design and operation of such amplifiers. The new fab will also facilitate the launch of the next (sixth) generation of transistors, which will be based on a 0.4 or 0.35 µm process. Murphy says that 0.14 µm technology will probably be exploited in generation seven.
Philips will begin sampling generation-five products in the final quarter of 2004. "We already have prototypes in the lab, and a few weeks ago at the MTT-S [IEEE International Microwave Symposium/Exhibition in Fort Worth, Texas] we held a live demonstration of the [prototype s] 30% efficiency and 17 dB gain." Philips is currently talking to several lead customers and is fine tuning the technology. Murphy expects that lead customers will implement the transistors by mid-2005 as they ramp up product manufacture.
Metallization changeA key feature of generation-five technology is the use of a patented four-level level aluminum/copper metallization system (figure 1). Murphy says that this gives the components a very long mean time to failure (MTTF). This is a key parameter because reliability is one of the most important features of the devices in PAs. Philips used gold metallization in its previous generations of LDMOS.
Aluminum/copper metallization allows for junction temperatures of 185 ºC - instead of the standard 160 ºC using gold - with the same MTTF. If the transistor is used at 160 ºC, it is four times as reliable as conventional LDMOS in W-CDMA operation.
Also on the temperature front, Philips has reduced the thermal resistance of its devices from 0.76 to 0.5 K/W. According to Murphy, Philips set 0.5 K/W as its thermal-resistance target for generation-five devices, and decreasing the resistance further is a key focus for the company. "There are a lot of design tricks and knowledge of the technology that we have to achieve this. This involves getting the overall design right: the thickness of the silicon and the type of heat sink used. These are all areas of current investigation."
Several of Philips s competitors have had greater success in reducing thermal resistance, with Agere Systems, US, claiming a value of 0.35 K/W in its 180 W device. However, Murphy notes that while competitors may quote better thermal-resistance values, he believes that "for a 100 W W-CDMA device at an average output power of 26 W, the fifth-generation device with a thermal resistance of 0.5 K/W combined with the record efficiency of 30% at a record gain of 17 dB gives the industry s leading proven performance".
According to a recent market survey by ABI Research - semiconductor-industry analysts - Philips is currently the number-two supplier of RF power transistors used in base-station PAs. Murphy believes that the company has about a 25% market share. The business was number four in the market in 2000, controlling a 10% share, so Philips s commitment to LDMOS technology appears to be paying off. The market has long been dominated by US giant Motorola (now Freescale), but relative newcomers such as Agere Systems are also making significant inroads.
Other companies producing LDMOS PAs include Germany s Infineon and French/Italian group STMicroelectronics, which ABI says are joint third in the market. Infineon has just announced the latest in its line of GOLDMOS (LDMOS) PAs. The 100 W, 2.1 GHz devices offer 16.5 dB gain, 30% efficiency and a thermal resistance of 0.38 K/W (figure 2).
These firms must retain the loyalty of people like Peter Kenington, who is director of advanced technology at US-based PA module maker Andrew and therefore a purchaser of power transistors. "LDMOS is by far the most dominant technology at present, with about 80-90% of the infrastructure market," he said, adding that the rest of the market is occupied by GaAs FETs, with some legacy designs using bipolar transistors.
"LDMOS is currently winning because of price," said Kenington. He explains that the industry is under financial pressure and any technological advantages offered by an alternative to LDMOS must be sufficient to offset the added expense of the new technology. In particular, these advantages must significantly reduce the overall cost of the PA unit.
It is cheaper because it is a mature technology that is used in volume production, but as Kenington pointed out, "LDMOS is not standing still...every [new] generation of LDMOS is a small improvement on the previous one."
The technology benefits from ongoing increases in peak and average drain voltages. For example, a device might operate at a mean voltage of 28 V but experience regular peaks of up to 40 V. W-CDMA PAs operate in a high peak-to-mean mode and future LDMOS devices will be able to accommodate the peaks by momentarily allowing a boost in drain voltage.
