Technical Insight
RFMD sets its sights on the GSM transceiver market
As the market-share leader in cellular power amplifiers, RF Micro Devices continues to implement new technologies and introduce smaller and more highly integrated products. The company is now set to potentially double its dollar content within each handset by moving into the GSM/GPRS transceiver market. Compound Semiconductor spoke with Jerry Neal, RFMD co-founder and executive vice-president of marketing and strategic development.
Compound Semiconductor: There is a trend towards integrating more components into modules, especially for GSM handsets. What is the driving force behind such developments?
Jerry Neal: That s correct - an example is our introduction of the RF3177 transmit module for quad-band GSM/GPRS handsets, which is a laminate-based module that contains all the functionality needed to connect the transceiver with the antenna. The package measures only 9 x 10 x 1.5 mm, but it contains the quad-band GaAs HBT power amplifier (PA), a CMOS power controller, a GaAs PHEMT switch with four receive ports, and all the associated filtering, duplexing and control functions. We re doing the system engineering and optimizing the performance within the module, and the customer ends up with a single placement that takes care of all those functions. If the customer works with a PA die they have to work out the matching and choose the proper switches and filters, so our approach offers a great simplification. All the large makers of handsets are finding that they need to concentrate on the system aspects, for example developing software and incorporating functionality such as MP3 players, Bluetooth or GPS. The systems are becoming so complex that manufacturers need to leave the component work to someone who has expertise in building integrated modules. The ability to put together the front-end components and optimize the cost and performance used to be seen as a differentiator among handset manufacturers, but this is no longer the case. GSM/EDGE is ahead of CDMA in terms of what has to be included in the PA module. However, the general trend is to incorporate all the technology into a single very small package. CS: Some of your new PA products use Lead Frame Module (LFM) technology - can you explain what this is?
JN: Most of the PA modules currently manufactured by RFMD contain a GaAs die and passive components surface-mounted onto a laminate substrate. With LFM, we eliminate the substrate and integrate the passive components into a custom GaAs die, which sits alongside the PA die. LFM technology has a number of advantages. We can manufacture extraordinarily small modules; our new CMDA PA modules (RF3163, RF3164, RF3165) have a footprint of just 3 x 3 mm and are only 0.9 mm thick. Also, the manufacturing cost is very low, since instead of placing multiple passives we have to place a single die. This simplifies the supply chain; we don t have to procure 20 different components, because everything is included in our own GaAs die. We can also incorporate a CMOS power control IC into the LFM package together with the GaAs die, eliminating the need for components such as directional couplers and detector diodes. An example is the RF3146 quad-band PA module for GSM/DCS/PCS applications. CS: What is the significance of pre-production shipments of the Polaris Total Radio* transceiver?
JN: This is a huge event for RFMD; we ve been working on this product for about three years. Frankly, it s late; we took a modulation path that was rather risky. We knew if it worked out it would give us a real advantage in the marketplace, but it was very difficult to do. The fractional N-based modulation architecture is unique, and reduces transmit power consumption by around 30-50% compared with standard translational loop and direct IQ modulators. RFMD has a share of slightly more than 50% in cellular PAs, which is a market worth around $1.2-1.3 billion. By entering the GSM/GPRS transceiver market, which is around the same size, we are effectively doubling our available market in handsets. Competitors such as Skyworks, Silicon Labs and Infineon have been gaining good business from this market, and we feel we will be able to take market share from them because of the technical capabilities of the product. Our transceiver is manufactured using SiGe technology, which offers excellent linearity and integration capabilities. It s a two-chip transceiver that has a high receive sensitivity across all four bands, which is of critical importance to handset manufacturers since receive sensitivity is the primary determinant of dropped calls. CS: Several companies have recently announced SiGe and CMOS PAs. What role does silicon have in the PA arena?
