Aicnn Implementing Typical Cnn Algorithms With Analogtoinformation Conversion Architecture

• doi: 10.1109/ISVLSI.2017.23
• title: AICNN: Implementing Typical CNN Algorithms with Analog-to-Information Conversion Architecture
• publisher: IEEE
• isbn: 978-1-5090-6763-3
• issn: 2159-3477
• rank: 1358
• access_type: LOCKED
• content_type: Conferences
• abstract: AICNN architecture is presented in this work to map the state-of-the-art machine-learning algorithms of CNN to power-constrained embedded hardware. As the combination of analog-to-information conversion and typical CNN algorithms, AICNN can realize ultra-highly efficient computation by using massive parallel analog signal processing circuits, which could also significantly reduce ADC devices cost of converting sensors' outputs. As a design example, the specific AICNN-3 implementation is evaluated, which realize the minimum system of typical CNN task using AICNN architecture, with SMIC 0.18 μm CMOS process. Simulation results show that the AICNN-3 can classify a 28x28 MNIST image with only 1.47nJ. Compared with baseline implementation on CPU, the AICNN-3 could achieve 67000x energy-efficiency improvement, however the accuracy loss is less than 1%. Moreover, the influences of devices mismatch and process variations are evaluated using Monte Carlo statistical method, for the imperfection of analog processing paradigm, as well as the scalability of AICNN architecture is also discussed.
• article_number: 7987499
• pdf_url: https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=7987499
• html_url: https://ieeexplore.ieee.org/document/7987499/
• abstract_url: https://ieeexplore.ieee.org/document/7987499/
• publication_title: 2017 IEEE Computer Society Annual Symposium on VLSI (ISVLSI)
• conference_location: Bochum, Germany
• conference_dates: 3-5 July 2017
• publication_number: 7985547
• is_number: 7987474
• publication_year: 2017
• publication_date: 3-5 July 2017
• start_page: 80
• end_page: 85
• citing_paper_count: 8
• citing_patent_count: 0
• download_count: 1025
• insert_date: 20170724
• index_terms:
  • ieee_terms:
    • Convolution
    • Neurons
    • Kernel
    • Very large scale integration
    • Robots
  • author_terms:
    • CNN
    • Analog-to-Information Conversion
    • Analog Signal Processing
    • Energy efficiency
  • dynamic_index_terms:
    • Convolutional Neural Network
    • Types Of Algorithms
    • Convolutional Neural Network Algorithm
    • Typical Convolutional Neural Network
    • Converter Architecture
    • Simulation Results
    • Signal Processing
    • Energy Consumption
    • Energy Expenditure
    • Convolutional Layers
    • Internet Of Things
    • Energy Cost
    • Real-time Performance
    • Real-time Processing
    • Real-time Operation
    • Max-pooling
    • Pooling Layer
    • Fully-connected Layer
    • Neural Network Algorithm
    • Computing Units
    • Calculation Unit
    • Current Difference
    • Current Deviation
    • MNIST Dataset
    • Drain Current
    • Analog Domain
    • Differential Pair
    • Scale-invariant Feature Transform
    • Kirchhoff’s Current Law
    • Kirchhoff’s Circuit Laws
    • Current Mirror
    • Tail Current
    • Analog Circuits
    • Low Power Consumption
    • Internet Of Things Systems
    • Power Consumption
• isbn_formats:
  • format: Print on Demand(PoD) ISBN, value: 978-1-5090-6763-3, isbnType: New-2005
  • format: Electronic ISBN, value: 978-1-5090-6762-6, isbnType: New-2005
• authors:
  • Author Name: Kaige Jia
    Affiliation: Dept. of Electronic Engineering, Tsinghua University, China
    Author URL: https://ieeexplore.ieee.org/author/37085990712
    ID: 37085990712
    Order: 1
    Author Affiliations:
    • Dept. of Electronic Engineering, Tsinghua University, China
  • Author Name: Zheyu Liu
    Affiliation: Dept. of Electronic Engineering, Tsinghua University, China
    Author URL: https://ieeexplore.ieee.org/author/37085778930
    ID: 37085778930
    Order: 2
    Author Affiliations:
    • Dept. of Electronic Engineering, Tsinghua University, China
  • Author Name: Fei Qiao
    Affiliation: Dept. of Electronic Engineering, Tsinghua University, China
    Author URL: https://ieeexplore.ieee.org/author/37272195200
    ID: 37272195200
    Order: 3
    Author Affiliations:
    • Dept. of Electronic Engineering, Tsinghua University, China
  • Author Name: Xinjun Liu
    Affiliation: Dept. of Mechanical Engineering, Tsinghua University, China
    Author URL: https://ieeexplore.ieee.org/author/37085996613
    ID: 37085996613
    Order: 4
    Author Affiliations:
    • Dept. of Mechanical Engineering, Tsinghua University, China
  • Author Name: Qi Wei
    Affiliation: Dept. of Electronic Engineering, Tsinghua University, China
    Author URL: https://ieeexplore.ieee.org/author/37582779500
    ID: 37582779500
    Order: 5
    Author Affiliations:
    • Dept. of Electronic Engineering, Tsinghua University, China
  • Author Name: Huazhong Yang
    Affiliation: Dept. of Electronic Engineering, Tsinghua University, China
    Author URL: https://ieeexplore.ieee.org/author/37291236500
    ID: 37291236500
    Order: 6
    Author Affiliations:
    • Dept. of Electronic Engineering, Tsinghua University, China

Image Sensor

• sensor_type: CMOS
• resolution: 1920x1080
• dynamic_range: 70 dB
• pixel_size: 2 µm
• dark_current: 10 e-/s

Optical Data

• focal_length: 50 mm
• aperture: f/2.8
• field_of_view: 70°
• distortion: 5%

Performance Metrics

• frame_rate: 30 fps
• signal_to_noise_ratio: 40 dB
• sensitivity: ISO 800
• shutter_speed: 1/60s
• power_consumption: 150 mW
• noise: 5 e-

Applications & Benefits

• cell_imaging: Used in medical imaging for cell analysis and diagnostics
• benefits: High sensitivity and resolution, compact size for integration into small devices

Supporting Organizations

• supported_by: IEEE, ACM, University of California

Manuscript Details

• publication_date: 3-5 July 2017

Relevancy Score

• score: 9
• missing_fields:
  • none

Processed JSON Filename

request_11f30707-ae42-4f0e-8f92-33c859beceb2-aicnn_implementing_typical_cnn_algorithms_with_analogtoinformation_conversion_architecture.json