Artist's illustration using binary numbers

A new computer chip that mimics how the human mind works is making its way from the space program to American industry and may end up in millions of American cars in years to come.

Computer scientists at NASA's Jet Propulsion Laboratory, Pasadena, CA, have made advanced neural network technology breakthroughs that can solve diagnostic problems in industries from automobiles and aerospace to manufacturing and electricity production.

JPL and the Ford Motor Co. have signed a licensing agreement for use of an advanced neural network technology to diagnose misfiring under the hoods of Ford automobiles, among its many potential applications. With the advent of this new chip, vehicles should show a reduction in emission levels.

The smart fit between JPL's neural net hardware and Ford's automotive engineering algorithm expertise will enhance the industrial giant's ability to meet ever-stricter Clean Air Act requirements as they apply to continuous onboard diagnostics and control, officials said.

In addition, the chip is designed to improve fuel economy, resulting in financial savings for car owners. Ford engineers do not predict a price increase for installation of the chip because JPL designed a computationally powerful neuroprocessor that could be mass-produced in a highly cost-effective way. The technology also improves customer satisfaction by virtually eliminating distracting false alarms about misfiring that vehicle dashboards can signal with current under-the-hood diagnostic technology.

JPL and Ford scientists say the chip represents the first significant change in the way computing is done on vehicles since computers were first introduced into automobiles in the 1970s.

"Neural networks are a new discipline, and diagnostics, prognostics and control is a huge field. Ford's application is but the tip of the iceberg of this chip's potential use in American industry as a whole," said Tom Hamilton, program manager at JPL's Dual-Use Technology Office, one of JPL's many technology transfer arms. "JPL is proud to be able to make this revolutionary technology available for U.S. business."

The licensing agreement comes on the heels of a less formal Technology Cooperation Agreement that had existed between JPL and Ford since 1993. Under terms of that agreement, JPL and Ford engineers worked cooperatively to refine applications of the emerging technology to Ford's specific needs.

The new license provides Ford with rights to intellectual property of the chip for auto industry applications, while JPL, which has applied for patents to the technology, retains general rights. JPL is managed by the California Institute of Technology, which serves as the party of record for this license.

Neural systems were inspired by the architecture of nervous systems of animals, which use neurons, a form of parallel processing elements, to process large volumes of information simultaneously. In vehicle applications, artificial neural networks will "learn" both how to diagnose problems like engine misfires and control the engine to optimize fuel economy and emissions.

"What JPL has brought to the table is expertise in designing and building what are known as neural network 'application- specific integrated circuits'," said Dr. Raoul Tawel, who led the development at JPL for the chip. "With Ford, we are implementing highly complex neural network software code in dedicated hardware logic. This brings about a tremendous boost in computational ability compared to traditional software-based approaches, enabling real-time onboard diagnostics for the first time."

For misfire diagnostics, it is necessary to observe and diagnose every engine firing event, estimated at over one billion in the life of each car.

In addition, the diagnostic error rate has to be extremely small, less than one in a million, in order to avoid sending false alarm signals to the driver. The new chip will accomplish that task by "learning" diagnostic tasks during the vehicle development process, bypassing the need to develop conventional software that, in any event, can neither perform these tasks as well nor be implemented in large production volumes with standard microprocessors. The neural network chip, designed to carry out parallel neuron computations efficiently, overcomes the computational barriers that prevent this technology from being exploited today.

A detailed, technical explanation of the technology written by Tawel and Drs. Ken Marko and Lee Feldkamp of Ford's neural network team, among several others, is available on the Web. "Custom VLSI ASIC for Automotive Applications with Recurrent Networks" can be accessed at

For further information about JPL's technology transfer programs, visit

News Media Contact