Steve Jones, an RF systems engineer at Agere Systems, told Compound Semiconductor that Agere is pushing LDMOS voltage by increasing the device s channel length. However, he concedes that "designing for higher breakdown voltages is a compromise". This process reduces the gain of the transistor, and operation at higher voltages can degrade the long-term stability of the device. Unfortunately these two factors are key features for PA designers.
Reliability challenge for GaNWhile LDMOS producers like Philips and Agere believe that LDMOS will continue to dominate the market, they are also pursuing alternative technologies. Jones says that Agere is looking at GaAs MOSFETs while Philips has a research group in Nijmegen, the Netherlands, that is focused on GaN. "Philips sees GaN as a potential technology in the future and we are tracking it closely," said Murphy. "We believe that the biggest challenge with GaN is reliability. You can look back several years to all the hype about SiC - there were even some prototypes on the market - but SiC never reached its potential."
Kenington believes that GaN technology and, to a lesser extent, GaAs PHEMT present the biggest threat to LDMOS. "GaN is becoming a more mature technology but there are no commercial products available at the moment," he explained. "There are test samples and pre-production units, but what is being produced at the moment is not yet suitable for commercial base stations."
Kenington believes that current GaN prototypes could be used in the earlier stages of a PA, but not in the output stage of high-power modules. "They cannot yet compete with LDMOS in terms of absolute power level," he explained. However, Kenington points out that the power capability of each successive generation is increasing: "GaN is improving in leaps and bounds, whereas improvements in LDMOS are much more incremental."
Kenington says that Andrew has a close relationship with several companies developing GaN, and is monitoring their progress. There is currently a lot of claim and counter-claim regarding linearity - while GaN makers say that the linearity of their devices is better than LDMOS, the latter is continually improving. "None of the GaN devices that we have seen displays a dramatic improvement over LDMOS," said Kenington, who believes that LDMOS could ultimately be a better choice in certain situations. "It is going to come down to a matter of cost," he stressed.
GaN device makers, including Japan-based Eudyna Devices and US firm Nitronex, say that they will be shipping commercial products by early 2005 (see Compound Semiconductor July 2004). "In some cases I think this is a reasonable time frame," said Kenington. He adds that in other cases this might be a little optimistic. "However, I believe that they will all get there. From a purely technological perspective we could see GaN in commercial systems by the end of 2005 and these could be in the field by the middle of 2006."
Nitronex is sampling two GaN devices at 10 and 20 W that offer a gain of 11.5 dB and 30% efficiency. The company plans to sample a 36 W device later this year (see table).
Availability alone will not induce PA makers like Andrew to replace LDMOS with a different technology. Kenington says that the benefits of GaN are not absolutely revolutionary and therefore the industry will not be prepared to pay a premium for the new technology. In fact, GaN producers may initially have to sell their transistors at a loss to make them competitive with LDMOS.
Beyond the issue of price, Kenington comments that some of the advertised benefits of GaN have yet to be proved. Theoretically, GaN should be more efficient than LDMOS. Its wide bandgap means that devices can be operated at higher voltages than LDMOS, which should make it easier to match the output of the transistors.
Tower-mounted PAsAnother key benefit of GaN is that the devices can be run at higher temperatures than LDMOS. While Kenington concedes that this would give GaN the edge in tower-mounted-amplifier configurations, he observes that "virtually nothing is tower-mounted at the moment because of reliability concerns". He explains that the cost of sending out a rigger to replace a PA at the top of a mast negates any savings gained in the reduced purchase price of the unit. "Network operators are currently reluctant to have anything active on masts and this is only changing slowly. It will be a while before we see tower-mounted PAs and it is not on account of the device technology but rather the complexity of the PAs."
