Aida: Associative DNN Inference Accelerator

Leonid Yavits; Roman Kaplan; Ran Ginosar

Aida: Associative DNN Inference Accelerator

Leonid Yavits
, Roman Kaplan
, Ran Ginosar

Research output: Working paper / Preprint › Preprint

48 Downloads (Pure)

Abstract

We propose AIDA, an inference engine for accelerating fully-connected (FC) layers of Deep Neural Network (DNN). AIDA is an associative in-memory processor, where the bulk of data never leaves the confines of the memory arrays, and processing is performed in-situ. AIDA area and energy efficiency strongly benefit from sparsity and lower arithmetic precision. We show that AIDA outperforms the state of art inference accelerator, EIE, by 14.5x (peak performance) and 2.5x (throughput).

Original language	Undefined/Unknown
State	Published - 20 Dec 2018

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

SDG 7 Affordable and Clean Energy

Keywords

cs.DC

Access to Document

1901.04976v1

Cite this

@techreport{1d27ceb2312c4c6d9de36546d5c7d0d6,

title = "Aida: Associative DNN Inference Accelerator",

abstract = " We propose AIDA, an inference engine for accelerating fully-connected (FC) layers of Deep Neural Network (DNN). AIDA is an associative in-memory processor, where the bulk of data never leaves the confines of the memory arrays, and processing is performed in-situ. AIDA area and energy efficiency strongly benefit from sparsity and lower arithmetic precision. We show that AIDA outperforms the state of art inference accelerator, EIE, by 14.5x (peak performance) and 2.5x (throughput). ",

keywords = "cs.DC",

author = "Leonid Yavits and Roman Kaplan and Ran Ginosar",

year = "2018",

month = dec,

day = "20",

language = "???core.languages.und???",

type = "WorkingPaper",

}

TY - UNPB

T1 - Aida

T2 - Associative DNN Inference Accelerator

AU - Yavits, Leonid

AU - Kaplan, Roman

AU - Ginosar, Ran

PY - 2018/12/20

Y1 - 2018/12/20

N2 - We propose AIDA, an inference engine for accelerating fully-connected (FC) layers of Deep Neural Network (DNN). AIDA is an associative in-memory processor, where the bulk of data never leaves the confines of the memory arrays, and processing is performed in-situ. AIDA area and energy efficiency strongly benefit from sparsity and lower arithmetic precision. We show that AIDA outperforms the state of art inference accelerator, EIE, by 14.5x (peak performance) and 2.5x (throughput).

AB - We propose AIDA, an inference engine for accelerating fully-connected (FC) layers of Deep Neural Network (DNN). AIDA is an associative in-memory processor, where the bulk of data never leaves the confines of the memory arrays, and processing is performed in-situ. AIDA area and energy efficiency strongly benefit from sparsity and lower arithmetic precision. We show that AIDA outperforms the state of art inference accelerator, EIE, by 14.5x (peak performance) and 2.5x (throughput).

KW - cs.DC

M3 - פרסום מוקדם

BT - Aida

ER -