Solid-State Imaging Element Analog Convolution Control
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The convolution operation in CNNs leads to increased processing load and power consumption due to repeated weighting and summing of pixel values, which is undesirable.
Innovation Solution
A solid-state imaging element with a photoelectric conversion element, floating diffusion, and transfer control element, where the transfer control element is controlled by a driving voltage based on convolution coefficients, allowing for controlled transfer of signal charge and reducing processing load through analog processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If digital convolution operations are performed by repeatedly weighting and summing pixel values, then image classification accuracy is maintained, but processing load and power consumption increase
Solution Approach 1:
The patent replaces digital computational operations with optical analog processing. Light intensity directly modulates the amount of signal charge transferred from photoelectric conversion elements to floating diffusions, eliminating the need for digital multiplication and addition operations. This substitution of optical physics for digital computation dramatically reduces power consumption and processing load while maintaining convolution operation functionality.
Solution Approach 2:
The patent changes the operational parameters from digital domain to optical domain. Instead of using digital values and performing computational operations, the system uses light intensity as the controlling parameter to directly regulate signal charge transfer amounts. This parameter transformation enables analog implementation of convolution coefficients through optical means, reducing energy consumption.
2Adaptability or versatility
If signal charge is transferred from each photoelectric conversion element to individual floating diffusions, then convolution weighting is achieved, but device complexity increases
Solution Approach 1:
The patent makes floating diffusions serve multiple functions: they act as both charge storage nodes and convolution operation output nodes. By controlling which photoelectric conversion elements transfer charge to which floating diffusions through the optical control mechanism, the same floating diffusion structure can implement different convolution kernels and weighting patterns without requiring additional dedicated hardware for each operation.
Solution Approach 2:
The patent introduces dynamic control of charge transfer through light intensity modulation. The transfer control elements can dynamically adjust the amount of charge transferred based on real-time light intensity, enabling flexible implementation of different convolution coefficients without changing the physical structure. This dynamic behavior allows the same hardware configuration to adapt to various convolution operations.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach reduces power consumption and processing load by performing convolution operations through analog processing in the imaging element, enabling efficient image classification in CNNs.
Implementation Method 1
a photoelectric conversion element that photoelectrically converts received light into signal charge corresponding to an amount of received light
Data Source
AI summary
Power consumption in realizing a convolutional neural network (CNN) is reduced.A solid-state imaging element according to the present technology includes a photoelectric conversion element that photoelectrically converts received light into signal charge corresponding to the amount of received light, a floating diffusion that holds the signal charge obtained by the photoelectric conversion element, a transfer control element that controls transfer of the signal charge from the photoelectric conversion element to the floating diffusion, and a control unit that controls application of a drive voltage to the transfer control element on the basis of a convolution coefficient in a CNN.