GaN s superior thermal properties could also allow PA modules to be made smaller, thus reducing cooling-related operating expenses. In addition, a single transistor could be used where two are used today. This could be an important improvement because matching one device is easier (and possibly cheaper) than matching two devices. Kenington emphasizes that the important question is whether a single device implementation is cheaper on a dollar per watt basis.
Looking further into the future, GaN and GaAs PHEMT transistors could be better suited to a new wave of PA architectures that employ efficient but nonlinear devices. Such designs have been around for many years and use clever techniques to achieve linear amplification using nonlinear component signals.
Kenington says that these systems could be used for W-CDMA - provided that a few existing problems can be solved. While LDMOS makers concede that their current PAs may not be suitable for these advanced architectures, Murphy - program manager at Philips - is confident that future generations of LDMOS could be made to fit the bill.
Simon Tonks, principal consultant in the Wireless Technology Practice of the UK s PA Consulting Group, has performed laboratory tests on the performance of LDMOS and GaAs transistors. Tonks concluded that LDMOS is best when used in a particular application with stabilized conditions and a limited power range. While this is fine for GSM, he points out that the output power of a W-CDMA base station varies according to the traffic loading in a specific cell. This means that an LDMOS PA may not always operate in its "sweet spot" and its performance would suffer. GaAs PAs offer better performance over a range of operating conditions and therefore could be more suitable for W-CDMA.
Despite its shortcomings for W-CDMA, a combination of inertia and incremental improvements will maintain LDMOS s market lead for the next few years. But in the future the cost-versus-performance advantage of this legacy technology will be eroded by a new generation of transistors. When this will occur, and the exact nature of the usurping technology, remains unclear.
A shining example of LDMOS evolution is the fifth-generation LDMOS power transistor recently unveiled by Philips Semiconductor, headquartered in the Netherlands. Based on a 0.4 µm process, Philips claims that these transistors offer 30% efficiency for 100 W W-CDMA applications and 17 dB gain for 2.5G PCS/DCS systems.
Mark Murphy, who is the program manager for RF power base stations at Philips, told Compound Semiconductor that the 30% efficiency rating for W-CDMA is a "world record". He also says that increasing the gain to 17 dB from the industry-standard 14-15 dB was an important achievement. Higher gain allows PA designers to reduce the number of amplifier stages by pushing more of the amplification into the final stages of the PA. This reduces the overall cost of implementation.
According to Murphy, Philips has recently moved its process to a "leading-edge CMOS fab that can go down to 0.14 µm". He says that this has allowed the company to achieve significant improvements in the design and operation of such amplifiers. The new fab will also facilitate the launch of the next (sixth) generation of transistors, which will be based on a 0.4 or 0.35 µm process. Murphy says that 0.14 µm technology will probably be exploited in generation seven.
Philips will begin sampling generation-five products in the final quarter of 2004. "We already have prototypes in the lab, and a few weeks ago at the MTT-S [IEEE International Microwave Symposium/Exhibition in Fort Worth, Texas] we held a live demonstration of the [prototype s] 30% efficiency and 17 dB gain." Philips is currently talking to several lead customers and is fine tuning the technology. Murphy expects that lead customers will implement the transistors by mid-2005 as they ramp up product manufacture.
Metallization changeA key feature of generation-five technology is the use of a patented four-level level aluminum/copper metallization system (figure 1). Murphy says that this gives the components a very long mean time to failure (MTTF). This is a key parameter because reliability is one of the most important features of the devices in PAs. Philips used gold metallization in its previous generations of LDMOS.
Aluminum/copper metallization allows for junction temperatures of 185 ºC - instead of the standard 160 ºC using gold - with the same MTTF. If the transistor is used at 160 ºC, it is four times as reliable as conventional LDMOS in W-CDMA operation.
Also on the temperature front, Philips has reduced the thermal resistance of its devices from 0.76 to 0.5 K/W. According to Murphy, Philips set 0.5 K/W as its thermal-resistance target for generation-five devices, and decreasing the resistance further is a key focus for the company. "There are a lot of design tricks and knowledge of the technology that we have to achieve this. This involves getting the overall design right: the thickness of the silicon and the type of heat sink used. These are all areas of current investigation."
