Sorting Circuit Hardware Acceleration for Low Latency

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing machine learning sorting algorithms face inefficiencies and increased latency due to reliance on central processing units (CPU) for sorting operations, which can hinder faster and more efficient sorting processes.

Innovation Solution

A sorting circuit comprising an input buffer, a sorting buffer, a comparing circuit, and a processing circuit that compares input values with previous and sorted values to selectively insert or discard them, minimizing latency by operating independently of CPU overhead and optimizing the sorting buffer's content in descending order.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If sorting operations are performed through CPU, then general-purpose processing is achieved, but sorting latency increases and efficiency decreases

Engineering Contradiction:
Improvesorting efficiencyVSAvoidsorting latency
Core Design Contradiction:
ProductivityVSLoss of time

Solution Approach 1:

The patent replaces the mechanical CPU-based sorting system with a specialized sorting circuit that uses hardware comparators and buffers to perform sorting operations in parallel, eliminating the sequential processing bottleneck of CPU-based sorting and achieving faster sorting latency through dedicated hardware acceleration

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The sorting circuit is segmented into multiple independent functional units including input buffers, comparators, and sorting buffers that operate in parallel. Each comparator independently compares input values with sorted values, and the sorting buffer independently manages the sorted sequence, allowing simultaneous processing of multiple sorting operations without CPU intervention

Inventive Principle:
Principle #1Segmentation

2Speed

If sorting operations are performed through CPU, then flexibility in processing different data types is maintained, but sorting speed decreases

Engineering Contradiction:
Improvesorting speedVSAvoidsorting circuit complexity
Core Design Contradiction:
SpeedVSDevice complexity

Solution Approach 1:

The sorting circuit is designed to be self-sufficient with dedicated input buffers that automatically capture incoming values, comparators that autonomously perform comparisons, and sorting buffers that self-manage the sorted sequence. This self-service architecture eliminates the need for CPU control overhead while maintaining high sorting speed through autonomous hardware operation

Inventive Principle:
Principle #25Self-service

Data Source

PatentUS20240303035A1Sorting circuit and operating method thereof
Publication Date: 2024.09.12 SK HYNIX INC
  • US20240303035A1 patent drawing
  • US20240303035A1 patent drawing
  • US20240303035A1 patent drawing

AI summary

A sorting circuit includes an input buffer, a sorting buffer, a comparing circuit and a processing circuit. The input buffer sequentially receives values and sequentially store therein the values. The sorting buffer stores therein selected ones from among the values as each of the values is inserted thereto or discarded. The comparing circuit compares an input value currently received from among the values with each of a previous value and sorted values and output an input value comparison result. The previous value is a value that the input buffer receives immediately before receiving the input value, from among the values, and the sorted values are values stored in the sorting buffer immediately before the previous value is discarded or inserted into the sorting buffer. The processing circuit selectively inserts, based on the input value comparison result, the input value into the sorting buffer.