Memory Cell Value Recalibration for ANN Error Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Process variations in memory cells lead to errors in the output of artificial neural networks (ANNs) due to variations in voltage-conductance characteristics, causing progressive error accumulation during MAC operations.
Innovation Solution
The values stored in memory cells, such as weights of an ANN, are updated based on process variation characteristics to recalibrate and reduce error accumulation, by incrementing or decrementing the values to account for the variations in memory cells.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If values are stored in memory cells for ANN operations, then the computing device can perform neural network operations, but process variations cause voltage-conductance characteristic deviations leading to output errors
Solution Approach 1:
The patent applies preliminary action by measuring process variation characteristics of memory cells before storing ANN values, and pre-compensating for these variations by adjusting the values to be stored. This upfront calibration eliminates the need for real-time correction during neural network operations, thereby maintaining both productivity and reliability.
Solution Approach 2:
The patent changes the parameters (values) to be stored in memory cells based on measured process variation characteristics. By adjusting these parameters according to actual memory cell behavior, the system compensates for manufacturing variations and ensures accurate neural network outputs despite hardware imperfections.
2Measurement precision
If process variation characteristics are measured and values are updated to compensate, then output accuracy is improved, but additional processing steps and time are required
Solution Approach 1:
The compensation process is performed in advance during memory cell initialization or configuration, rather than during actual neural network computations. This timing strategy minimizes the impact on computational productivity while achieving the goal of improving measurement precision and output accuracy.
3Reliability
If values are incremented or decremented to account for process variations, then error propagation is reduced, but the complexity of the storage and retrieval process increases
Solution Approach 1:
The patent modifies the values to be stored in memory cells based on measured process variation characteristics. This parameter adjustment approach systematically compensates for manufacturing variations without requiring complex real-time correction circuits, thereby improving reliability while maintaining relatively simple device architecture.
Data Source
AI summary
Modifying values stored in memory cells based on process variation can include receiving process variation information for a first plurality of memory cells of a memory array. A plurality of values to be stored in the memory cells can be modified based on the process variation information. The plurality of instructions to store the plurality of values in a second plurality of memory cells can be compiled. The plurality of modified values can also be stored in a second plurality of memory cells at run time.