TI also said it is now working with 90-nanometer test silicon that combines the DRP with the baseband processor and some of the power-management functions. Bill Krenik, advanced architectures manager at TI's wireless terminals business unit in Dallas, said the GSM/GPRS implementation cuts board space and power roughly in half.

Moreover, it requires no additional masks beyond the set used for TI's base 90-nm logic process. Only the high-voltage battery-charging function, the power amplifier and flash memory remain as the major functions off-chip.

"We now have a bunch of GSM/GPRS test silicon and the initial looks show very promising results," Krenik said. "Rather than seeing any new limitations, we are more confident where we can go with the DRP approach."

After hitting the GSM market with samples toward year's end and starting commercial shipments in late 2005, TI plans DRP implementations of an integrated wireless LAN chip, and an Edge chip for high-bandwidth cell phones. In 2006, TI expects to attack the CDMA market with an integrated solution.

CDMA is a full-duplex architecture that requires better noise isolation and "stricter" performance from low-noise amplifiers and voltage-controlled oscillators (VCOs), he said. On the plus side, CDMA is amenable to a digital RF solution because "the wider bandwidth relaxes other aspects" of the chip design.

"We have enough real technical data from our test chip evaluation that we see no issue with going into the CDMA domain," Krenik said.

Rather than discuss the cell phone design, the ISSCC papers describe the transmit and receive functions in TI's single-chip Bluetooth implementation, made in 130-nm CMOS. The papers do include descriptions of technologies being used in the cell phone DRP, such as TI's digitally controlled oscillator (DCO), Krenik said, a key part of the cell phone DRP. It replaces the VCO, one of the most difficult-to-design parts in a conventional RF device.

Since a conventional VCO could not be built in a digital process, TI created the digitally controlled oscillator. The analog function is accomplished with a digitally controlled bank of switching varactors, with what TI calls lower "noise feedthrough" than that of today's discrete varactors. Varactors — shorthand for variable reactors — require a smooth, monotonic change of capacitance with voltage, and doing high-quality conventional varactors in a 90-nm digital CMOS process would mean extra masks, Krenik said.

The ISSCC paper discussing the transmitter describes a varactor bank being switched and dithered to control the DCO frequency. But when a voltage is applied, gated-capacitor varactors tend to change abruptly in capacitance, accompanied by transfer characteristics that are not monotonic, causing stability problems.

Krenik said TI engineers realized that a solution lay in the architecture of gated-capacitor varactors, which have both high- and low-capacitance regions. The change in capacitance with a high voltage is almost flat, nearly zero. But capacitance increases and decreases at certain places across the transfer curve.

The design team worked with those shifts, Krenik said, to create a varactor with the required noise-transmission characteristics: The amount of noise coupled to the control node of the varactor is low and relatively immune to ground noise and power-supply noise.

By switching quickly between the low-capacitance and high-capacitance points, TI achieved an average value of capacitance at the tank circuit within the VCO that proved to be noise-immune, Krenik said.

Will Strauss, an analyst at Forward Concepts (Tempe, Ariz.) said he designed board-level radios early in his career. "Switched-capacitor filters have been around for a long time, but they haven't been used in modern designs to my knowledge," Strauss said. "They were very prevalent in low-frequency designs, of 100 KHz or lower, but it was believed that they had some upper-frequency bounds. They are not easy to do in a 90-nm process, I can tell you that, but TI seems to have figured out a way to do it."

Krenik said the result is a much simpler transmitter design, one that eliminates many of the analog, voltage control, phase-detector and loop-filter functions and reduces the external discretes and passives on the board.

"The only real analog in the entire transmitter is the digitally controlled oscillator itself," Krenik said. "The phase compare is digital. The loop filter is digital. We insert the modulation digitally and control the VCO through a digital control function. We have just a handful of dc oscillators and a very small buffer amp to amplify the signal and drive the pin, which goes over to the PA [power amplifier]. This is a case where we took a very complicated analog function and made it digital."

One significant benefit is that while a discrete RF chip draws about 80 to 90 milliamps, the RDP portion of the integrated chip requires only about half that amount. The RDP occupies about 15 percent of the total die area.

The business implications of an integrated baseband with RF solution are being hotly debated now.

Morry Marshall, an analyst with Semico Research (Phoenix), said, "The rest of the industry is skeptical. One drawback is that if the feature set changes, it is awkward to make changes when you have a single-chip solution."

Strauss at Forward Concepts and Marshall at Semico agreed that a large market will continue to exist for low-cost GSM handsets that mainly handle voice calls and do not include such functions as a camera.

By freeing up board space and reducing total power, Krenik argued, TI's single-chip solution lets phone manufacturers add functionality as needed. Also, as process technology advances to the 65-nm node and beyond, the digital RF processor architecture will support further integration and cost reductions.

"The DRP will scale more easily than an analog radio," he said, because it reduces susceptibility to noise coupling and supports a cleaner board for smaller and more-ergonomic phone designs.

In the cell phone industry, where "cost is king," Krenik said, cutting cost is the biggest advantage of an integrated design.

"If TI can do it at lower costs, that's great. But I am skeptical," Marshall said. "It may work for a straight voice phone, but not for the feature-rich phones that the market will be demanding."