A Computer Like Your Brain

Top: The Three-Dimensional Artificial Neural Network processor is capable of recognizing objects in real time and in highly cluttered background scenes. Photo provided by Jet Propulsion Laboratory.

A new NASA-developed computing device allows machines to work much like the brain. This technology may allow fast-thinking machines to make decisions based on what they see. A planetary rover might use this technology to avoid obstacles, select scientifically interesting spots to explore just by what it sees and navigate through terrain on its own without review from ground controllers. A spacecraft might use the technology to avoid hazards and identify a pre-selected landing site with very high precision.

�This may well be recognized as a quantum leap in the pursuit of intelligent vision, allowing machines to be significantly more autonomous,� said Dr. Anil Thakoor, supervisor of the Bio-Inspired Technology and Systems Group at NASA�s Jet Propulsion Laboratory in Pasadena, California.

The device works much like the brain, whose power comes from the complex networks of interconnections called �synapses� between brain cells. Networks of these brain cells, called neurons, allow humans to make instant decisions based on an observed image or scene. The new processor captures the same capability to process images in real time as a scene unfolds.

The Three-Dimensional Artificial Neural Network processor is capable of recognizing objects in real time and in highly cluttered background scenes. It can process an image and is capable of a certain degree of judgment about the objects, much the same way a person looks at a variety of objects and makes judgments about the nature of those objects.

Two technologies give the compact processor an unprecedented ability to process a stream of images in a way similar to that used by the human eye-brain combination. One is the JPL-pioneered, highly interconnected networks of ultra-low-power electronic synapses on very large-scale-integrated (VLSI) chips that mimic the core of a brain. The other is the three-dimensional stacking of those chips in a sugar cube-sized package developed by Irvine Sensors Corporation of Costa Mesa, California. Irvine Sensors is a successful NASA SBIR firm that has commercialized the stacked chip technology.

The device achieves a computing speed of more than a trillion operations per second, using only eight watts of power. That is more than a thousand times faster than a typical commercially available desktop computer that consumes more than 100 watts of power. Engineers believe potential commercial benefits for the new technology may be found in public safety and in creating a personal computer that can respond to users� emotional states by simply recognizing the users� facial expressions. The development may also be useful to the video game industry in improving interactive technologies.