Fractal Calculating Device Homogeneous Architecture Programming Efficiency

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current machine learning calculating devices face challenges in programming efficiency due to heterogeneity, parallelism, and hierarchy, limiting the widespread adoption of machine learning technology despite advancements in power efficiency.

Innovation Solution

A fractal calculating device with a homogeneous and serial hierarchical structure, utilizing a fractal von Neumann architecture and a fractal instruction set, decomposes instructions into serial and parallel sub-instructions to simplify programming and optimize operations across multiple layers of calculation units.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Use of energy by moving object

If machine learning calculating devices use heterogeneous, parallel, and hierarchical structures to improve performance, then power efficiency increases, but programming efficiency deteriorates

Engineering Contradiction:
Improvepower efficiencyVSAvoidprogramming efficiency
Core Design Contradiction:
Use of energy by moving objectVSEase of operation

Solution Approach 1:

The patent applies universality by designing a unified instruction set architecture that can control calculation units across different layers and scales. The same instruction set can program both simple and complex devices, eliminating the need for different programming approaches for heterogeneous structures. This allows developers to write programs once and execute them efficiently on devices with varying degrees of parallelism and hierarchy.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent segments the calculation system into multiple layers of calculation units, each capable of independent operation but coordinated through the unified instruction set. This segmentation allows the system to maintain high performance through parallel processing while simplifying programming by treating each layer as a standardized component that follows the same instruction protocol.

Inventive Principle:
Principle #1Segmentation

2Productivity

If machine learning calculating devices increase peak performance through complex architecture, then computing power improves, but device complexity increases

Engineering Contradiction:
Improvepeak performanceVSAvoidarchitecture complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent implements a nested hierarchical structure where calculation units are organized in multiple layers, with each layer containing calculation units of the next level. This nesting allows the system to achieve high peak performance through parallel processing at multiple levels while maintaining manageable complexity through the unified instruction set that operates consistently across all layers.

Inventive Principle:
Principle #7Nested doll (Nesting)

Solution Approach 2:

The patent adds a hierarchical dimension to the calculation architecture, organizing calculation units in multiple layers rather than a single flat structure. This dimensional change enables the system to scale performance by adding layers while keeping each individual layer relatively simple and uniform, thus managing overall complexity.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

3Ease of operation

If machine learning calculating devices use simplified homogeneous structure, then programming efficiency improves, but computing scalability deteriorates

Engineering Contradiction:
Improveprogramming efficiencyVSAvoidcomputing scalability
Core Design Contradiction:
Ease of operationVSAdaptability or versatility

Solution Approach 1:

The patent achieves universality by designing a single instruction set that can effectively program calculation units regardless of their position in the hierarchy or the overall scale of the device. This universal approach maintains programming simplicity while enabling the system to scale from small to large configurations by simply adding or removing calculation units according to the fractal pattern.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent enables scalability through parameter changes in the hierarchical structure, where the same basic calculation unit design can be replicated and organized in different numbers and configurations. By changing structural parameters (number of layers, units per layer) rather than the fundamental unit design, the system achieves scalability while maintaining programming simplicity.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentUS11841822B2Fractal calculating device and method, integrated circuit and board card
Publication Date: 2023.12.12 CAMBRICON TECH CO LTD
  • US11841822B2 patent drawing
  • US11841822B2 patent drawing
  • US11841822B2 patent drawing

AI summary

A fractal computing device according to an embodiment of the present application may be included in an integrated circuit device. The integrated circuit device includes a universal interconnect interface and other processing devices. The calculating device interacts with other processing devices to jointly complete a user specified calculation operation. The integrated circuit device may also include a storage device. The storage device is respectively connected with the calculating device and other processing devices and is used for data storage of the computing device and other processing devices.