Fujitsu transceiver chip eliminates SAWs from 3G handset designs

Obviously cost is a major competitive issue for 3G handset silicon. Less obviously, the real issue is total assembled cost, not just the cost of the chips. With die size and margins driven as low as possible, silicon vendors to the handset business are now turning their attention to another major cost in the big picture: the cost of purchasing, testing, and inserting the external components that support the big chips. This is leading to radical changes in analog circuit designs as the RF gurus try to find ways to eliminate what used to be accepted as necessary external filters and to reorganize the signal paths.

One of the best-equipped design teams to do this is the former Motorola RF design group in Arizona. Despite being passed from Motorola Semiconductor to Freescale, and now sold to Fujitsu Microelectronics America, the team has held together its core players and its expertise. It is large and accomplished. And with the introduction of the MB86L01A 3G/2G transceiver chip this morning, it has achieved a major milestone: elimination of all the inter-stage surface acoustic wave (SAW) filters from the 3G transmit and receive paths, and integration of all the low-noise amplifiers (LNAs.) The Fujitsu part appears to be the first multi-band transceiver to do 3G without the SAW parts, though it does still employ SAWs on the 2G receive bands.

This eliminates quite a collection of external parts: up to eight SAW filters with all of their matching circuitry. In addition, eliminating the SAWs on the 3G receive paths makes it possible for the designers to move the LNAs onto the chip and to employ multi-mode power amplifiers (PAs) eliminating even more packages.

Remarkably, the design team accomplished all this with a standard TSMC 90 nm process, without special process modules, according to senior director of RF engineering Vivek Bhan. All of the advances, he said, were accomplished with architectural changes and circuit-design improvements, not process tweaks.

At the block diagram level the transceiver looks pretty much like any other multi-band part. There is an industry-standard DigRF 3G digital interface to the baseband chip, a multi-band PLL/VCO block, a transmit modulator, a 7-channel receiver block, 2 GSM and 4 WCDMA transmit drivers, and a power controller. Hidden between the blocks in the public block diagram are far more interesting things.

For one, Bhan said that there is very significant digital signal processing horsepower in the transceiver chip between the DigRF interface and the DAC. For another, there is elaborate power control, not only to accept feedback from the power detectors in the external PAs, but also to switch circuits in and out as needed within the transceiver itself. And there is elaborate software control via either an SPI or GPO interface to the external processor. Fujitsu has provided an application-mode-oriented API for the control processor that allows handset designers to set up the chip for a particular operating mode with a single command, rather than having to go through a litany of individual command registers, setting each function within the chip at the API level. All the flexibility is there, but you don’t need to know about it.

Then there are the major advances. Bhan said that there is no magic internal filter that replaces the external SAWs. Instead, the RF designers made the SAWs unnecessary by carefully analyzing all of the phase-noise sources in the transmit and receive chains and attacking each one. On the receive side, this meant reorganizing the signal-path lineup significantly, Bhan said. It also meant applying an entirely new architecture to implement the LNAs in 90 nm silicon CMOS. Internal noise sources the designers could minimize. External sources they had to deal with by use of clever IP.

In the process of rethinking the design, the architects chose to make the transceiver as independent of the baseband processor as possible. For example, rather than depending on a feedback signal from the baseband chip to control the receiver AGC, the transceiver does its own gain control. "If you are careful, there is enough information available in the chip to set the AGC internally," Bhan observed. Similarly, you can configure the chip to use either open- or closed-loop power control with the PAs.

All of these achievements did not cost performance, Bhan insists. He claims that the receiver has world-class EVM performance, making the chip suitable for high-data-rate applications in smart phones. And the flexibility is certainly there, covering WCDMA bands I through XI, excluding band VII, supporting EGPRS and HSPA, and the necessary 2G stuff. That makes the chip ideal, Fujitsu says, for quad-band UTMS and EDGE service.

As competition tightens for handset sockets, it appears that the winning skill set will have to include not just an understanding of RF IC design and the air interface standards, but also a comprehensive view of what handset manufacturers will have to do to use the chip in a real phone. Increasingly in this, as in so many other markets, the silicon vendor is becoming the systems expert.