JN: At the moment, we don t see any compelling reason not to use GaAs in the PA. RFMD has been competing for years against silicon LDMOS in GSM applications, particularly from Hitachi, and the competition from CMOS and SiGe is nothing different. One of the companies that just announced CMOS PAs said they had samples ready to ship; well, RFMD ships 1.5 million GaAs HBT-based PAs every day. Our customers expect levels of reliability and ruggedness that in the past has not been possible with silicon. We are convinced that, in the cellular area, silicon will be the dominant technology in the future. Our purchase of a minority stake in Jazz Semiconductor should indicate to people that we believe in silicon. However, looking at ruggedness, the handset PA is a very tough environment, and it s very difficult to get the ruggedness from silicon that you can from GaAs. CS: What about cost and on-chip integration issues?
JN: RFMD now produces GaAs ICs on 6 inch wafers in very high volume, at very low cost, and we don t believe that SiGe or RF CMOS have a significant cost advantage. Also, we have a cycle time in the order of two weeks, and silicon manufacturers can t come close to that because of the complexity of their processes. Supply chain issues are extremely important, and when our customers drop another 10 million pieces on us we can respond very quickly. Silicon can t respond in the same way. RFMD has its own GaAs HBT factories and also makes its own epitaxial material, so we have a tight feedback loop between the production and the starting material, and we can adjust very quickly and optimize the process. Silicon has an advantage in terms of integration, but increasingly customers don t want stand-alone PAs, they want a quad-band module with power control, filtering and switches - the technology does not exist to integrate that onto a single die. For an optimum switch and filter, you have to have external components to the PA. We believe that integration, rather than cost, is the major advantage of silicon. Some day all of the front-end functionality will be integrated on one silicon die - but that day is a long way away. CS: Does your investment in building two GaAs fabs commit you to using GaAs HBT technology in PAs for the foreseeable future?
JN: GaAs fabs are very economical compared with silicon, and our first fab paid for itself in a year and a half. Also, it was necessary to make the investment in order to have the ability to supply our customers. We would not have achieved our current market share if we hadn t built our factories and didn t have the capacity to ramp quickly. We have seen so many examples where people have let technology be the driver, but we believe that technology has to follow the market. There are many other things we can do in our factory, such as manufacturing switches or filters, and we can address infrastructure applications using emerging GaN technology. One branch of our roadmap allows us to migrate out of GaAs HBTs for PAs; if that s the way the market goes, we would not fight it. *Polaris Total Radio is a trademark of RFMD, LLC.
Jerry Neal: That s correct - an example is our introduction of the RF3177 transmit module for quad-band GSM/GPRS handsets, which is a laminate-based module that contains all the functionality needed to connect the transceiver with the antenna. The package measures only 9 x 10 x 1.5 mm, but it contains the quad-band GaAs HBT power amplifier (PA), a CMOS power controller, a GaAs PHEMT switch with four receive ports, and all the associated filtering, duplexing and control functions. We re doing the system engineering and optimizing the performance within the module, and the customer ends up with a single placement that takes care of all those functions. If the customer works with a PA die they have to work out the matching and choose the proper switches and filters, so our approach offers a great simplification. All the large makers of handsets are finding that they need to concentrate on the system aspects, for example developing software and incorporating functionality such as MP3 players, Bluetooth or GPS. The systems are becoming so complex that manufacturers need to leave the component work to someone who has expertise in building integrated modules. The ability to put together the front-end components and optimize the cost and performance used to be seen as a differentiator among handset manufacturers, but this is no longer the case. GSM/EDGE is ahead of CDMA in terms of what has to be included in the PA module. However, the general trend is to incorporate all the technology into a single very small package. CS: Some of your new PA products use Lead Frame Module (LFM) technology - can you explain what this is?
JN: Most of the PA modules currently manufactured by RFMD contain a GaAs die and passive components surface-mounted onto a laminate substrate. With LFM, we eliminate the substrate and integrate the passive components into a custom GaAs die, which sits alongside the PA die. LFM technology has a number of advantages. We can manufacture extraordinarily small modules; our new CMDA PA modules (RF3163, RF3164, RF3165) have a footprint of just 3 x 3 mm and are only 0.9 mm thick. Also, the manufacturing cost is very low, since instead of placing multiple passives we have to place a single die. This simplifies the supply chain; we don t have to procure 20 different components, because everything is included in our own GaAs die. We can also incorporate a CMOS power control IC into the LFM package together with the GaAs die, eliminating the need for components such as directional couplers and detector diodes. An example is the RF3146 quad-band PA module for GSM/DCS/PCS applications. CS: What is the significance of pre-production shipments of the Polaris Total Radio* transceiver?
