A 1.15-TOPS 6.57-TOPS/W Neural Network Processor for Multi-Scale Object Detection With Reduced Convolutional Operations
- 13 January 2020
- journal article
- research article
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Signal Processing
- Vol. 14 (4) , 634-645
Abstract
For automated driving cars, we present a 40-nm dedicated object detection processor with only three operations: 3 × 3 convolution, 1 × 1 convolution, and 4 × 4 deconvolution. Multi-scale object detection at high recognition accuracy is possible by virtue of the deconvolution feature and concatenation. The input memory for a feature map has 8-bit width. A multiplier for the inputs has 8-bit precision. Partial-sum memory, however, has 16-bit width to suppress detection accuracy deterioration in a layer with 1024 channels in the target network. By fixed-point bit precision, the external memory bandwidth and internal memory capacity are reduced. Optimized parallelization in input and output channels reduces the external memory bandwidth to 0.062 billion accesses per 1280 × 384 image with internal memory capacity of 400 kB. The detection error is 1.9% of that using single-precision floating point. The maximum operating frequency is 500 MHz at a supply voltage of 1 V. Its peak performance is 1.15 TOPS. The maximum energy efficiency is 6.57 TOPS/W at 174 MHz and 0.6 V.Keywords
Funding Information
- Toyota Motor Corporation
This publication has 20 references indexed in Scilit:
- 14.5 Envision: A 0.26-to-10TOPS/W subword-parallel dynamic-voltage-accuracy-frequency-scalable Convolutional Neural Network processor in 28nm FDSOI2017
- 14.2 DNPU: An 8.1TOPS/W reconfigurable CNN-RNN processor for general-purpose deep neural networks2017
- EyerissACM SIGARCH Computer Architecture News, 2016
- EIEACM SIGARCH Computer Architecture News, 2016
- Faster R-CNN: Towards Real-Time Object Detection with Region Proposal NetworksIEEE Transactions on Pattern Analysis and Machine Intelligence, 2016
- You Only Look Once: Unified, Real-Time Object Detection2016
- Fast R-CNN2015
- ShiDianNao2015
- Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation2014
- Selective Search for Object RecognitionInternational Journal of Computer Vision, 2013