Hierarchical Memory Packing Using Sensitivity-Based Bit Reduction

Solution Overview

Problem

Current computer systems face challenges in reducing data movement between levels of a hierarchical memory system, leading to increased access times and power consumption, as they often require moving large amounts of data across different memory levels, which is costly in terms of both time and energy.

Innovation Solution

The implementation of systems and methods that utilize sensitivity analysis and Massively Asymmetric Precision (MAP) data representations to reduce the number of bits transferred between memory hierarchies by identifying and eliminating 'garbage' bits in numerical data, allowing for more efficient data storage and movement by dynamically adapting the bit length based on the sensitivity of the data elements.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If data is stored in hierarchical memory structures with multiple levels, then storage capacity is increased, but data movement between levels increases access time and power consumption

Engineering Contradiction:

Improvestorage capacityVSAvoidaccess time

Core Design Contradiction:

Quantity of substanceVSLoss of time

Solution Approach 1:

The patent extracts and eliminates unnecessary 'garbage bits' from numerical data representations. By performing sensitivity analysis on numerical data, the system identifies and removes bits that do not contribute to computational accuracy, thereby reducing the total data volume that needs to be moved between memory levels while maintaining storage capacity for meaningful information.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent dynamically changes the bit length parameter of numerical data based on sensitivity analysis results. By adapting the precision of numerical representations to the actual needs of computations, the system reduces data size for storage in hierarchical memory structures, thereby decreasing data movement volume and associated access time penalties.

Inventive Principle:

Principle #35Parameter changes

2Adaptability or versatility

If data is moved between multiple memory levels, then storage flexibility is improved, but power consumption increases

Engineering Contradiction:

Improvestorage flexibilityVSAvoidpower consumption

Core Design Contradiction:

Adaptability or versatilityVSUse of energy by moving object

Solution Approach 1:

The patent extracts and eliminates unnecessary 'garbage bits' from numerical data representations. By performing sensitivity analysis on numerical data, the system identifies and removes bits that do not contribute to computational accuracy, thereby reducing the total data volume that needs to be moved between memory levels while maintaining storage capacity for meaningful information.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The patent dynamically changes the bit length parameter of numerical data based on sensitivity analysis results. By adapting the precision of numerical representations to the actual needs of computations, the system reduces data size for storage in hierarchical memory structures, thereby decreasing data movement volume and associated power consumption.

Inventive Principle:

Principle #35Parameter changes

3Measurement precision

If full precision data is maintained throughout the system, then computational accuracy is preserved, but data movement volume increases

Engineering Contradiction:

Improvecomputational accuracyVSAvoiddata movement volume

Core Design Contradiction:

Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent dynamically changes the bit length parameter of numerical data based on sensitivity analysis results. By adapting the precision of numerical representations to the actual needs of computations, the system reduces data size for storage in hierarchical memory structures, thereby decreasing data movement volume while preserving computational accuracy where required.

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent applies different precision levels to different portions of numerical data based on their sensitivity characteristics. By performing sensitivity analysis, the system identifies which bits locally contribute to computational accuracy and which do not, allowing selective retention or elimination of bits in different data regions, thereby reducing overall data movement volume while maintaining accuracy where needed.

Inventive Principle:

Principle #3Local quality

Data Source

PatentUS11922021B2Arrangements for storing more data in memory when using a hierarchical memory structure

Publication Date: 2024.03.05 INTELLECTUAL PROPERTY SYSTEMS LLC

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.