Stacked Image Sensor AI Layout for Heat-Aware Parallel Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing photoelectric conversion apparatuses face issues with high power consumption and heat generation due to signal processing based on neural network calculation models, leading to noise in signal outputs and decreased image quality, as well as limited processing speed.
Innovation Solution
The apparatus incorporates a layered structure with AI processing units arranged to disperse heat and perform parallel processing, using multiple AD conversion circuits and AI processing units to reduce heat impact on pixel arrays and enhance processing speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If signal processing based on neural network calculation models is performed, then processing capability is improved, but heat generation increases causing noise and decreased image quality
Solution Approach 1:
The patent divides the second substrate into multiple regions, each containing different types of signal processing circuits (e.g., ADC circuits, machine learning processing circuits). This segmentation distributes heat generation across different areas rather than concentrating it in one location, reducing thermal impact on the pixel array while maintaining processing capability.
Solution Approach 2:
The patent utilizes the spatial dimension by arranging signal processing circuits in a layered structure on the second substrate, positioned at different distances from the pixel array. This dimensional arrangement allows heat to dissipate in multiple directions and reduces the thermal load on sensitive pixels while preserving processing functions.
2Speed
If multiple AD conversion circuits and signal processing units are added, then processing speed is improved, but device complexity increases
Solution Approach 1:
The patent designs signal processing circuits that can perform multiple functions: analog-to-digital conversion, machine learning processing, and general signal processing. This multi-functionality allows a single circuit design to handle various processing tasks, increasing processing speed without proportionally increasing device complexity.
Solution Approach 2:
The patent combines multiple processing functions (ADC conversion and machine learning processing) into an integrated layered structure on the second substrate. By merging these functions into a unified architecture rather than separate components, the system achieves faster processing while controlling overall complexity through integration.
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 configuration effectively reduces heat non-uniformity and noise in image data, improving image quality and increasing the speed of machine learning processing.
Implementation Method 1
a photodiode D1 serving as a photoelectric conversion portion
Data Source
AI summary
A photoelectric conversion apparatus includes a first substrate including a pixel array including a plurality of pixels, a second substrate layered on the first substrate and including an AD conversion portion including a plurality of AD conversion circuits configured to convert a signal output from the first substrate into a digital signal, wherein the second substrate further includes a plurality of signal processing units including a first signal processing unit and a second signal processing unit both configured to perform machine learning processing, wherein each of a plurality of sets includes a plurality of AD conversion circuits that differ between the plurality of sets, wherein the first signal processing unit is arranged to correspond to one of the plurality of sets, and wherein the second signal processing unit is arranged to correspond to another one of the plurality of sets.