Neural Network Computing Chip Translation Circuit Non-Negative Data Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current neural network computing chips face inefficiencies due to the need to split input data into positive and negative components for processing, resulting in increased delay and resource consumption, as they require separate calculations for each component and subsequent aggregation.
Innovation Solution
A neural network computing chip with a translation circuit that processes input data to convert negative values into zero or positive values, allowing for a single calculation step using a computing circuit, thereby eliminating the need for dual operations and reducing resource consumption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If input data is split into positive and negative components for separate calculation, then the computing circuit can process data within its value constraints, but the processing delay increases and computing efficiency decreases
Solution Approach 1:
The translation circuit performs preliminary action by converting negative values in input data to positive values before the data reaches the computing circuit. This advance transformation ensures that the computing circuit receives only non-negative values, eliminating the need for separate positive and negative data processing paths and reducing overall processing delay.
Solution Approach 2:
The translation circuit acts as an intermediary between the input data and the computing circuit. It mediates the data format by transforming negative values to positive values, allowing the computing circuit to operate within its constraints without requiring complex dual-path processing architecture.
2Measurement precision
If separate calculations are performed for positive and negative input data, then accurate results can be obtained, but resource consumption increases
Solution Approach 1:
The translation circuit merges the processing of positive and negative values into a single transformation step. By converting all negative values to positive values upfront, the system combines what would otherwise require separate calculation paths into one unified processing flow, reducing computing resource consumption while maintaining accuracy through subsequent bias adjustment.
3Device complexity
If the computing circuit processes only non-negative values, then the hardware design is simplified, but the input data must undergo additional transformation
Solution Approach 1:
The system segments the processing function into two specialized components: the translation circuit handles value transformation (converting negatives to positives), while the computing circuit handles the multiply-accumulate operation. This segmentation allows each component to be optimized for its specific function, simplifying the computing circuit design while maintaining overall processing efficiency.
Data Source
AI summary
A neural network computing chip includes a translation circuit and a computing circuit. After input data is processed by the translation circuit, a value of each element of the data input into the computing circuit is not a negative number, thereby meeting a value limitation condition imposed by the computing circuit on the input data.