JN: This is a huge event for RFMD; we ve been working on this product for about three years. Frankly, it s late; we took a modulation path that was rather risky. We knew if it worked out it would give us a real advantage in the marketplace, but it was very difficult to do. The fractional N-based modulation architecture is unique, and reduces transmit power consumption by around 30-50% compared with standard translational loop and direct IQ modulators. RFMD has a share of slightly more than 50% in cellular PAs, which is a market worth around $1.2-1.3 billion. By entering the GSM/GPRS transceiver market, which is around the same size, we are effectively doubling our available market in handsets. Competitors such as Skyworks, Silicon Labs and Infineon have been gaining good business from this market, and we feel we will be able to take market share from them because of the technical capabilities of the product. Our transceiver is manufactured using SiGe technology, which offers excellent linearity and integration capabilities. It s a two-chip transceiver that has a high receive sensitivity across all four bands, which is of critical importance to handset manufacturers since receive sensitivity is the primary determinant of dropped calls. CS: Several companies have recently announced SiGe and CMOS PAs. What role does silicon have in the PA arena?
JN: At the moment, we don t see any compelling reason not to use GaAs in the PA. RFMD has been competing for years against silicon LDMOS in GSM applications, particularly from Hitachi, and the competition from CMOS and SiGe is nothing different. One of the companies that just announced CMOS PAs said they had samples ready to ship; well, RFMD ships 1.5 million GaAs HBT-based PAs every day. Our customers expect levels of reliability and ruggedness that in the past has not been possible with silicon. We are convinced that, in the cellular area, silicon will be the dominant technology in the future. Our purchase of a minority stake in Jazz Semiconductor should indicate to people that we believe in silicon. However, looking at ruggedness, the handset PA is a very tough environment, and it s very difficult to get the ruggedness from silicon that you can from GaAs. CS: What about cost and on-chip integration issues?
JN: RFMD now produces GaAs ICs on 6 inch wafers in very high volume, at very low cost, and we don t believe that SiGe or RF CMOS have a significant cost advantage. Also, we have a cycle time in the order of two weeks, and silicon manufacturers can t come close to that because of the complexity of their processes. Supply chain issues are extremely important, and when our customers drop another 10 million pieces on us we can respond very quickly. Silicon can t respond in the same way. RFMD has its own GaAs HBT factories and also makes its own epitaxial material, so we have a tight feedback loop between the production and the starting material, and we can adjust very quickly and optimize the process. Silicon has an advantage in terms of integration, but increasingly customers don t want stand-alone PAs, they want a quad-band module with power control, filtering and switches - the technology does not exist to integrate that onto a single die. For an optimum switch and filter, you have to have external components to the PA. We believe that integration, rather than cost, is the major advantage of silicon. Some day all of the front-end functionality will be integrated on one silicon die - but that day is a long way away. CS: Does your investment in building two GaAs fabs commit you to using GaAs HBT technology in PAs for the foreseeable future?
JN: GaAs fabs are very economical compared with silicon, and our first fab paid for itself in a year and a half. Also, it was necessary to make the investment in order to have the ability to supply our customers. We would not have achieved our current market share if we hadn t built our factories and didn t have the capacity to ramp quickly. We have seen so many examples where people have let technology be the driver, but we believe that technology has to follow the market. There are many other things we can do in our factory, such as manufacturing switches or filters, and we can address infrastructure applications using emerging GaN technology. One branch of our roadmap allows us to migrate out of GaAs HBTs for PAs; if that s the way the market goes, we would not fight it. *Polaris Total Radio is a trademark of RFMD, LLC.
