photo by Gary Meek Associate Professor Paul Hasler, right, works with a student in a lab at the Georgia Tech Analog Consortium.

Commenting recently on issues facing neuromorphic engineers, Hasler noted that humans have at least 100 billion neurons in their brains, all powered by about 20 watts supplied by the nervous system.

A theoretical computer able to emulate the brain, he says, might use one Pentium chip – today’s digital-computing powerhouse – to emulate each neuron. That would translate into a power requirement of several trillion watts.

To succeed, he says, neuromorphic designs will likely require extensive use of low-power analog technology.

photo by Gary Meek Associate Professor Gabriel Rincón-Mora, left, makes a point to graduate student JinJyh Su about piezoelectric devices.

“And that, of course, is a simplified version.”

– Gabriel Rincón-Mora, associate professor, Georgia Tech School of Electrical and Computer Engineering

photo by Gary Meek Researcher Stephane Pinel, left, and Professor Joy Laskar display a high-speed multi-gigabit wireless IC.

“All of a sudden that becomes a rather non-trivial problem. I’m going to have a clock signal moving at a few gigahertz, I’m going to have data rates approaching a few gigabits per second, and I’m going to be moving on cheap board material that’s going to have impairments. And I can’t just solve the IO problem by what I’ll call a set of scaling rules, or rules that are necessarily exactly repeatable.”

– Joy Laskar, Schlumberger Chair in Microelectronics, School of Electrical and Computer Engineering; director, Georgia Electronic Design Center

photo by Gary Meek Associate Professor Linda Milor holds a silicon wafer.

“My strategy is to make it less of an art form. There have been many attempts to try to come up with automatic synthesis of analog circuits, but they haven’t worked – yet.”

– Linda Milor, associate professor, School of Electrical and Computer Engineering; Georgia Tech Analog Consortium

That means analog circuits can be preferable for many tasks suited to future mobile devices, including speech recognition, audio processing, and image and video processing.

Working with ECE colleague Paul Hasler, Anderson has been researching reconfigurable analog and mixed-signal systems at GEDC. The traditional problem with analog systems, he says, is that users cannot simply change function by changing what’s in memory, as they can with a digital system. Instead, they have to go through a costly and time-consuming redesign process.

GEDC researchers, he says, have developed analog chips that can be reconfigured on the fly to perform a variety of tasks. Given what amounts to a software change, this type of analog chip can do a different type of processing.

Such reconfigurable analog chips are not as adaptable as digital chips, but are more adaptable than previous analog designs. In an audio analog chip, for example, one algorithm might clean up audio, then switch to modem processing with a simple software change.