Image Sensor Memory Sharing for Neural Network Computing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural network computing devices face challenges in efficiently processing large amounts of data due to memory limitations, leading to increased chip size and limited computing performance.
Innovation Solution
Integration of a neural network computing device with an image sensor device that shares a memory, allowing the image sensor device to operate in a second mode to store neural network model-related data, thereby increasing available memory resources for the neural network computing device.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If memory size is increased to facilitate processing large amounts of data, then computing performance is improved, but chip size increases and implementation becomes difficult
Solution Approach 1:
The patent merges the memory resources of the image sensor device with the neural network computing device. The image sensor device includes a memory that can store both pixel values and neural network model-related data, allowing the two devices to share memory capacity rather than each having separate dedicated memory, thus avoiding chip size increase while improving computing performance
Solution Approach 2:
The memory in the image sensor device is designed with multi-functionality to serve dual purposes: storing pixel values during normal image sensing operations and storing neural network model-related data during computing operations. This universal memory resource allows the system to expand available memory capacity without adding separate dedicated memory components
2Quantity of substance
If memory size is increased to store more data, then data processing capability is improved, but device complexity increases
Solution Approach 1:
The image sensor device's memory is designed to perform multiple functions - storing pixel values during imaging operations and storing neural network model data during computing operations. This multi-functional memory approach increases data capacity without requiring separate dedicated memory subsystems, thereby avoiding increased device complexity
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 integration prevents the need for larger chip sizes and enhances neural network computing performance by increasing the amount of data that can be processed, improving overall device performance.
Implementation Method 1
a pixel array configured to receive optical signals and convert the received optical signals into electrical signals
Implementation Method 2
an analog-to-digital conversion circuit configured to convert a pixel signal received from the pixel array into a digital signal including a pixel value
Data Source
AI summary
An electronic device, including: a neural network computing device configured to perform a computing operation corresponding to a neural network model; and an image sensor device including: a pixel array configured to receive optical signals and convert the received optical signals into electrical signals; and a memory, wherein based on a mode of the image sensor device being a first mode, the memory is configured to store a pixel value, and wherein based on the mode of the image sensor device being a second mode, the memory is further configured to store neural network model-related data