A massively parallel, multiple-SIMD architecture for implementing artificial neural networks
L.B. Jump · 2003
The author presents a massively parallel architecture for implementing large scale artificial neural network (ANN) models. The architecture is multiple-SIMD (single instruction, multiple data), modular and programmable. Processing elements are organized locally as SIMD vector processors with a local ring communication structure. These vector processors are embedded in a global communication structure and each processing ring executes a potentially different program. The design of a prototype system employing this architecture for a programmable ANN workstation is discussed. Measurements and results taken from a hardware prototype are presented.>