...
  • 主题
高级搜索  >
历史搜索 清空历史
首页> 外文期刊>Journal of Low Power Electronics >0.5V Sinh-Domain Design of Activation Functions and Neural Networks
【24h】

0.5V Sinh-Domain Design of Activation Functions and Neural Networks

机译：激活函数和神经网络的0.5V正弦域设计

获取原文
获取原文并翻译 | 示例
           
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

Activation Function (AF) is the heart of Artificial Neural Network (ANN). It is typical building block of ANN which is difficult to implement in hardware. Besides, while attempting to design the AF in hardware, the designs should comply with the modern VLSI design. The two important requirements for contemporary VLSI design are: the designs should be implemented in only MOS technology and should operate at ultra-low voltage. One of the promising circuit design techniques for achieving these goals is companding. In this paper, the low-voltage design of three AFs, Tanh, Unipolar Sigmoidal, and Bipolar Sigmoidal, using Sinh-Domain (SD) technique are introduced. The SD blocks have been implemented using MOS transistors in weak inversion which ensures the low voltage operation of the circuits in addition to that provided by the companding technique itself. The AFs are subsequently used to design the neural networks which have been trained in MATLAB environment to perform AND, OR, NOT, NAND, NOR, and XOR logic functions. To extend the application of trained gates to multilayer neural network, the two XOR gates have been configured to design 3-bit grey to binary code converter which as a result becomes a two-layer network. The performance of the AFs and neural network has been evaluated through simulation results, using the HSPICE software with the MOS transistor models provided by the 0.35 μm AMS CMOS process.
机译：激活功能（AF）是人工神经网络（ANN）的心脏。它是人工神经网络的典型构建模块，很难在硬件中实现。此外，在尝试以硬件设计自动对焦时，设计应符合现代VLSI设计。当代VLSI设计的两个重要要求是：设计应仅以MOS技术实现，并且应在超低电压下运行。实现这些目标的有希望的电路设计技术之一是压扩。本文介绍了使用Sinh-Domain（SD）技术对Tanh，单极S型和双极S型三个AF进行的低压设计。 SD模块已使用MOS晶体管以弱反转的方式实现，除了压扩技术本身提供的功能外，还确保了电路的低压操作。 AF随后用于设计在MATLAB环境中经过训练的神经网络，以执行AND，OR，NOT，NAND，NOR和XOR逻辑功能。为了将经过训练的门的应用扩展到多层神经网络，这两个XOR门已配置为设计3位灰度到二进制代码转换器，结果成为两层网络。使用HSPICE软件和0.35μmAMS CMOS工艺提供的MOS晶体管模型，通过仿真结果评估了AF和神经网络的性能。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号