Stickification With Anywhere Padding for Granular Tensor Data Access

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing data manipulation processes in artificial intelligent hardware are inefficient and require optimization to enhance performance and resource utilization.

Innovation Solution

Implementing stickification operations with anywhere padding to read and write data at a granular level, annotating regions of activation and weight tensor data, and using padding to optimize data manipulation on artificial intelligent hardware.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If traditional data manipulation processes are used in artificial intelligent hardware, then the implementation is simpler, but the processing efficiency is lower and computational overhead is higher

Engineering Contradiction:

Improvedata manipulation efficiencyVSAvoidoperation complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The patent segments data manipulation operations into distinct phases: stickification (organizing data into contiguous blocks), padding (adding filler elements), and unstickification (separating data). This segmentation allows each operation to be optimized independently, improving overall data manipulation efficiency while maintaining manageable complexity through modular design

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies preliminary action by performing stickification and padding operations before data processing to pre-organize data in optimal formats. This preliminary preparation reduces computational overhead during actual processing by ensuring data is already in the required contiguous or padded format, eliminating the need for repeated reorganization operations

Inventive Principle:

Principle #10Preliminary action

2Manufacturing precision

If granular level data read and write operations are implemented, then data manipulation precision is improved, but computational overhead increases

Engineering Contradiction:

Improvedata manipulation precisionVSAvoidcomputational overhead

Core Design Contradiction:

Manufacturing precisionVSUse of energy by moving object

Solution Approach 1:

The patent merges multiple granular data operations into unified stickification and unstickification operations. Instead of individually managing each data element's positioning and padding, the system combines these tasks into batch operations that process contiguous blocks of data simultaneously, maintaining precision while reducing the total number of operations and computational overhead

Inventive Principle:

Principle #5Merging (Combining)

3Productivity

If stickification operation with padding is used, then resource utilization is improved, but operation complexity increases

Engineering Contradiction:

Improveresource utilizationVSAvoidoperation complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The patent applies parameter changes by dynamically adjusting padding parameters and stickification block sizes based on data characteristics and hardware constraints. This allows the system to optimize resource utilization for different data types and workloads while managing complexity through parameterized configurations rather than hard-coded complex logic

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS12468947B2Stickification using anywhere padding to accelerate data manipulation

Publication Date: 2025.11.11 INTERNATIONAL BUSINESS MACHINE CORPORATION

AI summary

Embodiments are provided for efficient realization of memory-bound operations in a computing system by a processor. Data may be read from and written to a memory at a granular level using a stickification operation. One or more regions of activation and weight tensor data on the memory may be annotated by coupling the stickification operation with padding.