Matrix Multiplier MAC Cell Zero-Value Operation Disable
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial intelligence and neural network operations incur significant power consumption due to numerous zero-value matrix multiplication operations, especially with the use of rectified linear units (ReLU) and pruning techniques, which convert negative values to zero and prune many values to zero, leading to redundant calculations.
Innovation Solution
A matrix multiplier with a computing array of MAC cells that disables multiplication accumulation operations when either input value is a specified value (e.g., zero), reducing power consumption by selectively enabling or disabling operations based on input values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If matrix multiplication operations are performed for all input values in neural networks, then calculation completeness is maintained, but power consumption increases significantly due to redundant zero-value operations
Solution Approach 1:
The patent applies partial action by disabling multiplication accumulation operations only for MAC cells where at least one input value is zero, rather than performing all operations. This selective execution reduces power consumption while maintaining calculation completeness for non-zero values, directly resolving the contradiction between energy efficiency and computational thoroughness
2Productivity
If ReLU activation function and pruning techniques are used to optimize neural networks, then model efficiency improves, but the number of zero-value operations increases leading to higher power consumption
Solution Approach 1:
The patent converts the harmful effect of increased zero-value operations (caused by ReLU and pruning) into a benefit by detecting zero inputs and disabling corresponding MAC cell operations. This transforms the previously harmful redundancy into an opportunity for power savings, allowing the system to maintain model efficiency while eliminating the associated energy waste
Data Source
AI summary
A matrix multiplier and an operation method thereof are provided. The matrix multiplier includes a plurality of first input lines, a plurality of second input lines and a computing array. The computing array includes a plurality of multiplication accumulation (MAC) cells. A first MAC cell of the plurality of MAC cells is coupled to a first corresponding input line of the plurality of first input lines and a second corresponding input line of the plurality of second input lines to receive a first input value and a second input value to perform a multiplication accumulation operation. When at least one of the first input value and the second input value is a specified value, the multiplication accumulation operation of the first MAC cell is disabled.