Scipediacontent: Scipediacontent moved page Draft Content 203738670 to Yang et al 2018a

2021-02-01T21:51:13Z

Scipediacontent moved page Draft Content 203738670 to Yang et al 2018a

Scipediacontent: Created page with " == Abstract == hough ReRAM-based convolutional neural network (CNN) accelerators have been widely studied, state-of-the-art solutions suffer from either incapability of trai..."

2021-02-01T21:51:07Z

Created page with " == Abstract == hough ReRAM-based convolutional neural network (CNN) accelerators have been widely studied, state-of-the-art solutions suffer from either incapability of trai..."

New page

== Abstract ==

hough ReRAM-based convolutional neural network (CNN) accelerators have been widely studied, state-of-the-art solutions suffer from either incapability of training (e.g., ISSAC [1]) or inefficiency of inference (e.g., PipeLayer [2]) due to the pipeline design. In this work, we propose AtomLayer---a universal ReRAM-based accelerator to support both efficient CNN training and inference. AtomLayer uses the atomic layer computation which processes only one network layer each time to eliminate the pipeline related issues such as long latency, pipeline bubbles and large on-chip buffer overhead. For further optimization, we use a unique filter mapping and a data reuse system to minimize the cost of layer switching and DRAM access. Our experimental results show that AtomLayer can achieve higher power efficiency than ISSAC in inference (1.1×) and PipeLayer in training (1.6×), respectively, meanwhile reducing the footprint by 15×.

== Original document ==

The different versions of the original document can be found in:

* [http://dl.acm.org/ft_gateway.cfm?id=3195998&type=pdf http://dl.acm.org/ft_gateway.cfm?id=3195998&type=pdf]

* [https://dblp.uni-trier.de/db/conf/dac/dac2018.html#QiaoCYSL18 https://dblp.uni-trier.de/db/conf/dac/dac2018.html#QiaoCYSL18],
: [https://ieeexplore.ieee.org/document/8465832 https://ieeexplore.ieee.org/document/8465832],
: [https://doi.acm.org/10.1145/3195970.3195998 https://doi.acm.org/10.1145/3195970.3195998],
: [https://doi.org/10.1145/3195970.3195998 https://doi.org/10.1145/3195970.3195998],
: [https://dl.acm.org/citation.cfm?id=3195998 https://dl.acm.org/citation.cfm?id=3195998],
: [https://academic.microsoft.com/#/detail/2809171749 https://academic.microsoft.com/#/detail/2809171749]

* [https://dl.acm.org/doi/pdf/10.1145/3195970.3195998 https://dl.acm.org/doi/pdf/10.1145/3195970.3195998],
: [http://dx.doi.org/10.1145/3195970.3195998 http://dx.doi.org/10.1145/3195970.3195998] under the license http://www.acm.org/publications/policies/copyright_policy#Background

← Older revision	Revision as of 21:51, 1 February 2021
(No difference)

Yang et al 2018a - Revision history

Scipediacontent: Scipediacontent moved page Draft Content 203738670 to Yang et al 2018a

Scipediacontent: Created page with " == Abstract == hough ReRAM-based convolutional neural network (CNN) accelerators have been widely studied, state-of-the-art solutions suffer from either incapability of trai..."