Nestor and Intel Show Neural Net “Recognition Accelerator”

The Ni1000, a neural network chip jointly developed by Intel and Nestor, has been delivered to the Defense Advanced Research Projects Agency (DARPA) and Office of Naval Research (ONR). The 1024-neuron, 3-million transistor neural network chip can perform 20 billion integer operations per second. Additionally, the two companies have launched a beta-site program that will enable VARs, OEMs, and system integrators to begin application development projects.

Intel and Nestor claim that Ni1000-based “recognition accelerator” can achieve character recognition rates at up to 10,000 characters per second, far greater than the 10 to 100 characters per second typical of most PC-based recognition software. The chip uses a large block of Flash memory so that learned patterns can be “memorized” and quickly “recalled” for real-time pattern recognition applications. Learning capability is implemented on-chip in the form of a 16-bit microcontroller.

In a related announcement, Nestor introduced version 1.3 of its NestorReader, an intelligent character recognition software package. The new version provides: recognition of alphanumeric fields, including Canadian postal codes, product codes, and machine-part numbers; symbol recognition, including dollar signs, floating decimal points, and the like; character field matching, which is useful for catalog order entry and time card processing; and courtesy amount reading, remittance processing and order entry from unconstrained, handwritten checks. For more information, contact:

Intel Neural Hotline
408-765-9235
or
Nestor Inc.
1 Richmond Square
Providence, RI 02906
401-331-9640