Dynamic Binary Signal Coding for Zero-Skipping MAC Operations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computing architectures for artificial neural networks face high energy consumption due to the large number of multiply-accumulate (MAC) operations, especially when dealing with digital signals that have a high density of zero values, as current coding methods do not account for the nature of the signals, leading to unnecessary operations.
Innovation Solution
A dynamic precision coding method that decomposes digital signals into binary words, transmits only non-zero pairs, and optimizes the size of binary words based on the statistical distribution of zero values to minimize energy consumption, reducing the number of MAC operations by eliminating zero binary words and sharing addresses across samples.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If conventional coding methods are used for digital signals in neural networks, then the implementation is simple, but energy consumption is high due to unnecessary MAC operations on zero values
Solution Approach 1:
The patent extracts and removes zero binary words from the digital signal before transmission and processing. By identifying and eliminating zero-valued binary words, the system prevents unnecessary MAC operations at the computing unit, directly reducing energy consumption while maintaining signal integrity through selective transmission of only non-zero components
Solution Approach 2:
The patent implements dynamic precision coding where the number of bits transmitted varies based on the actual signal characteristics. Instead of fixed-precision transmission, the system adaptively adjusts the number of binary words to transmit based on the density of zero values, optimizing energy consumption dynamically according to signal properties
2Use of energy by moving object
If the number of MAC operations is reduced by eliminating zero values, then energy consumption decreases, but the coding method must account for signal characteristics increasing complexity
Solution Approach 1:
The patent applies local quality by treating different portions of the digital signal differently based on their zero-density characteristics. Instead of uniformly processing all binary words, the system identifies regions with high zero density and applies selective elimination only to those portions, preserving non-zero components while removing redundant zero transmissions
Solution Approach 2:
The patent changes the parameter of binary word transmission by dynamically adjusting the number of bits transmitted based on signal characteristics. The system modifies transmission parameters adaptively, sending fewer bits when zero density is high and maintaining full precision when needed, thus optimizing energy consumption while preserving necessary signal fidelity
3Productivity
If binary words are decomposed and selectively transmitted, then the number of MAC operations is reduced, but the coding and transmission process becomes more complex
Solution Approach 1:
The patent segments the digital signal into individual binary words after decomposition, allowing selective identification and transmission of non-zero components. This segmentation enables the coding device to process and transmit only relevant data portions, improving processing efficiency while the modular nature of segmentation keeps the complexity manageable through systematic handling of discrete binary units
Data Source
AI summary
A computer-implemented method for coding a digital signal intended to be processed by a digital computing system includes the steps of: receiving a sample of the digital signal quantized on a number Nd of bits, decomposing the sample into a plurality of binary words of parameterizable bit size Np, coding the sample through a plurality of pairs of values, each pair comprising one of the binary words and an address corresponding to the position of the binary word in the sample, transmitting the pairs of values to an integration unit in order to carry out a MAC operation between the sample and a weighting coefficient.