Recognizing patterns is something computers do very well. Unlike variables such as context and concepts, which computers are notoriously poor at processing - as most people who have searched the Internet can attest - computers excel at identifying complex patterns of words and symbols. But doing something well and doing it fast are not necessarily the same thing.
To remedy the disparity, a Colorado Springs start-up called Neo-Core recently announced technology to bridge the gap between accuracy and speed in pattern-recognition software. The innovation lies in a new processing architecture that allows computers to associatively - rather than sequentially - recognize and manipulate data. Sequential processing is linear - each piece of data in a data set must be queried to arrive at a pattern. Neo-Core's associative-processing technology, however, generates 32-bit icons for data, and associates these icons with known patterns, up to 100 million associations per second per processor. The result is performance "orders of a magnitude faster than what's out there today," said Mike Young, vice president of marketing at Neo-Core. For example, the throughput in the average router is approximately 250 thousand packets per second. Neo-Core can increase that rate to 30 million per port, Young said.
Neo-Core has its roots in parallel-processing computing, an architecture employed in large distributed systems and mainframes. In fact, the chief technologist at Neo-Core is Chris Brandin, considered one of the pioneers of parallel processing systems from the 1970s.
With broad ambitions for the technology, Neo-Core is in discussions with browser and search-engine companies to integrate it with table-lookup and IP address look-up applications in Web browsers. Neo-Core's associative processor must be embedded directly on an integrated circuit.
The company, which declined to disclose its potential partners, also hopes its technology will eventually be used in image-recognition software, fingerprint-identification devices, and radar image-recognition applications. But possible applications also include conventional enterprise applications, such as data communications, network address translation, database indexing and searching, and data scrubbing.
Neo-Core identifies filtering, in particular, as a strength, which can dynamically assign filters in protocol analysis programs. Once again, the result is increased speed in which the technology identifies the patterns required for filtering.
To date, Neo-Core has made public one company that has signed on. UTMC Microelectronic Systems, a maker of chip technology for the aerospace industry and the military, plans to release an integrated circuit containing Neo-Core code customized for specific military applications by Q1 1998.
"The Neo-Core circuit starts out as a tough sell, because it's not intuitively obvious," said Ron Hehr, strategic marketing manager at UTMC, which is also based in Colorado Springs. "It's a significant departure from the architecture [the military] is currently using, so you have to show them the potential - the performance is just awesome. When they see the speed, they buy into it. But you do have to go into it with a clean sheet of paper."
UTMC will co-develop the military applications of the Neo-Core technology, Hehr said, while Neo-Core continues to concentrate on commercial vehicles for its technology. One of the military applications might include "target identification systems for weapons systems," Hehr said.
"Our business model is to identify the applications - existing or potential - and present the solution to a relevant company. We prefer to develop a strategic collaboration with companies so we can help them architect their products," said Young.