Several of Philips s competitors have had greater success in reducing thermal resistance, with Agere Systems, US, claiming a value of 0.35 K/W in its 180 W device. However, Murphy notes that while competitors may quote better thermal-resistance values, he believes that "for a 100 W W-CDMA device at an average output power of 26 W, the fifth-generation device with a thermal resistance of 0.5 K/W combined with the record efficiency of 30% at a record gain of 17 dB gives the industry s leading proven performance".
According to a recent market survey by ABI Research - semiconductor-industry analysts - Philips is currently the number-two supplier of RF power transistors used in base-station PAs. Murphy believes that the company has about a 25% market share. The business was number four in the market in 2000, controlling a 10% share, so Philips s commitment to LDMOS technology appears to be paying off. The market has long been dominated by US giant Motorola (now Freescale), but relative newcomers such as Agere Systems are also making significant inroads.
Other companies producing LDMOS PAs include Germany s Infineon and French/Italian group STMicroelectronics, which ABI says are joint third in the market. Infineon has just announced the latest in its line of GOLDMOS (LDMOS) PAs. The 100 W, 2.1 GHz devices offer 16.5 dB gain, 30% efficiency and a thermal resistance of 0.38 K/W (figure 2).
These firms must retain the loyalty of people like Peter Kenington, who is director of advanced technology at US-based PA module maker Andrew and therefore a purchaser of power transistors. "LDMOS is by far the most dominant technology at present, with about 80-90% of the infrastructure market," he said, adding that the rest of the market is occupied by GaAs FETs, with some legacy designs using bipolar transistors.
"LDMOS is currently winning because of price," said Kenington. He explains that the industry is under financial pressure and any technological advantages offered by an alternative to LDMOS must be sufficient to offset the added expense of the new technology. In particular, these advantages must significantly reduce the overall cost of the PA unit.
It is cheaper because it is a mature technology that is used in volume production, but as Kenington pointed out, "LDMOS is not standing still...every [new] generation of LDMOS is a small improvement on the previous one."
The technology benefits from ongoing increases in peak and average drain voltages. For example, a device might operate at a mean voltage of 28 V but experience regular peaks of up to 40 V. W-CDMA PAs operate in a high peak-to-mean mode and future LDMOS devices will be able to accommodate the peaks by momentarily allowing a boost in drain voltage.
Steve Jones, an RF systems engineer at Agere Systems, told Compound Semiconductor that Agere is pushing LDMOS voltage by increasing the device s channel length. However, he concedes that "designing for higher breakdown voltages is a compromise". This process reduces the gain of the transistor, and operation at higher voltages can degrade the long-term stability of the device. Unfortunately these two factors are key features for PA designers.
Reliability challenge for GaNWhile LDMOS producers like Philips and Agere believe that LDMOS will continue to dominate the market, they are also pursuing alternative technologies. Jones says that Agere is looking at GaAs MOSFETs while Philips has a research group in Nijmegen, the Netherlands, that is focused on GaN. "Philips sees GaN as a potential technology in the future and we are tracking it closely," said Murphy. "We believe that the biggest challenge with GaN is reliability. You can look back several years to all the hype about SiC - there were even some prototypes on the market - but SiC never reached its potential."
Kenington believes that GaN technology and, to a lesser extent, GaAs PHEMT present the biggest threat to LDMOS. "GaN is becoming a more mature technology but there are no commercial products available at the moment," he explained. "There are test samples and pre-production units, but what is being produced at the moment is not yet suitable for commercial base stations."
Kenington believes that current GaN prototypes could be used in the earlier stages of a PA, but not in the output stage of high-power modules. "They cannot yet compete with LDMOS in terms of absolute power level," he explained. However, Kenington points out that the power capability of each successive generation is increasing: "GaN is improving in leaps and bounds, whereas improvements in LDMOS are much more incremental."
