In recent years, the technological node used to implement FPGA devices has led to very high performance in terms of computational capacity and in some applications these can be much more efficient than CPUs or other programmable devices. The clock managers and the enormous versatility of communication technology through digital transceivers place FPGAs in a prime position for many applications. For example, from real-time medical image analysis to high energy physics particle trajectory recognition, where computation time can be crucial, the benefits of using frontier FPGA capabilities are even more relevant. This paper shows an example of FPGA hardware implementation, via a firmware design, of a complex analytical algorithm: The Hough transform. This is a mathematical spatial transformation used here to facilitate on-the-fly recognition of the trajectories of ionising particles as they pass through the so-called tracker apparatus within high-energy physics detectors. This is a general study to demonstrate that this technique is not only implementable via software-based systems, but can also be exploited using consumer hardware devices. In this context the latter are known as hardware accelerators. In this article in particular, the Xilinx UltraScale+ FPGA is investigated as it belongs to one of the frontier family devices on the market. These FPGAs make it possible to reach high-speed clock frequencies at the expense of acceptable energy consumption thanks to the 14 nm technological node used by the vendor. These devices feature a huge number of gates, high-bandwidth memories, transceivers and other high-performance electronics in a single chip, enabling the design of large, complex and scalable architectures. In particular the Xilinx Alveo U250 has been investigated. A target frequency of 250 MHz and a total latency of 30 clock periods have been achieved using only the 17 ÷ 53% of LUTs, the 8 ÷ 12% of DSPs, the 1 ÷ 3% of Block Rams and a Flip Flop occupancy range of 9 ÷ 28%.
Optimization and High-performance Hardware Implementation of Multipath Interference Compensation Algorithm
