Not too long ago, hotshot hardware-design engineers accrued bragging rights based on how many millions of transistors their chips had or how fast they were. Thanks to Moore's Law, transistors and speed are now less frequently the limiting factors of chip design. Rather, minimizing power and its attendant heat now indicates a clever design. And few applications face as many power constraints as the RF nodes in wireless-sensor networks, which are resilient, self-healing, and able to operate for years with no maintenance on standard AAA or coin-cell batteries.

As a professor of electrical engineering at the University of California—Berkeley, Kris Pister, also chief technical officer of wireless-sensor start-up Dust Networks, directs research into both the hardware and the software that these networks need. We talked with Pister and his graduate student Ben Cook about their research into ultralow-power radio design, which Cook recently presented at the International Solid-State Circuits Conference (Reference 1).

What was Cook's challenge?

Pister: I told him to ignore all the conventional wisdom in radio design and trade off almost anything on the performance side to get to low power. He didn't end up trading much in the way of performance and still achieved low power: His radio burns 300 µW in receiving mode at 2.4 GHz, whereas the best commercial chips we're using right now at Dust Networks burn 30 mW. There's nothing in his radio that you can point to as radically lowering power consumption, but, if you look in CMOS-RF textbooks, all the basic ideas he used are there. It's just that most current radios have discarded them. Everything focuses instead on high performance at the expense of power.

Can you give three design ideas that really made a difference in power consumption?

Cook: First, use a simple modulation scheme. For our applications, an extremely high data rate is unnecessary, so we can use BFSK [binary frequency-shift keying]. Plus, we added more frequency separation than is common. For example, Bluetooth also uses a form of BFSK, which has a data rate of, say, 1 Mbps, but it jumps between frequencies of only 320 kHz, and, as a result, it's a lot harder to demodulate than 1 Mbit with a frequency separation of a couple of megahertz.

Second, passive-voltage gain is important. This idea flies in the face of current conventional wisdom that you've got to get your power gain at the beginning of the received-signal chain. CMOS devices take voltage, rather than power, as an input, and they have an almost purely capacitive input, so they can't absorb any power. If you apply your signal to the gate of a CMOS transistor, voltage amplification and getting that voltage gain using passive components that consume zero are the most important factors in overcoming the noise of that device.

Pister: A subthread running through the design process was the additional goal of using as few off-chip components as possible. You hear a lot of people talking about single-chip radios, but that idea usually means a single silicon chip with all the transistors on it and then a bunch of passive components—crystals, filters, inductors, and other components—off the chip. He was shooting for not just the lowest power, but also the lowest cost in the smallest form factor in these radios, with a single chip with the absolute minimum number of components.

Cook: Passive-voltage gain is not a universal solution. Some systems need power gain. For instance, cell phones have to interface to up-chip parts, such as SAW [surface-acoustic-wave] filters, and they need power gain in their amplifying stages, but we keep everything on-chip, and, therefore, we can interface to high impedances on the chip. So, we're not fixed to 50Ω, which RF components have traditionally used. We have flexibility with using voltage gain instead of power.

Third, minimize the overhead power consumption. For almost all radio topologies, you use a VCO [voltage-controlled oscillator] as the core of the transceiver, and we wanted to make it as low-power as possible, as well as eliminating as many other high-power blocks as possible. So, in the receiver, there's no RF amplifier in the front end, which is kind of a novel thing. Instead, the front end is a passive-voltage-amplifying network that goes straight to a passive mixer. Making everything passive in the front end drastically reduced power and still yielded good performance.