UMTS Multiband Architectures: Handling Complexity in the RF Front End William Mueller Strategic Marketing Manager Avago Technologies Wireless Semiconductor Division 18 April 2008

Agenda Definition of “Front End” Architectural Progressions Comparisons and Tradeoffs Comments on Components Wish List

The “Front End” Power Amplifiers GaAs HBT GaAs FET Si MOSFET Duplexers, Diplexers, Filters acoustic (SAW, BAW, FBAR) dielectric (ceramic, substrate) Switches FET (GaAs, SoS) diode mechanical (MEMS) Other passives, substrate, … Specialized Technologies Connecting the RFIC to the Antenna. Avago Technologies (formerly HP Semiconductor, then Agilent Technologies) is a leading supplier of duplexers and power amplifiers for mobile handsets

Baseline: QB GSM Architecture For 2G, most GSM phones were based on a common simple architecture supporting all worldwide bands 3 components for 4 bands Uses 10 pins on RFIC Experience from GSM sets expectations: “adding a band is easy and inexpensive”. Cost of front end dominated by the PA

Moving to Multi-Band UMTS: QB GSM + 5 Bands UMTS Proliferation of new UMTS bands has created a much more challenging RF front end 24 components for 4 + 5 bands Uses 37 pins on RFIC Adding FDD bands isn’t as easy as adding GSM bands was. PA cost no longer as dominant The result is complex: routing, cost, and size issues

Higher Isolation Removes Filters A move to improved RFIC along with higher isolation duplexers eliminates both Tx and Rx filters from the UMTS chains. ISO Tx band / Rx band originally 50/40, but now need 55/50. Many RFICs adopt differential Rx ports. 14 components for 4 + 5 bands Uses 25 pins on RFIC New technology into duplexers Routing of multiple differential Rx lines causes major headaches Still a lot of components. High Isolation duplexers

Implementing Differential The IC being differential doesn’t mean the duplexer has to be. 4 element network to match to RFIC and provide DC block Is really 100 W 200 W 100 W Any matching at duplexer must retain balance Paired high impedance differential lines can be difficult to route while maintaining magnitude and phase balance, especially if they need to cross into lower board layers. 4 element network to match to RFIC and perform SE:DE transformation (DC block intrinsic with FBAR) Is really 200 W 50 W Single 50 ohm transmission line is straightforward to route even if crossing into lower board layers. This allows more flexibility in duplexer placement. Optional simple, flexible matching at the duplexer Phone performance achieved (Band VIII). 2937CH 3013CH 3088CH -109.5dBm -109.0dBm -109.0dBm

Integration Speeds Design Switch-filter and PA-switch-filter modules Intermediate integration for easier design, better performance. Dual band PAs for size and cost savings PA-duplexer has better efficiency than discretes due to interstage match Multiplexers improve broadband rejection, simplify switching PA switch integration (but are switch requirements standardized enough?) “by band” FEMs SE lines and baluns to solve routing 6 components for 4 + 5 bands Uses 25 pins on RFIC Multiplexers Intermediate integration simplifies design, improves performance. Dual band PAs

Multiband PAs Are Proposed to Cut Cost Move to Multi-Band PAs to eliminate PAs. However this also adds significant loss, from several sources: 1) IL of new switches 2) poorer PAE of PA due to wider loadline 3) poorer PAE to oversizing PA die to handle band with most loss 4) additional switch loss to stop noise leakage from common PA on UMTS bands with overlap 15 components for 4 + 5 bands Uses 22 pins on RFIC Switch-filter structures Multi-Band linear PA Significant impact on efficiency: ~100 mA added peak current (but only a few mA more average current)

Multi-Mode PAs to Go Even Further Move to Multi-Band Multi-Mode PAs to further simplify. Also uses Rx of duplexers for GSM Rx. Further impact on efficiency: 1) even poorer PAE to oversizing PA die to handle higher power of GSM: may broach peak current limit 2) “through” for GSM Tx creates noise leakage path for all bands: more importance on high isolation (hence lossy) switches. However sets stage for multi-mode UMTS-LTE Sensitivity impact from increased switch losses and higher loss of duplexer filters compared to Rx filters, including BW effects. DC-DC converter GSM Rx through duplexers? 15 components for 4 + 5 bands Uses 14 pins on RFIC Multi-Band Multi-Mode PA Costly insertion loss on GSM Tx (esp. low band) Architecture may move rather than remove cost Enough bandwidth for 700 MHz and Band VII?

