Digitize early, then process" has been the theme for most designs over recent decades. The singular role of analog circuitry is to bring the signal--which represents a physical parameter (temperature, pressure, EKG)--to the processor as quickly as possible, then step aside and let the algorithms, driven by clock cycles, get busy. The approach makes a lot of sense, given the capabilities of signal-chain components, converters and processors.

Or maybe not.

At last week's TI Developer Conference, Professor Rahul Sarpeshkar of MIT's Analog VLSI and Biological Systems Group gave a fascinating talk on his team's work on biomedical electronics for cochlear implants, brain-machine interfaces and the like. He emphasized that this was a very low-power world, since the electronics must run from power transmitted across a skin barrier, or an implanted battery that must last 10 to 20 years (even if the battery is rechargeable, you only get up to about 700 cycles). Go through the power numbers, and you are talking about picoamps, femtoamps and even attoamps from a subvolt source, rather than the microamps and nanoamps we normally associate with "low power."

How did his group achieve these goals? By moving the A/D transition point (if any) much further back in the signal chain, and performing much of the computation in the analog domain.

This is not just another academic, theoretical project: Sarpeshkar's group has designed, built and tested ICs that employ significant analog signal processing as well as amazing architectural innovations.

Further, since this is a university group, it has published detailed papers (www.rle. mit.edu/avbs) and done comprehensive analyses on the trade-offs of signal processing in the analog and digital domains, to find the optimum location of the transition point between the two. The results are fairly contrary to conventional thinking but have solid data and tangible proof to back them up. The analog circuitry is also far smaller than a digital equivalent.

There is some irony here. Decades ago--before processors became the heart of a system and software was the answer to every computational question--all-analog ICs were used for multiplication, division, rms-to-dc conversion, peak detection and log- arithmic calculations. They implemented accurate calculations, in real-time, for such apps as power measurement and closed-loop control.

Technology's progress and your corresponding design strategy are not a straight line; there are curves, bumps and inflection points. Sometimes, it comes back to where it began. You have to be open to the "old" when dealing with the "new." Just doing more of the same, admittedly with far better components, may not be the right answer.