Practical Chip Design

A Monday-afternoon panel at the GlobalPress Summit yesterday provided a multifaceted survey of low-power design, from the perspectives of tool providers, analog and power component providers, DRAM vendors, and one particularly interesting chip designer. Recorded and expanded a bit, the discussion would have been a clinic in the intricacies and the scope of design for energy-efficiency. But one very interesting point emerged from two comments made by two different panelists that suggests a radical way of thinking about low-power architecture.

One of the comments came from the aforementioned interesting chip designer: Michel De Mey, senior director of hearing/audio solutions at ON Semiconductor's recently acquired AMI Semiconductor group. In describing how one goes about designing a complete audio system--comprising microphone, pre-amp, A/D converter, DSP, D/A converter and power amplifier—to operate at 1 mW, De Mey commented that part of the architectural approach had been rethinking the partitioning of tasks between analog and digital hardware.

These days, almost everyone automatically thinks of any filtering operation, if it's not simply the anti-alias filter on the front end of an A/D, as a digital operation. But that assumption does not necessarily lead to low energy consumption. In some applications, an analog filter can provide a sufficiently accurate response curve with low enough in-band attenuation to be significantly more energy-efficient than an equivalent digital filter. The analog filters can, with forethought, even be digitally tuned.

The second point came from a quite different direction. National Semiconductor marketing manager Rick Zarr also discussed analog/digital partitioning decisions, but in a very different application. He pointed to a continuous-time sigma-delta converter that National announced earlier this year. National claims that the converter, which uses a continuous-time sigma-delta modulator in an otherwise fairly conventional architecture, requires 30 percent less power than similarly-speced devices.

The point is not, of course, that analog is inherently more energy-efficient than digital. It is that under some conditions, signal processing can be more efficient in the continuous-time than in the discreet-time domain. This fact demands that designers looking for low energy consumption stop and think about their choice of domains for signal-processing tasks, rather than automatically starting out the block diagram on the left-hand side of the sheet with an A/D.

Beyond this lies unknown land. What about processing of signals that aren't already analog? Can analog techniques be applied to inherently digital signals with energy savings? And what about tasks that go beyond simple signal conditioning? The observation that a sigma-delta modulator can be implemented as a continuous-time device raises the question of what other functions, beyond simple filters, might be done more efficiently in the analog domain? I suspect there are some great thesis projects here, and, for those who are paying attention to the academic work, some great products.