کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

Approximation Approaches for Efficient Hardware Implementation of Sigmoid and Tangent Hyperbolic Activation Function on FPGA: A Comaparison Study

نویسندگان :

Seyed Mohammad Ajami Behbahani¹ Shahrzad Ajabi² Yousef Seifi Kavian³ Mehdi Naderisooraki⁴

1- دانشگاه شهید چمران 2- دانشگاه شهید چمران اهواز 3- دانشگاه شهید چمران اهواز 4- دانشگاه شهید چمران اهواز

کلمات کلیدی :

FPGA implementation،hardware efficiency،sigmoid function،hyperbolic tangent function،nonlinear activation function،approximation computation

چکیده :

In recent years, as artificial intelligence has advanced, hardware implementation of neural networks with low power consumption and high computational throughput has become increasingly essential. In this context, optimizing neural network operations, particularly non-linear activation functions, plays a key role in improving performance and reducing hardware overhead. This paper investigates the implementation of sigmoid and hyperbolic tangent activation functions on FPGA platforms. Our analysis includes look-Up table method, piecewise linear approximations, iterative computation-based approaches and hybrid techniques. Finally, we compare each method's accuracy with MATLAB and subsequently evaluate its performance on an AMD Kintex-7 FPGA (XC7K325T-FFG676) in terms of resource utilization, power consumption, and latency.