F-TFM: Accelerating Total Focusing Method for Ultrasonic Array Imaging on FPGA

Bizhao Shi, Jieran Zhang, Guojie Luo

Abstract: The total focusing method (TFM) is a widely employed imaging algorithm utilized for processing ultrasonic signals in various applications. Currently, there is a growing demand for real-time efficient TFM processing. However, existing research struggles to achieve satisfactory throughputs and optimal energy efficiency. In this work, we present F-TFM, an efficient TFM imaging system on FPGA. Firstly, we analyze and explore the trade-off between re-computation and memory access in TFM and organize the efficient workflow for high-efficient processing on FPGA. Secondly, we propose an efficient accelerator architecture with carefully designed dataflow as well as optimized hardware parameter configurations. Evaluations demonstrate that F-TFM achieves remarkable improvements over existing works, while maintaining high image quality. Specifically, F-TFM achieves up to $5.43 \times$ speedups and $47.15\times$ energy efficiency improvements compared with the high-end GPU baseline and achieves up to $35.94 \times$ speedups compared with the FPGA baselines.