University of California, Berkeley, Civil & Environmental Engineering
Ultrasound remains a popular imaging modality due to its mobility and cost-effectiveness. As general purpose computing and DSPs are entering an era of multi-core architectures, the potential for parallel performance gains are significant when used properly. This paper explores the possibility of using a massively parallel processor array to meet real-time throughputs for mid-/back-end ultrasound processing. A many-core array of simple DSP cores, shared memories, and an FFT processor is shown to dissipate 87.79 mW for B-mode, 33.20 mW for color flow, and 29.24 mW for spectral doppler, while achieving a frame rate of 37.6 fps for B-mode and 12.5 fps for color flow.