Multi-band multi-mode PA Noise Leakage Paths B1 Noise Leakage paths GSM High: noise leakage around B1, B2, B3, B4/10, B9 B1: noise leakage around B2 B2: noise leakage path around B3/B9 GSM Low: noise leakage path around B5, B6, B8 B8: noise leakage path around B5/B6 Example: When broadcasting on B2, noise from the common high band PA at 1930-1990 MHz leaks across SW1 (1) and reaches the antenna switch (3) with minimal loss. It leaks across the antenna switch, and into the output of the B2 duplexer, then passes through the duplexer Rx filter (4) to add to the noise at B2 Rx. Similarly, noise in the 1930-1980 MHz band leaks across SW1 (5) and passes through the Tx filter of the B1 duplexer (6) with minimal loss. It then leaks across the antenna switch (7) and back through the B2 duplexer Rx filter (4), adding more noise. (6) B2, PCS Rx SW1 (4) (5) B3, DCS Rx (7) (1) (3) Multi-band multi-mode PA GSM High GSM Low B5, 800 Rx SW2 B8, 900 Rx

Switches and Isolation Switch isolation comes from adding series-shunt pairs of FETs. While a second pair of FETs doubles the isolation, it also doubles the insertion loss. Thus there is an insertion loss penalty for using high isolation switches. “Rule of thumb” typical performance: High band, one pair: I.L. = 0.4 dB typ, ISO = 18 dB typ. Low band, one pair: I.L. = 0.3 dB typ, ISO = 22 dB typ. Need 10 to 15 dB more ISO than duplexer filter supplies for minimal impact on isolation: >60 dB ISO required! (1st series) (2nd series) (1st shunt) (2nd shunt) Representative switches manufacturer I.L. ISO Type (P/N) Test freq I.L. @ high for ISO >60 Filtronics 0.85 max / 0.65 typ 30 min / 32 typ SP4T (FMS2016-001) 2 GHz 1.9 min / 1.3 typ M/A Com 0.6 typ 27 typ SP2T (MASWSS0161V3) 1.8 typ Peregrine 0.75 max / 0.55 typ 23.5 min / 27.5 typ SP4T (PE42641) 2.25 min / 1.65 typ Sony 0.6 max / 0.45 typ 25 min SP5T (a6808865) 1.9 GHz 1.8 max / 1.35 typ

Normalized Cost Comparison GSM +1 Band GSM + 3 Bands GSM + 5 Bands For QB GSM, PA cost dominates. However for added UMTS bands, cost is increasingly shared with duplexers For single band, discretes are cheaper than MB or MBMM – no surprise. For tri-band, there isn’t that much cost difference – but there is a performance difference. For penta-band, there is a cost savings from consolidating PAs – but going MM mostly just transfers cost from the PA to the switch.

Performance Comparison (Current Draw) UMTS B-1 peak and average current GSM peak current, low and high bands Multi-Mode impacts GSM current (>100 mA penalty!) Multi-Band doesn’t impact GSM Multi-Band impacts UMTS peak current (~70 mA) Impact on average current is relatively small (~5 mA) Multi-Mode doesn’t add much additional penalty to UMTS

PA Formats Optimized Efficiency Compatible Power Control Integration: cost, size, performance, time Combined PAs: Dual band Multi-Band Multi-Band Multi-Mode Conventional – optimized for min current at full power (battery peak current) PA only Best average current – optimized for min current at most common power levels (use time) PA plus switch GSM + T/R UMTS + routing combined “std elements” PA with coupler Bypass – optimized for min current at most common, plus has bypass state (use time) PA plus duplexer PA with coupler and detector

Duplexer Formats Design vs Band Properties Differential vs SE Integration: Easy: Broad-Band optimized Gentle roll-off B1, B4, B6, B9 Good for the IC vendor Hard on layout (routing) May cost Q (hence IL) Not truly needed for ISO PA plus duplexer Medium: Out-of-band optimized Moderate Roll-off B5, B7 Multiplexer Hardest: In-band optimized Steep Roll-off B2, B3, B8 Good for layout Requires baluns Lowest IL solutions Front End

Some Closing Thoughts Global solutions cost more (at least use more components) and don’t work as well as optimized local solutions. However they have better economies of scale and make better use of engineering resources. There is a difference between a global reference design that demonstrates the capability of a chipset and a phone that supports a particular geography. The industry is still learning what is needed to handle data while roaming . The answer may depend on whether future handsets are primarily voice appliances or data appliances. There is plenty of room for component suppliers to offer innovative new solutions to solve the emerging problems of multi-band multi-mode handsets.

It sure would be nice if…. 1. the folks who decide such things would be more open to discussing the trade-offs involved prior to adopting new “industry standards” for product format. I’d rather eat fusion cuisine than old chestnuts. 2. an opportunity existed for component houses to discuss possible trade-offs jointly with Service Providers and OEMs / IC houses / reference design creators, to set the right balance between performance and cost. I don’t always like to shop at Wal-Mart. 3. component suppliers had better access to information about IC pin-out so that complimentary parts could be intelligently matched to chipsets. Blind Man’s Bluff was never my favorite game. 4. And where is that small, free, lossless, infinite isolation switch, anyway?