The heart of the NNP family is the MIMD Neural Network Processor, which provides the basic unit of processing. Multiple NNPs can operate in parallel through a unique bus and connector design.

PCI, ISA, and VME busses are supported with interface boards that have up to two built in NNPs. Additional NNPs can be stacked on the baseboard. The VME board includes two TMS320C40 DSP Chips.

The boards are supported on a variety of PC and SGI platforms.

NNP Features

• True Multiple Instruction-Multiple Data (MIMD) architecture
• NNPs Stack to Max of 10 Processors
• 8,192 Neurons per system
• 32,768 Connections per processor
• 32 million multiply-accumulate cycles per second per processor
• Unique Neural Network Processor Language; neural network tools and compiler provided with the NNP.
• Program and run standard paradigms or your own unique paradigm
• Powerful architecture concept that avoids memory bottlenecks.

System Requirements

The following hardware platforms are supported:

• PC Platforms running Windows 98 or NT 4.0:
• PC with ISA Bus
• PC with PCI Bus
• Silicon Graphics Platforms running IRIX 5.2 and above:
• Indy using Bit3 Corporation GIO32 to VME bus adaptor
• Indigo using Bit3 Corporation GIO32 to VME bus adaptor
• Indigo 2 using internal EISA bus
• Onyx using internal VME bus
• Power Series with single IO3 board using internal VME bus
• VME
• VME 6U dual TMS320C40 DSP card with the ability to support one to four Neural Network Processors

Support Software

Accurate Automation Corporation's programming tools include an assembler, linker, debugger, and diagnostics program that allow users to quickly design and use networks on the NNP.

Applications

Neural Networks Processors are being used to speedily solve difficult problems in many areas. Some current applications of the NNP include: