An efficient approach to engage neural net hardware to PC

This paper addresses a major bottleneck in engaging artificial neural network (ANN) hardware as a processing element to a conventional PC architecture, and proposes an efficient solution. Adding ANN hardware as a special processing element to conventional PC would allow the benefit of parallel operations in artificial neural networks to be fully exploited. However, the major bottleneck for such an addition is the limited memory bandwidth of the conventional computers. The proposed approach attempts to supply a high memory bandwidth for the ANN processing element with minimal modification to the PC architecture