Kenington says that Andrew has a close relationship with several companies developing GaN, and is monitoring their progress. There is currently a lot of claim and counter-claim regarding linearity - while GaN makers say that the linearity of their devices is better than LDMOS, the latter is continually improving. "None of the GaN devices that we have seen displays a dramatic improvement over LDMOS," said Kenington, who believes that LDMOS could ultimately be a better choice in certain situations. "It is going to come down to a matter of cost," he stressed.
GaN device makers, including Japan-based Eudyna Devices and US firm Nitronex, say that they will be shipping commercial products by early 2005 (see Compound Semiconductor July 2004). "In some cases I think this is a reasonable time frame," said Kenington. He adds that in other cases this might be a little optimistic. "However, I believe that they will all get there. From a purely technological perspective we could see GaN in commercial systems by the end of 2005 and these could be in the field by the middle of 2006."
Nitronex is sampling two GaN devices at 10 and 20 W that offer a gain of 11.5 dB and 30% efficiency. The company plans to sample a 36 W device later this year (see table).
Availability alone will not induce PA makers like Andrew to replace LDMOS with a different technology. Kenington says that the benefits of GaN are not absolutely revolutionary and therefore the industry will not be prepared to pay a premium for the new technology. In fact, GaN producers may initially have to sell their transistors at a loss to make them competitive with LDMOS.
Beyond the issue of price, Kenington comments that some of the advertised benefits of GaN have yet to be proved. Theoretically, GaN should be more efficient than LDMOS. Its wide bandgap means that devices can be operated at higher voltages than LDMOS, which should make it easier to match the output of the transistors.
Tower-mounted PAsAnother key benefit of GaN is that the devices can be run at higher temperatures than LDMOS. While Kenington concedes that this would give GaN the edge in tower-mounted-amplifier configurations, he observes that "virtually nothing is tower-mounted at the moment because of reliability concerns". He explains that the cost of sending out a rigger to replace a PA at the top of a mast negates any savings gained in the reduced purchase price of the unit. "Network operators are currently reluctant to have anything active on masts and this is only changing slowly. It will be a while before we see tower-mounted PAs and it is not on account of the device technology but rather the complexity of the PAs."
GaN s superior thermal properties could also allow PA modules to be made smaller, thus reducing cooling-related operating expenses. In addition, a single transistor could be used where two are used today. This could be an important improvement because matching one device is easier (and possibly cheaper) than matching two devices. Kenington emphasizes that the important question is whether a single device implementation is cheaper on a dollar per watt basis.
Looking further into the future, GaN and GaAs PHEMT transistors could be better suited to a new wave of PA architectures that employ efficient but nonlinear devices. Such designs have been around for many years and use clever techniques to achieve linear amplification using nonlinear component signals.
Kenington says that these systems could be used for W-CDMA - provided that a few existing problems can be solved. While LDMOS makers concede that their current PAs may not be suitable for these advanced architectures, Murphy - program manager at Philips - is confident that future generations of LDMOS could be made to fit the bill.
Simon Tonks, principal consultant in the Wireless Technology Practice of the UK s PA Consulting Group, has performed laboratory tests on the performance of LDMOS and GaAs transistors. Tonks concluded that LDMOS is best when used in a particular application with stabilized conditions and a limited power range. While this is fine for GSM, he points out that the output power of a W-CDMA base station varies according to the traffic loading in a specific cell. This means that an LDMOS PA may not always operate in its "sweet spot" and its performance would suffer. GaAs PAs offer better performance over a range of operating conditions and therefore could be more suitable for W-CDMA.
Despite its shortcomings for W-CDMA, a combination of inertia and incremental improvements will maintain LDMOS s market lead for the next few years. But in the future the cost-versus-performance advantage of this legacy technology will be eroded by a new generation of transistors. When this will occur, and the exact nature of the usurping technology, remains unclear.
