提出一种浮点型数字信号处理器 (DSP)硬核结构，在兼容定点数运算的同时，也为浮点数运算提供较好支持。目前各大现场可编程门阵列 (FPGA)主流厂商在实现浮点数运算功能时均采用软核实现方式，即将浮点数运算算法映射到芯片上，通过逻辑资源和 DSP模块实现。相比于传统方法，提出的硬核结构在不占用 FPGA中其他逻辑资源情况下，仅利用 DSP模块便能完成浮点数运算。设计中，充分考虑负载和时延影响，插入多级流水线，显著提高浮点数的计算效率。采用中芯国际 (MCI)28 nm工艺设计并完成所提出的浮点型 DSP硬核结构。仿真结果表明，所提出的硬核结构的单个浮点数加法和乘法效率为 0.4 Gflops。

A floating-point Digital Signal Processor(DSP) hardcore structure is proposed. The hardcore structure not only is compatible with fixed-point operation, but also provides better support for floating-point operation. At present, the major manufacturers of Field-Programmable Gate Array(FPGA) use soft core to implement floating-point arithmetic, which maps floating-point arithmetic to the chip and realizes it through logical resources and DSP module. Compared with traditional methods, the proposed hard-core architecture can complete floating-point arithmetic only by using DSP module without occupying other logic resources in the FPGA. At the same time, in the design process of the DSP, the influence of load and time delay is fully considered, and the multi-level pipeline is reasonably inserted, which greatly improves the calculation efficiency of floating-point number. The floating-point DSP hard-core structure is designed and completed by the Semiconductor Manufacturing International(MCI) 28 nm technology. The simulation results of the proposed hard-core structure show that the single floating-point efficiency of addition and multiplication is 0.4 Gflops.