CNN model on stream processing platform
László Füredi, Péter Szolgay · 2009
Stream processor, Inc. (SPI) has introduced the Storm-1 stream processing system-on-a-chip (SoC) which contains 16 data-parallel processing arrays with five very long instruction word (VLIW) arithmetic logic units (ALUs) in each lane (altogether 80 processing units). In the paper the implementation of emulated-digital cellular neural networks universal machine (CNN-UM) simulation kernel on the Storm-1 stream processing platform (SPP) is described and optimized according to the special requirements of SPP. The area/speed/power tradeoffs of our solution and other emulated digital CNN implementations are also compared.