Skip to content

Latest commit

 

History

History
48 lines (36 loc) · 2.18 KB

RedEye Analog ConvNet Image Sensor Architecture for Continuous Mobile Vision.md

File metadata and controls

48 lines (36 loc) · 2.18 KB

Paper title:

RedEye: Analog ConvNet Image Sensor Architecture for Continuous Mobile Vision

Publication:

ISCA’2016

Problem to solve:

Continuous mobile vision has recently attracted attention, but it faces a daunting barrier: energy efficiency. This is largely due to the energy burden of analog readout circuit, data traffic, and intensive computation. The analog readout bottleneck need to be solved urgently.

Major contribution:

  1. The paper presents an analog convolutional image sensor called RedEye that performs layers of a convolutional neural network in the analog domain before quantization. It aims to improve energy efficiency in continuous mobile vision. The paper gives both the hardware architecture and analog circuit design of RedEye. RedEye mitigates analog design complexity, using a modular column-parallel design to promote physical design reuse and algorithmic cyclic reuse. RedEye uses programmable mechanisms to admit noise for tunable energy reduction.

  2. The paper uses a simulation-based framework to estimate task accuracy and energy consumption. This allows developers to optimize their RedEye programs and noise parameters for minimal energy consumption at sufficient task accuracy.

  3. Compared to conventional systems, RedEye reports an 85% reduction in sensor energy, 73% reduction in cloudlet-based system energy, and a 45% reduction in computation-based system energy. RedEye advances towards overcoming the energy-efficiency barrier to continuous mobile vision. While they design and simulate RedEye energy, noise, and timing performance, they do not yet provide a circuit layout of the RedEye architecture.

Lessons learnt:

  1. To address analog readout bottleneck, the key idea of the paper is to push early processing into the analog domain to reduce the workload of the analog readout.

  2. RedEye positions a convergence of analog circuit design, systems architecture, and machine learning, which allows it to perform early operations in the image sensor’s analog domain, moving toward continuous mobile vision at ultra low power.