This book discusses in detail low-voltage low-power designs for minimizing the hardware resources required by neural network implementations. The novel method presented in this book for an accurate realization of activation functions for artificial neural networks (ANNs), is based on specific superior-order approximation functions. The author describes analog implementations in CMOS technology to increase the speed of operation, while reducing the hardware resources required for obtaining these approximation functions. Original architectures presented in this book, used for implementing previous CMOS computational structures, allow for operation independent of technological errors and temperature variations. SPICE simulations confirm the theoretically estimated results for previously presented CMOS computational structures, developed for ANNs and artificial intelligence applications.

Describes low-voltage low-power designs for minimizing hardware resources required by neural network implementations Presents an analog synthesis of activation functions with many applications in artificial neural networks Analyzes high-accuracy CMOS computational circuits for implementing activation functions for artificial neural networks

Autorentext

Cosmin Radu Popa has appointed during his teaching activity, in the following positions (obtained from national contests): assistant professor (1997-2004), lecturer (2004-2012), associate professor (2012-2014) and professor (2014-to date) at the Faculty of Electronics, Telecommunications and Information Technology Bucharest from National University of Science and Technology Politehnica Bucharest, Romania. He is the course titular for the following disciplines: Analog Integrated Circuits, Circuits Intégrés, Electronic Devices and Circuits, Design Techniques for Analog VLSI Structures, Computer Aided Design of Microelectronic Circuits and Fundamentals of Electronics from the academic year 2003/2004 to date. His courses were taught in Romanian language, also in English and French languages, Cosmin Radu Popa teaching in his career many multinational student teams. Cosmin Radu Popa has a research experience of over 28 years and a full time postdoctoral research experience of over 20 years, having an independent scientific and research activity of large international visibility. The most important results of this research activity has been focused in the publication as single author of three monographs at SPRINGER Publishing House (Superior-Order Curvature-Correction Techniques for Voltage References (2009 year), Synthesis of Computational Structures for Analog Signal Processing (2011 year) and Current-Mode Analog Nonlinear Function Synthesizer Structures (2013 year), also of other 5 books in the field of analysis and design of VLSI CMOS analog integrated circuits, as well as in the coordination of 7 research projects obtained in CEEX, IDEAS, CNCSIS and POSDRU competitions, or in participation as a member of the research team in other 14 research projects.

Inhalt

Introduction.- Superior-order approximation functions for generating sigmoidal activation functions.- Superior-order approximation functions for generating radial basis activation functions.- Superior-order approximation functions for artificial neural networks applications.- Analysis and design of analog function synthesizers for implmenting sigmoidal activation functions.- Analysis and design of analog function synthesizers for generating radial basis activation functions.- Analysis and design of analog function synthesizers for artificial neural networks applications.- Low-voltage low-power current-mode CMOS computational circuits for implementing activation functions.- Conclusions.