基于DCORDIC算法的Tanh函数硬件设计与实现

吴绮; 金楚丰; 陈刚; 鲁华祥; 陈旭

Chinese

您当前的位置：

首页 >

文章列表页 >

基于DCORDIC算法的Tanh函数硬件设计与实现

更新时间：2026-09-04

- 基于DCORDIC算法的Tanh函数硬件设计与实现
- Hardware design and implementation of Tanh based on DCORDIC algorithm
- 微电子学与计算机 2020年37卷第12期页码：22-26
- 作者机构：
  
  1. 中国科学院机构半导体研究所,北京,100083
  2. 中国科学院大学机构,北京,101408
  3. 电子科学与技术大学,四川,成都,610054
  4. 中国科学院脑科学与智能技术卓越创新中心,上海,200031
  5. 半导体神经网络智能感知与计算技术北京市重点实验室,北京,100083
- 作者简介：
- 基金信息：
  
  国家自然科学基金项目 61701473中国科学院战略性先导科技专项 XDA18000000
- DOI：
  中图分类号：
- 纸质出版：2020
- 稿件说明：
移动端阅览
吴绮, 金楚丰, 陈刚, 鲁华祥, 陈旭. 基于DCORDIC算法的Tanh函数硬件设计与实现[J]. 微电子学与计算机, 2020, 37(12): 22-26, 32.

WU Qi, JIN Chu-feng, CHEN Gang, et al. Hardware design and implementation of Tanh based on DCORDIC algorithm[J]. 2020, 37(12): 22-26.
吴绮, 金楚丰, 陈刚, 鲁华祥, 陈旭. 基于DCORDIC算法的Tanh函数硬件设计与实现[J]. 微电子学与计算机, 2020, 37(12): 22-26, 32. DOI：

WU Qi, JIN Chu-feng, CHEN Gang, et al. Hardware design and implementation of Tanh based on DCORDIC algorithm[J]. 2020, 37(12): 22-26. DOI：

摘要

双曲正切函数(Tanh)在神经网络中经常被用作激活函数，而其非线性的特点导致难以用硬件电路直接计算实现.为了快速有效的实现Tanh函数，提出采用DCORDIC(Differential Coordinate Rotation Digital Computer)算法在不同模式下实现双曲正弦、双曲余弦函数和除法运算，并在设计中增加选择模块扩大输入数据的范围，最终设计实现的Tanh计算核具有超长流水线，当迭代次数为13时，计算误差在2以内.与现有的硬件实现方式相比，本文的实现方式具有更小的误差、更高的运算速度和更大的吞吐率

Abstract

Hyperbolic tangent function (Tanh) is often used as an activation function in neural networks

but its non-linear character makes it difficult to be realized by hardware circuit directly. In order to implement the Tanh function quickly and efficiently

the DCORDIC(Differential Coordinate Rotation Digital Computer) algorithm is proposed to implement the hyperbolic sine/cosine function and division operation in different modes

and a selection module is added to expand the range of input data. The final design implementation of the tanh computation core has a very long pipeline with a computational error of 2. Compared with the existing hardware implementation methods

the proposed implementation method has smaller error

higher computing speed and greater throughput.

关键词

Keywords

references

ZHANG M, VASSILIADIS S, DELGADOFRIAS J G, et al. Sigmoid generators for neural computing using piecewise approximations[J]. IEEE Transactions on Computers, 1996, 45(9): 1045-1049. DOI: 1045

LEBOEUF K, NAMINA H, MUSCEDERE R, et al. High Speed VLSI Implementation of the Hyperbolic Tangent Sigmoid Function[C]//2008 Third International Conference on Convergence and Hybrid Information Technology. Busan: IEEE, 2008: 1070-1073.

BASTERRETXEAK, TARELAJ M. Approximation of sigmoid functionand the derivative for artificial neurons[C]// Zeng Z G. advances in neural networks and applications. Athens: WSES Press, 2001: 397-401.

ZHU W J. Research and development trend of improved CORDIC algorithm[J]. Popular Science and Technology, 2010(4):17-18+25. DOI: 17

BASTERRETXEAK, TARELAJ M, CAMPOI D. Approximationof sigmoid function and the derivative for hardware implementation ofartificial neurons[J].IEE Proceedings - Circuits, Devices and Systems, 2004, 151(1):18-24. DOI: 18

LIN C, WANG J. A digital circuit design of hyperbolic tangent sigmoid function for neural networks[C]// 2008 IEEE international symposium on circuits and systems. Seattle: IEEE, 2008: 856-859.

PARHAMI B. Generalized signed-digit number systems: a unifying framework for redundant number representations[J]. IEEE Transactions on Computers, 1990, 39(1):89-98. DOI: 89

SONG Y K, GAO X H, ZHANG D L, et al.The piecewise non-linear approximation of the sigmoid function and its implementation in FPGA[J]. Application of Electronic Technique, 2017, 43(8):49-51. DOI: 49

CHANG A X, MARTINI B, CULURCIELLO E, et al. Recurrent Neural Networks Hardware Implementation on FPGA[J]. International journal of advanced research in electrical, electronics and instrumentation engineering, 2015, 5(1): 401-409. DOI: 401

SHRUGAL V, NISHA S, RICHA U.Hardware Implementation Of Hyperbolic Tan Using Cordic On FPGA[J].International Journal of Engineering Research and Applications (IJERA), 2013, 3(2):696-699. DOI: 696

GOMPERTSA, UKILA, ZURFIUHF. Development and Implementation of Parameterized FPGA-Based General Purpose Neural Networks for Online Applications[J].IEEE Trans. On IndustrialInformatics, 2011, 7(1):78-89. DOI: 78

浏览量

113

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

基于区块链的AI生成内容溯源框架设计研究

基于监控指标关联性的微服务异常检测

低功耗设计的双链电源关断策略的实现与验证

AlN/GaN混合缓冲层双异质结p-GaN HEMT直流特性仿真

注意力机制与改进密集卷积引导的信息隐藏模型