Charge-Domain Multiplier for Low-Noise ML Image Sensing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image sensing technologies introduce noise, latency, and energy consumption during the conversion of charge to voltage or current for machine learning applications, degrading image quality and increasing latency in critical applications.
Innovation Solution
Directly couple charge stored in reservoirs to multipliers of a machine learning input layer, utilizing charge domain structures like CCD shift registers and systolic arrays to perform weighted sums without digitization, reducing noise and latency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If charge is converted to voltage or current for machine learning applications, then the data can be processed by conventional digital circuits, but noise is introduced and image quality degrades
Solution Approach 1:
The patent extracts the charge domain processing capability directly into the machine learning input layer, removing the need for charge-to-voltage conversion. By taking out the conversion step that introduces noise, the system maintains charge in its native domain throughout processing, thereby eliminating the harmful noise effect while preserving reliability.
Solution Approach 2:
The patent introduces a charge domain multiplier as an intermediary component that can directly process charge signals without conversion. This mediator enables machine learning operations to be performed on charge domain data, preventing the introduction of noise that would occur during conventional voltage conversion while still allowing data processing.
2Productivity
If charge is converted to voltage or current for processing, then conventional digital circuits can be used, but latency increases in time-critical applications
Solution Approach 1:
The patent removes the charge-to-voltage conversion step from the processing pipeline, extracting this time-consuming operation. By eliminating the conversion stage, the system reduces latency significantly while maintaining the ability to process data at high speeds using charge domain multipliers and accumulators.
Solution Approach 2:
The patent enables continuous charge domain processing without interruption for conversion operations. By maintaining charge in its native domain throughout the machine learning pipeline, the system achieves continuous useful action without the breaks and delays inherent in repeated charge-to-voltage conversions, thereby improving productivity while reducing latency.
3Ease of operation
If charge is converted to voltage or current, then data can be read by conventional circuitry, but energy consumption increases
Solution Approach 1:
The patent extracts the readout function to operate directly in the charge domain, removing the energy-intensive charge-to-voltage conversion step. By taking out the conversion operation, the system maintains ease of operation through direct charge domain readout while significantly reducing power consumption associated with repeated conversions.
Solution Approach 2:
The patent enables the charge domain multiplier and accumulator to serve themselves by operating natively on charge signals. This self-service capability eliminates the need for external voltage conversion circuitry, reducing energy consumption while maintaining readability through direct charge domain interfaces that can be read by conventional circuitry when needed.
4Adaptability or versatility
If voltage or current conversion is performed for each machine learning input, then data can be processed, but the complexity of the system increases
Solution Approach 1:
The patent extracts the conversion circuitry from the system entirely by enabling direct charge domain processing. By taking out the voltage and current conversion components, the system maintains adaptability for machine learning processing while reducing device complexity through the elimination of multiple conversion stages and associated circuitry.
Solution Approach 2:
The patent creates a universal charge domain multiplier that can handle all machine learning input operations directly in the charge domain. This multi-functional component replaces multiple specialized conversion circuits, achieving versatility in machine learning processing while reducing overall system complexity through consolidation and direct charge domain operation.
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 maintains image fidelity and reduces power consumption by eliminating the need for voltage or current conversion, enhancing performance in time-critical applications.
Implementation Method 1
The buried pinned photodiode 10 may integrate electrons created when light is collected by the buried pinned photodiode 10 into a storage well SW region
Implementation Method 2
The transfer gate TG is lowered and raised in conformance with required electron flow. At some point the transfer gate TG lowers the potential barrier and the electrons spill from the storage well SW charge reservoir into the floating diffusion FD charge reservoir
Data Source
Figure 1
Figure 2A~2B
Figure 3
AI summary
A multiplier has a pair of charge reservoirs. The pair of charge reservoirs are connected in series, A first charge movement device induces charge movement to or from the pair of charge reservoirs at a same rate. A second charge movement device induces charge movement to or from one of the pair of reservoirs, the rate of charge movement programmed to one of add or remove charges at a rate proportional to the first charge movement device. The first charge movement device loads a first charge into a first of the pair of charge reservoirs daring a first cycle, The first charge movement device and the second charge movement device remove charges at a proportional rate from the pair of charge reservoirs during a second cycle until the first of the pair of charge reservoirs is depleted of the first charge. The second charge reservoir thereafter holding the multiplied result.