Display AI Convolution With Dynamic Bit-Width Shifting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The increase in hardware size and power consumption due to the use of artificial intelligence in display devices for processing 8K resolution content, which requires large multipliers and memory, leads to higher manufacturing costs.
Innovation Solution
Implementing a lightweight convolutional neural network with reduced bit width weights and performing dynamic shift operations to reduce bit width during neural network computations, thereby reducing hardware requirements and power consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If artificial intelligence models are used to process 8K resolution content, then image quality is improved, but hardware size increases
Solution Approach 1:
The patent applies parameter changes by reducing the bit width of weights in the convolutional neural network from conventional precision (e.g., 32-bit) to lower precision (e.g., 8-bit or 4-bit). This parameter modification allows the AI model to maintain sufficient image processing capability while significantly reducing the memory and register requirements, thereby decreasing hardware size without substantially compromising image quality
Solution Approach 2:
The patent implements partial action by using reduced precision weights that provide sufficient processing capability for the intended application. Instead of using full precision weights, the system uses just enough precision (reduced bit width) to achieve acceptable image quality, thereby reducing hardware requirements while maintaining adequate performance
2Measurement precision
If artificial intelligence models are used to process 8K resolution content, then image quality is improved, but power consumption increases
Solution Approach 1:
The patent reduces power consumption by changing the precision parameter of weights to a lower bit width. This reduction decreases the computational complexity and the amount of data that needs to be processed and transferred, thereby reducing power consumption while maintaining adequate image quality through the use of reduced precision arithmetic operations
3Measurement precision
If conventional precision weights are used in convolutional neural networks, then processing accuracy is maintained, but hardware resources increase
Solution Approach 1:
The patent changes the precision parameter of weights from conventional high precision to reduced precision (lower bit width). This parameter change reduces the quantity of hardware resources required, including memory capacity and register size, while maintaining sufficient processing accuracy for image and video processing applications through carefully designed reduced precision weight representations
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method of performing a convolution operation is provided. The method includes obtaining a lightweight convolutional neural network with a reduced bit width of weights, and inputting input data to the convolutional neural network and performing neural network computations to obtain output data. The neural network computations includes determining a shift distance based on a value of input feature data, performing a shift operation to reduce a bit width of the input feature data based on the shift distance, performing a convolution operation on the input feature data with the reduced bit width and the weights with the reduced bit width, wherein the convolution operation includes a shift operation of restoring the bit width, and obtaining output feature data with the restored bit width.