Toroidal neural processor TNP

This section describes the background to the TNP architecture and then proceeds to describe the support it offers for neural model emulation. 

The TNP architecture was inspired by the need for a low-cost massively parallel processing system that could emulate a large variety of neural models. The role envisaged for the TNP architecture is to serve as a tool allowing the exploration of neurocomputing problems. 

Such an architecture must be flexible enough to satisfy the different forms of interconnect and computational requirements demanded by the various neural models. The suitability of the linear array architecture for providing interconnect flexibility has been addressed above. 

If the TNP design is to support a wide range of neural network models, then it must implement a number of key features. These include  

The requirement for a simple instruction set was primarily motivated by the desire to observe the primitive operations of neural algorithms. Initial experimentation with various instruction sets indicated that the use of a reduced instruction set numbering around 8-12 instructions can satisfy most of the computational requirements of neural models. 

---