Exploiting Parallelism for CNN Applications on 3D Stacked Processing-In-Memory Architecture
- 31 August 2018
- journal article
- research article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Parallel and Distributed Systems
- Vol. 30 (3) , 589-600
Abstract
Deep convolutional neural networks (CNNs) are widely adopted in intelligent systems with unprecedented accuracy but at the cost of a substantial amount of data movement. Although the emerging processing-in-memory (PIM) architecture seeks to minimize data movement by placing memory near processing elements, memory is still the major bottleneck in the entire system. The selection of hyper-parameters in the training of CNN applications requires over hundreds of kilobytes cache capacity for concurrent processing of convolutions. How to jointly explore the computation capability of the PIM architecture and the highly parallel property of neural networks remains a critical issue. This paper presents Para-Net, that exploits Parallelism for deterministic convolutional neural Networks on the PIM architecture. Para- Net achieves data-level parallelism for convolutions by fully utilizing the on-chip processing engine (PE) in PIM. The objective is to capture the characteristics of neural networks and present a hardware-independent design to jointly optimize the scheduling of both intermediate results and computation tasks. We formulate this data allocation problem as a dynamic programming model and obtain an optimal solution. To demonstrate the viability of the proposed Para-Net, we conduct a set of experiments using a variety of realistic CNN applications. The graph abstractions are obtained from deep learning framework Caffe. Experimental results show that Para-Net can significantly reduce processing time and improve cache efficiency compared to representative schemes.Keywords
Funding Information
- National Natural Science Foundation of China (61502309, 61702357)
- Guangdong Natural Science Foundation (2016A030313045, 2017B030314073)
- Shenzhen Science and Technology Foundation (JCYJ20150529164656096, JCYJ20170302153955969, JCYJ20170817100300603, JCYJ20170818100006280, JCYJ20170816093943197)
- Guangdong Provincial General University National Development Program (2014GKXM054)
- Natural Science Foundation of SZU (803/000026080154, 827-000073)
- Natural Science Foundation of Tianjin Municipality (18JCQNJC00300)
- Institute of Computing Technology, Chinese Academy of Sciences (CARCH201608)
This publication has 35 references indexed in Scilit:
- Simultaneous Multikernel GPU: Multi-tasking throughput processors via fine-grained sharing2016
- Practical Near-Data Processing for In-Memory Analytics Frameworks2015
- Data reorganization in memory using 3D-stacked DRAM2015
- Going deeper with convolutions2015
- Optimizing FPGA-based Accelerator Design for Deep Convolutional Neural Networks2015
- Caffe2014
- Memory access optimized routing scheme for deep networks on a mobile coprocessor2014
- DianNaoACM SIGPLAN Notices, 2014
- Memory-Aware Task Scheduling with Communication Overhead Minimization for Streaming Applications on Bus-Based Multiprocessor System-on-ChipsIEEE Transactions on Parallel and Distributed Systems, 2013
- Synchronous circuit optimization via multidimensional retimingIEEE Transactions on Circuits and Systems II: Analog and Digital Signal Processing, 1996