How Ultra-Low-Power Operation is Achieved

1. The use of special-purpose programmable architectures that balance flexibility with efficiency well.

2. The use of analog basis functions for efficient preprocessing before digitization or signal-to-symbol conversion.

3. The use of slow-and-parallel collective analog and mixed-signal architectures.

4. The exploitation of exponential subthreshold or 'leakage' currents to do analog and digital processing due to the high speed per watt in this regime, and increasingly faster operation (>1GHz) in modern nanometer processes.

5. The balance of computation and communication costs (bits processed versus bits transmitted).

6. The use of adaptive-and-feedback architectures to reduce errors in signals and devices, to focus resources on informative portions of the input space, and to operate at optimal points in topology, current, voltage, or device-parameter space.

7. The reduction of information to be processed by exploiting knowledge and learning and by mapping function to structure well.

8. The balance of resources needed for robust operation in the presence of thermal noise, 1/f noise, device parameter variations, power-supply noise, and temperature variations with the resources needed for implementing the actual processing.
9. The use of revolutionary biologically-inspired architectures that implement novel algorithms and signal processing and that force us to think outside the box.

Several of the listed engineering principles, e.g., a robustness-efficiency or flexibility-efficiency tradeoff apply to biological design as well.

Why engineers should care about biology

Biological systems have developed over hundreds of millions of years of evolution to perform sensory, motor, and chemical tasks extremely efficiently and robustly whilst using very little power, in very little volumes, and in real time. The average neuronal cell in the human brain consumes less than a nW of power, and the average cell in the body uses approximately 1pW of power. The entire brain and body are put together with energy-efficient neurons and cells to robustly perform complex information-processing tasks in the chemical, mechanical, or electrical domains with about 12W and 100W of power, respectively. The human inner ear or cochlea is a clever custom analog computer that does more than 1GFLOPS of filtering and gain-control computations at 14uW of power due to its integration of microfluidics, micromechanics, microelectronics, and electrochemistry. The cochlea could run on a AA battery for 15 years. Much can be learned from biology to develop efficient technologies, to integrate technologies well, to develop sophisticated control systems, to learn to architect systems that can perform efficiently and reliably with unreliable devices, to build systems that automatically learn and adapt to a changing environment, and to build systems that can recover gracefully from failure. Indeed, engineers, physicists, and chemists working on MEMS, low-power electronics, biomedical devices, sensitive molecular detection, chemical synthesis, fabrication, energy harvesting, and robotics are all humbled when they realize how hard it is to replicate some of nature's awesome feats!

It is important to copy nature insightfully, i.e., to keep the baby and throw out the bathwater. Blind mimicry can lead to a degradation in engineering performance since the constraints and purpose for which the biological system was designed may not be exactly that for which the engineering system is designed. Often, we also don’t know enough about nature to understand why it is architected the way it is or its architecture may have be a frozen accident of evolution. Thus, it is still important to try to evaluate one’s overall success when one is done by the same metrics used to judge traditional engineering architectures. As in all interdisciplinary fields, it is important to synergistically combine the creativity and excitement generated by new non-traditional thinking with the discipline and knowledge of older ideas.

Healthcare is a natural area for applying biologically inspired technologies since we are trying to engineer systems that perform the normal functions of biological ones, so mimicking the biology can be helpful in fixing it. Our lab has done work on designing analog cochlear implant processors and cochlea- inspired algorithms for the deaf that have dramatically lowered power and improved performance in noise. We have used inspiration from how neurons do pattern recognition to create a time-based A-to-D for biomedical applications that is currently at or very near the state of the art in energy efficiency (0.12pJ per quantization level). In collaboration with biologists, we have begun to work on extremely energy-efficient prosthetics for curing paralysis by decoding movement intentions from neurons in the brain and using these outputs to control prosthetic or natural limbs. In non-healthcare areas, we are working on taking inspiration from the architecture of the inner ear or cochlea to design a very energy-efficient 'RF Cochlea' that can detect signals rapidly over a very wide carrier-frequency range; such RF cochleas have applications in universal, software, or cognitive radios. We are investigating how cells process information with sophisticated DNA-protein networks to design efficient hybrid analog-digital cellular processors for various applications.

Back to Home Page