Design of Dual-path Fused Floating-point Add-Subtract Unit

A Fused Floating-point Add-Subtract Unit is a module that takes two operands as inputs and provides both the sum and difference of the inputs as the outputs simultaneously. This unit is used widely in Digital Signal Processing (DSP) applications like Fast Fourier Transform (FFT) and Discrete Cosine Transform (DCT) butterfly operations. The unit is designed for Single-precision, Double-Precision and bfloat16 floating point data formats. The Dual-path algorithm is used to improve the performance of floating point fused addition/subtraction. The far and close path modules (Dual-path design) designed for higher performance is taken as a reference design and timing analysis is done individually on far and close path timing arcs. Depending on the timing analysis, a modified close-path design is proposed which does not compromise performance much when compared to the reference design and also consumes lesser area and lower power. The reference and modified designs are synthesized using a 28nm standard-cell library. The proposed design requires approximately 33% less area and 17% less power than the reference design for both Single-precision and Double-precision designs, and requires 27% less area and 12% less power than the reference design for bfloat16 design.Electrical and Computer Engineerin