Memory Data Packing Using Sensitivity-Based Precision Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computing systems face challenges in reducing data movement between levels of a hierarchical memory system, leading to increased costs and power consumption due to inefficient data storage and transmission practices.
Innovation Solution
The implementation of systems and methods that utilize sensitivity analysis and Massively Asymmetric Precision (MAP) data representations to identify and eliminate 'garbage' bits in numerical data, allowing for reduced bit length transmission and storage, thereby minimizing data movement between memory hierarchies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If data is stored and transmitted using standard precision formats, then numerical accuracy is maintained, but data movement volume and storage requirements increase
Solution Approach 1:
The patent applies local quality by assigning different precision levels to different data elements based on their sensitivity requirements. Sensitive data elements maintain higher precision while less sensitive elements use lower precision, optimizing the balance between numerical accuracy and data volume without uniform precision reduction
Solution Approach 2:
The system dynamically changes the precision parameter of data representations based on sensitivity analysis results. By adjusting the number of bits allocated to represent different data elements according to their actual sensitivity requirements, the system reduces overall data movement volume while maintaining necessary numerical accuracy for sensitive operations
2Reliability
If full precision data is transmitted between memory hierarchies, then computational accuracy is preserved, but power consumption and transmission costs increase
Solution Approach 1:
The patent implements local quality by applying different precision strategies to different data elements based on their sensitivity characteristics. Critical data elements use full precision transmission while less critical elements use reduced precision, thereby reducing overall power consumption without compromising computational accuracy where needed
Solution Approach 2:
The system changes the precision parameter dynamically based on sensitivity analysis, adjusting the number of bits transmitted for each data element according to its actual sensitivity requirements. This parameter adaptation reduces power consumption by minimizing data movement volume while preserving computational accuracy for sensitive operations
3Reliability
If standard data formats are used for storage, then data integrity is maintained, but memory capacity and storage efficiency decrease
Solution Approach 1:
The patent applies local quality by storing data elements with different precision levels according to their sensitivity requirements. Sensitive data elements are stored with higher precision to maintain integrity, while less sensitive elements use compressed lower-precision representations, optimizing memory capacity utilization without compromising data integrity where needed
Solution Approach 2:
The system changes the storage precision parameter based on sensitivity analysis results, allocating more bits to sensitive data elements and fewer bits to less sensitive ones. This parameter adaptation increases effective memory capacity by reducing storage requirements for non-critical data while maintaining data integrity for sensitive operations
Data Source
AI summary
Data employed in computations is processed so that during computations more of the data can be fit into or maintained in a smaller but higher speed memory than an original source of the data. More specifically, a sensitivity value is determined for various items of the data which reflect the number of bits in the data items that are not garbage bits, and only information in the data items that are indicated by the sensitivity value to not be garbage bits are necessarily effectively retained. At least the information that is not garbage bits and the corresponding associated sensitivity are packed together. The results of computations that are performed using the data items as at least one of the operands for the computation are associated with a sensitivity that is derived from the individual sensitivities of the operands used in the computation.