A Low-Voltage, Low-Power Reconfigurable Current-Mode Softmax Circuit for Analog Neural Networks

Electronics

Massimo Vatalaro¹  Marco Lanuzza¹  Tatiana Moposita¹  Felice Crupi¹  Sebastiano Strangio²  Lionel Trojman³  Andrei Vladimirescu³ 

[1] Dipartimento di Ingegneria Informatica, Modellistica, Elettronica e Sistemistica, Università della Calabria, 87036 Rende, Italy;Dipartimento di Ingegneria dell’Informazione, Università di Pisa, Via G. Caruso 16, 56122 Pisa, Italy;Institut Supérieur d’Électronique de Paris, 10 rue de Vanves, 92130 Issy les Moulineaux, France;

关键词: softmax;    activation functions;    deep neural networks;    machine learning;   

DOI  :  10.3390/electronics10091004

来源: DOAJ

【 摘 要 】

This paper presents a novel low-power low-voltage analog implementation of the softmax function, with electrically adjustable amplitude and slope parameters. We propose a modular design, which can be scaled by the number of inputs (and of corresponding outputs). It is composed of input current–voltage linear converter stages (1st stages), MOSFETs operating in a subthreshold regime implementing the exponential functions (2nd stages), and analog divider stages (3rd stages). Each stage is only composed of p-type MOSFET transistors. Designed in a 0.18 µm CMOS technology (TSMC), the proposed softmax circuit can be operated at a supply voltage of 500 mV. A ten-input/ten-output realization occupies a chip area of 2570 µm² and consumes only 3 µW of power, representing a very compact and energy-efficient option compared to the corresponding digital implementations.

【 授权许可】

Unknown