신경회로망을 위한 병렬 DSP설계

초록

Artificial neural network has been utilized for various applications. it is generally constructed at the learning stage and implemented or tested at the recognition stage. The learning stage of the network is computationally expensive. For the high speed calculation, thr design of custom made VLSI chip and accelerator board based on DSP chips have been proposed. This paper presents an accelerator board for the artificial neural network applications. It is designed based on the DSP chips and peripheral circuits are integrated by FPGAs. The board is designed as a modular structure to provide flexibility in expansion and modification. The performance of the board is evaluated by applying back propagation algorithm.