Low-Power RF for SOC Design
I’ll be the first to tell you that RF isn’t my thing. I’m a digital guy. However, even someone who lives in Boolean space like me can find things of interest in today’s keynote address on “Low-Power RF Transceiver Design Strategies for SOC” given by Kaveh Kianush at the International SOC Conference being held in Tampere, Finland. Kianush is the CTO and VP at Catena Radio Design. He launched right into the issue of the day: higher integration may be great for digital circuits but they create headaches for RF designers. Symptoms include more crosstalk (digital noise injection into power and signal lines), unacceptable power characteristics, and cost issues.
The biggest challenge Kianush addressed in his keynote is the problem of co-located radios (GPS, Bluetooth, and cellular for example). You want to put multiple transceivers on one die for maximum integration and lowest cost, but getting all those radios to play together nicely is a big problem because of interference. Further, trying to build such radios on scaled silicon is also a problem because the reduced Vdd that always seems to accompany smaller device geometries reduces the radios’ signal-handling abilities, creates more leakage (because of thinner oxides), and increases 1/f (flicker) noise.
Device scaling doesn’t help RF transmitter power dissipation because, as it turns out, the transmitter power is dominated by government regulation. It’s not really a function of technology the way digital power dissipation is.
RF transceivers on SOCs combine analog, RF, mixed-signal, and DSP circuits. It’s an ugly brew. So there’s a trend towards making the transceiver as digital as possible. This trend points to software-defined radio (SDR), which appears to be this decade’s Holy Grail for the RF world. A mostly digital transceiver can use DSP to compensate for the imperfections of low-cost analog circuits using calibration and correction techniques.
In some RF applications, the average power dissipation is determined not by the active power but by the standby power, where the RF transceivers are powered off. Only the processors and buses are active. That’s because there are long standby periods and short active periods for some radio links such as ZigBee. Kianush showed a ZigBee example where the radio is on for 1 msec with standby periods lasting 100 msec to 4 sec. In this example, said Kianush, a 1 microamp standby current would result in10x more energy drain than the 20 mA transmit current, because of the low transmit duty cycle.
Process migration also contributes to leakage. Going from 180 nm to 130 nm can increase leakage up to 10x with a constant supply voltage. The culprit is thinner gate oxide, which is down to five atomic layers (1.2 nm) at 90nm lithographies. One way to circumvent this issue is to run the system’s nonvolatile memory on a separate power supply and switch off everything else during standby operation. One bright spot: the 45nm processes with high-K dielectrics based on Hafnium have orders of magnitude lower leakage than 65nm processes thanks to the thicker gate oxides made possible by the new dielectric material.
For RF transmitter circuits, the efficiency is largely determined by the efficiency of the power amp (always abbreviated as “PA” in RF discussions). GSM, Bluetooth, and ZigBee transmitters can use Class C PAs, which are not very linear but very efficient relative to the highly linear Class A PAs required by mobile cellular radios. Receiver power consumption is determined by dynamic range requirements, controlled by the relationship between the noise floor and the maximum expected signal.
Finally, Kianush stated something obvious, but it did jog my brain. His company is endeavoring to design inductor-less transmitters and receivers. Why? “Because the digital guys can put thousands of gates onto the same piece of silicon needed by just one inductor.” And that’s why it’s a digital world.
