Integrated Accelerators for Neural Vector Computation and Retrieval

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing systems face inefficiencies in integrating data computation, storage, and retrieval processes, particularly in artificial intelligence applications involving vector data, leading to suboptimal performance and resource utilization.

Innovation Solution

A data computation and retrieval accelerator system is introduced, comprising a data processing accelerator and a data retrieval accelerator, which process input data using machine learning algorithms, enabling operations such as neural network algorithms and vector data retrieval, with integrated memory structures like SRAM and DRAM configurations, and potential integration across semiconductor dies or PCB boards.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If data computation, storage, and retrieval are performed using separate conventional systems, then each system can operate independently, but system efficiency and resource utilization deteriorate due to integration inefficiencies

Engineering Contradiction:

Improvesystem efficiencyVSAvoidintegration complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The patent combines data computation, storage, and retrieval functions into a single integrated accelerator system. The data processing accelerator and data retrieval accelerator are merged at the hardware level, allowing seamless data flow between computation and retrieval operations without external memory interfaces, thereby improving system efficiency while managing integration complexity through unified architecture design

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The integrated accelerator system performs multiple functions including neural network computation, vector data storage, and similarity search retrieval within a single device. This multi-functionality eliminates the need for separate systems, improving overall productivity while the universal memory structure handles diverse data types efficiently

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Speed

If vector data processing is performed using traditional database systems, then data storage capacity is maintained, but computation and retrieval speed deteriorate due to lack of specialized acceleration

Engineering Contradiction:

Improveretrieval speedVSAvoiddata storage capacity

Core Design Contradiction:

SpeedVSQuantity of substance

Solution Approach 1:

The patent replaces traditional mechanical database I/O operations with hardware-accelerated computation and retrieval. The data retrieval accelerator uses specialized hardware circuits to perform vector similarity searches directly in memory, substituting software-based database queries with dedicated hardware acceleration, thereby dramatically improving retrieval speed while maintaining storage capacity

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The unified memory structure acts as an intermediary between data processing and retrieval operations. Instead of transferring data between separate computation and storage systems, the unified memory allows both accelerators to access data simultaneously, eliminating data transfer bottlenecks and improving retrieval speed without compromising storage capacity

Inventive Principle:

Principle #24Intermediary (Mediator)

3Productivity

If multiple neural network operations are performed sequentially using conventional processors, then computational accuracy is maintained, but processing time increases due to lack of parallel acceleration

Engineering Contradiction:

Improvecomputation speedVSAvoidprocessing time

Core Design Contradiction:

ProductivityVSLoss of time

Solution Approach 1:

The patent segments the neural network computation into distinct processing stages handled by specialized units within the data processing accelerator. Different neural network layers and operations are divided into parallel processing segments, allowing simultaneous execution of multiple computational tasks while maintaining the accuracy requirements of each segment

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The integrated accelerator system enables continuous data flow from computation to retrieval operations without idle waiting time. The data processing accelerator generates output vectors that are immediately available to the data retrieval accelerator, eliminating gaps between computational stages and maximizing the continuity of useful action, thereby reducing overall processing time while maintaining computational accuracy

Inventive Principle:

Principle #20Continuity of useful action

Data Source

PatentUS20250328530A1System for accelerating data computation and retrieval

Publication Date: 2025.10.23 D NOTITIA INC

AI summary

Provided are a method and system of operating a machine learning algorithm with vector data in a data computation and retrieval system including a data processing accelerator that processes input data using machine learning and a data retrieval accelerator. The method includes operating a first neural network algorithm using input data, retrieving vector data similar to a result obtained by operating the first neural network algorithm, and operating a second neural network algorithm using the retrieved vector data as input data.