Partial-Sum Read Control for Low-Energy MAC Memory Access
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing hardware implementations of multiply accumulate (MAC) operations in artificial neural networks consume excessive computational energy due to full read operations, which can be optimized by reducing read energy consumption without compromising accuracy.
Innovation Solution
Implement a dynamic read energy (RRE) mechanism that monitors partial sum accumulation to determine if reduced read energy conditions are met, allowing for energy-efficient MAC operations by using lower read voltage or skipping read operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If full read operations are used in MAC operations, then accuracy is maintained, but energy consumption increases
Solution Approach 1:
The patent applies partial action by performing read operations selectively rather than fully. The dynamic read energy mechanism monitors the partial sum after each multiplication and decides whether to perform subsequent read operations based on whether the partial sum already indicates the final result will be zero or non-zero. This partial approach reduces energy consumption while maintaining accuracy by avoiding unnecessary read operations that would not change the outcome.
Solution Approach 2:
The patent implements dynamics through the dynamic read energy (RRE) mechanism that adapts read operation behavior based on real-time partial sum evaluation. The system dynamically adjusts between performing full read operations and skipping read operations depending on the current state of computation. This dynamic adaptation allows the system to optimize energy consumption for each specific MAC operation while ensuring accuracy is maintained when necessary.
2Use of energy by moving object
If reduced read energy is used, then energy consumption decreases, but read accuracy may be compromised
Solution Approach 1:
The patent applies feedback through the dynamic read energy mechanism that continuously monitors the partial sum after each multiplication operation. This feedback loop evaluates whether the current partial sum state indicates that subsequent read operations would be unnecessary (when the result is already determined to be zero) or necessary (when the result may be non-zero). Based on this feedback, the system dynamically adjusts read energy usage, applying reduced read energy only when accuracy can be maintained, thus resolving the contradiction between energy reduction and accuracy preservation.
Data Source
AI summary
Embodiments include monitoring a partial sum of a multiply accumulate calculation for certain conditions. When the certain conditions are met, a reduced read energy is used to read out memory contents instead of the regular read energy used. The reduced read energy may be obtained by reducing a pre-charge voltage, withholding a pre-charge voltage or providing a ground signal, and/or by reducing voltage hold times (i.e., reducing the time a pre-charge voltage is provided and/or discharged).


