2014：DianNao a small

DianNao a small"/>

2014：DianNao a small

文章目录

• Abstract
• • 下一段
  • 第三段
• 1. Introduction
• • 第二段

• 我是在这个网址上找到这篇文章的！

.1145/2654822.2541967

• 属于

Abstract

• ML
• pervasive in a broad range of domains,
• in a broad range of systems (embedded to data centers)

• At the same time,
  • a small set of ml algorithms (especially Convolutional and Deep Neural Networks, i.e., CNNs and DNNs)
• are proving to be sota across many applications.

• As architectures evolve towards
  • heterogeneous multi-cores
  • composed of a mix of cores and accelerators,
• a ml accelerator can achieve
  • the rare combination of efficiency (due to the small number of target algorithms)
  • and broad application scope.

下一段

• Until now, most ml accelerator designs
  • focused on
  • efficiently implementing the computational part of the algorithms.
• However,recent sota CNNs and DNNs
  • are characterized by their large size.

这么大咋办？

• design an accelerator
  • for large-scale CNNs and DNNs,
• emphasis on the impact of memory on accelerator design, performance and energy.

第三段

• possible to design an accelerator with
  • a high throughput,
  • capable of performing 452 GOP/s (key NN operations such as synaptic weight multiplications and neurons outputs additions)
  • in a small footprint of 3.02 m m 2 mm^2 mm2 and 485 mW;
• compared to a 128-bit 2GHz SIMD processor,
  • 117.87x faster,
  • reduce the total energy by 21.08x.

• The accelerator characteristics are obtained
  • after layout at 65nm.
• Such a high throughput in a small footprint
  • can
  • open up the usage of sota ml algorithms
  • in a broad set of systems and
  • for a broad set of applications.

1. Introduction

• architectures evolve towards
  • heterogeneous multi-cores
  • composed of a mix of cores and accelerators,
• designing accelerators
  • which realize the best possible tradeoff
  • between flexibility and efficiency
  • is becoming a prominent issue.

第二段

• The first question
  • for which category of applications
• one should primarily design accelerators ?

• Together with the architecture towards
  • accelerators,
• a second simultaneous and significant trend
  • in high-performance and embedded applications is developing:

• many of the emerging high-performance
  • and embedded applications,
  • from image/video/audio recognition to automatic translation, business analytics, and all forms of robotics rely on ml techniques.

• This trend even starts to percolate in our community
  • where it turns out that
• half of the benchmarks of PARSEC [2],
  • a suite partly introduced to highlight the emergence of new types of applications,
  • can be implemented using machine-learning algorithms [4].

PARSEC里的一半都可以用ml实现！

• a third and equally remarkable trend in ml
  • where a small number of techniques,
  • based on nn (especially CNN [27] and DNN
    [16]),
    proved in the past few years to be state-of-the-art across a broad range of applications [25].

• a unique opportunity to
  • design accelerators
  • which can realize the best of both worlds:
• significant application scope together with
• high performance and efficiency
• due to the limited number of target algorithms.