SoCs no panacea for comms infrastructure

Like "military intelligence," the "communications infrastructure SoC" is almost, but not quite, an oxymoron. Unlike battery-powered handsets, the room-size cellular basestations and telecom switching stations that make up the real communications infrastructure have scant need for highly integrated, small-form-factor, low-power ICs. And with the capital-spending environment being what it is, builders of communications infrastructure systems are not big consumers of SoC devices.

However, this is not to imply that system builders are not interested in the advantages of SoC integration. Both basestations and switching centers are densely packed with electronics. Telecom system builders look at devices that will increase channel density without raising power consumption or heat dissipation inside a closed cabinet.

That is certainly the case in central-office switching stations, where manufacturers are interested in increasing the channel density and, in some cases, overlaying data call services like DSL and analog modems on the existing voice lines. Change here is slow, since it must be made without disruptions to the existing services.

In contrast, cellular basestations must build in "hooks" and accommodations for new services. The migration from 2G to 2.5G and 3G mobile services will demand multicarrier RF transmission with increased Internet and packet data traffic. Existing basestations must be rewired not only to accommodate more channels but also to digitize RF and baseband information that was previously handled as a high-frequency analog signal.

Let us look into several strategies for integrating basestation and central-office functions into new-generation silicon. Because of their interface to a 48V circuit, codecs and voice-processing systems harness a high-voltage fabrication process. In some cases, this requires a bipolar interface; in others, a wide-geometry CMOS (3m and 5m) offers a higher breakdown voltage than deep-submicron CMOS. But, as Dru Cabler and Don Beech of voice codec manufacturer Legerity Inc. note, finer-geometry CMOS enables the channel density of voice-processing circuits to transition from two to four, four to eight, eight to 16 and so forth.

With cellular basestations, manufacturers are exploring ways for bipolar processes--particularly SiGe - to replace GaAs in RF transceivers. GaAs is relatively inexpensive and offers low noise and good power transfer characteristics at gigahertz frequencies. Power MOSFETs, particularly LDMOS transistors, have replaced the GaAs antenna drivers in cellular basestations operating at 2.1GHz and below, but everything else in the transmitter-receiver loops is subject to review. BiCMOS, with SiGe as the bipolar implant, is increasingly being explored as the means of obtaining some of the higher channel densities suggested by CMOS, along with a high-frequency bipolar process.

Brad Brannon and Paul Hendriks of Analog Devices Inc. explain the elements involved in the construction of 2.5G and 3G basestations in their contribution. "Within the receive and transmitter chains, processes such as GaAs are good for low-noise devices," they write, "but these processes are not sufficient for high-density functions such as ADCs. Conversely, CMOS is an excellent process for data converters, but provides poor noise performance." Brannon and Hendriks call for a "smart partitioning" scheme that selects processes for their optimum behavior in each element of the basestation.

Similarly, PMC-Sierra Inc.'s Laurie Wallace describes a technique for linearizing the output of basestation power amplifiers. The technique - which involves applying at the transmitter predistortion that comes out as an entirely linear signal at the receiver - depends on the application of DSP wave-shaping technology. But this does not imply that DSPs could be integrated on the same chip with the power amp. "In cellular basestation systems, discrete devices are the norm," Wallace says. "Power amplifiers rely on discrete components to perform functions such as crest factor reduction to digital predistortion."

About the only place for deep-submicron CMOS - 90nm technology - is in the protocol dissection telcos must perform as they convert the voice-based asynchronous transfer mode protocol packets into Internet Protocol (IP) packets. The Internet functions like a conveyor belt for packets, putting blocks from multiple senders on the same transmission channel and sorting them at the central office. But ATM and IP use different size packets, with different headers and address references, and sorting them out requires high-speed processing.

- Stephan Ohr

EE Times