Multiple approximate instances in neural processing units for energy-efficient circuit synthesis: work-in-progress

Tanfer Alan, Jorge Castro-Godínez, Jörg Henkel

Producción científica: Capítulo del libro/informe/acta de congresoContribución a la conferenciarevisión exhaustiva

Resumen

We present an architectural approach toward energy-efficient synthesis of circuits used in neural processing units. Neural network applications are shown to tolerate varying operand precisions between different inputs, accuracy targets, their phases, and learning methods, without significantly impacting the classification accuracy. Using multiple instances of systolic arrays at different precisions, we show that significant energy gains are possible beyond the conventional approach, using the same circuit for all precisions.

Idioma originalInglés
Título de la publicación alojadaProceedings - 2021 International Conference on Compilers, Architectures, and Synthesis for Embedded Systems, CASES 2021
EditorialAssociation for Computing Machinery, Inc
Páginas3-5
Número de páginas3
ISBN (versión digital)9781450383783
DOI
EstadoPublicada - 30 sept 2021
Evento2021 International Conference on Compilers, Architectures, and Synthesis for Embedded Systems, CASES 2021 - Virtual, Online, Estados Unidos
Duración: 8 oct 202115 oct 2021

Serie de la publicación

NombreProceedings - 2021 International Conference on Compilers, Architectures, and Synthesis for Embedded Systems, CASES 2021

Conferencia

Conferencia2021 International Conference on Compilers, Architectures, and Synthesis for Embedded Systems, CASES 2021
País/TerritorioEstados Unidos
CiudadVirtual, Online
Período8/10/2115/10/21

Huella

Profundice en los temas de investigación de 'Multiple approximate instances in neural processing units for energy-efficient circuit synthesis: work-in-progress'. En conjunto forman una huella única.

Citar esto