Low Area FPGA Implementation of Hyperbolic Tangent Function

The hyperbolic tangent function is a crucial aspect of navigation, communication, and activation function in neural networks and other modern digital signal processing systems. This paper presents a new techniques methodology for hardware implementation of the hyperbolic tangent function. Many techniques are used to implement the tanh function, such as the CORDIC algorithm, Tylor series, polynomial curve fitting, and look-up table. The ROM approach and traditional and modified polynomial curve fitting method are the techniques used in this paper. The main aim of this work is to minimize resource utilization, low power consumption, and high-speed execution of a hyperbolic tangent function. The Spartan 3A-3N FPGA is the platform used to synthesize the proposed designs. The suggested system based on the traditional polynomial curve fitting achieved the highest speed of one clock cycle; it consumed a considerable amount of area, utilizing 590 L UTs and 345 occupied slices. Whereas the modified polynomial curve fitting had a large area consumption of 108 LUTs and 66 occupied slices, along with 22 Flip Flops, with a delay of six clock cycles. In contrast, the proposed design based on the ROM approach achieved accurate results with small area utilization: 47 look-up tables and 25 occupied slices. Furthermore, the approach achieves high speed, which consumes only one clock cycle. In addition, the power optimization percentage when comparing the ROM approach to the traditional polynomial, and modified polynomial curve fitting are (30 %) and (25%), respectively. And the operating frequency for the ROM approach is about (315 MHz), whereas for the modified curve fitting is about (81 MHz).