Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Eyeriss: A Spatial Architecture for Energy-Efficient Dataflow for Convolutional Neural Networks #23

Open
meton-robean opened this issue Jan 14, 2020 · 2 comments

Comments

@meton-robean
Copy link
Owner

参考:
能效优先:eyeriss CNN加速器的设计思路

@meton-robean
Copy link
Owner Author

论文贡献:

  • 针对CNN,提出了新型的数据流模式,row-stationary, 挖掘数据重用的潜力,最小化了数据移动过程中的能量消耗

  • 按照数据移动的能耗,将加速器的存储分成四个层级,off chip-DRAM, global buffer, i/o FIFO, RF,越常被重用的数据,存储的层级越低,访问的开销越小。

image

  • 传统CNN加速器架构只针对权重或者累加和进行数据的重用。而eyeriss可以通过配置,实现不同存储层次上的权重和累加数据的重用,避免了反复写回片外存储带来的巨大开销
    image

  • 加速器上有两个时钟,一个是global buffer访问内存的时钟,一个是PE做计算的核心时钟。区分开来的好处是可以异步访问内存,方便预取

  • 每个PE上都有一个状态机,当操作数就绪时,每个PE可以立即执行运算操作,而无需等待所有PE就绪。

@meton-robean
Copy link
Owner Author

思考:

  • Eyeriss 提出的四层存储结构很好地降低了数据移动的开销
  • 每个PE上的状态机也避免了PE数量过多,导致等待其他PE就绪时间过长
  • 论文通篇围绕 提高数据重用 而设计,目的性强,没有冗余结构

Repository owner locked and limited conversation to collaborators Jan 14, 2020
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Projects
None yet
Development

No branches or pull requests

1 participant