CMOS Image Sensor Spectral Filters for AI Feature Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional CMOS image sensors sacrifice light sensitivity for color imagery due to red, green, and blue color filters, leading to reduced light reception and increased blurring in low-light conditions, which is inadequate for modern applications requiring rich spectral diversity for artificial intelligence and machine learning.
Innovation Solution
A CMOS imaging sensor design that employs a pixel cell with spectral filters configured to filter light based on unique transmission functions, allowing for higher photo-electron generation efficiency and spectral diversity, enabling the creation of information-rich feature vectors for AI and ML applications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If red, green, and blue color filters are applied to imaging sensor pixels, then color imagery is achieved, but light sensitivity is reduced
Solution Approach 1:
The patent removes traditional RGB color filters from the imaging sensor pixels, extracting the color filtering function entirely. Instead of using physical filters to achieve color imagery, the system captures full-spectrum light data and uses computational algorithms to synthesize color information post-capture, thereby maintaining maximum light sensitivity while still producing color images.
Solution Approach 2:
The patent replaces the mechanical/optical filtering system (physical color filters) with a computational/software-based system. Color information is generated through digital processing and algorithms rather than physical light blocking, substituting a mechanical approach with an informational/computational one that preserves light sensitivity.
2Loss of information
If red, green, and blue color filters are applied to imaging sensor pixels, then color imagery is achieved, but image blurring increases
Solution Approach 1:
By removing the physical color filters that cause optical blurring, the patent eliminates the source of image degradation. The extraction of the filtering function allows light to reach the sensor without being scattered or blocked by filter materials, preserving fine detail and reducing blurring while color information is reconstructed computationally.
Solution Approach 2:
The patent substitutes the mechanical filtering approach with a computational color synthesis approach. Instead of physically filtering light (which causes blurring), the system uses digital processing to generate color information, replacing an optical-mechanical process with an informational process that preserves image sharpness.
3Loss of information
If spectral bandpass filtering is used in imaging sensors, then color photography is achieved, but light reception is reduced
Solution Approach 1:
The patent extracts and removes spectral bandpass filtering from the optical path entirely. Instead of using filters to create color channels, the system captures the complete spectrum and uses computational methods to derive color information, taking out the filtering step that would otherwise block portions of the light spectrum.
Solution Approach 2:
The patent replaces optical spectral filtering with computational spectral analysis. Rather than using physical filters to separate wavelengths (which reduces light quantity), the system uses digital processing to analyze and reconstruct spectral information from full-spectrum light capture, substituting an optical-mechanical approach with a computational one.
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
The design enhances AI-efficacy by maintaining light sensitivity while providing spectral diversity, allowing for improved image quality and data richness suitable for AI and ML applications, even in low-light conditions.
Implementation Method 1
spectral filters that are each configured to filter incoming light based on a transmission function
Implementation Method 2
A complementary MOS (CMOS) sensor is sensitive to (e.g., can detect) the entire visible light spectrum with relatively high efficiency
Data Source
AI summary
An imaging device capable of producing images or data with relatively high spectral diversity, allowing for creation of information-rich feature vectors, is provided. Among other things, such information-rich feature vectors may be applied to a range of artificial intelligence and machine learning applications. The imaging device may include a substrate having a baseline spectral responsivity function, multiple pixels forming a cell fabricated on the substrate, and spectral filters each configured to filter light based on a transmission function corresponding to a substantially broad portion of the baseline spectral responsivity function. The spectral filters may be notch filters. Each of the multiple pixels in the cell may be configured to receive light through each of the spectral filters. The transmission function of each of the spectral filters may be substantially different for each of at least a majority of the multiple pixels in the cell.