AI Chip Data Stream Network for Bandwidth-Mismatched Modules
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing artificial intelligence chips face low resource utilization due to data acquisition through instruction sets, leading to inefficient resource consumption.
Innovation Solution
The artificial intelligence chip is designed with a plurality of computing modules connected in an operation order, forming a data stream network where data flows according to a preset direction, and includes data stream dams to manage bandwidth mismatches between modules, along with local and global data stream storage modules to facilitate efficient data processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If data is acquired by means of instruction sets, then the artificial intelligence chip can process data according to operation rules, but resource utilization rate decreases
Solution Approach 1:
The patent extracts the data acquisition function from the traditional instruction-set-based control mechanism and implements it through dedicated data input interfaces and data stream networks. This separation allows computing modules to focus on computation while data flows through specialized pathways, eliminating the resource overhead of instruction interpretation and improving overall resource utilization.
Solution Approach 2:
The patent introduces data stream dams as intermediary components between computing modules. These dams act as buffers and regulators that manage data flow between modules with different bandwidths, enabling efficient data transmission without requiring complex instruction-based coordination, thus improving resource utilization while maintaining processing capability.
2Productivity
If computing modules are connected in data stream network with preset data flowing direction, then resource utilization rate increases, but bandwidth mismatch between modules causes data transmission inefficiency
Solution Approach 1:
Data stream dams serve as intermediary components positioned between computing modules with different bandwidth characteristics. These dams buffer data flows, regulate transmission rates, and match bandwidths between adjacent modules, ensuring that data transmission efficiency is maintained even when connected modules have different processing speeds and bandwidth capabilities.
Solution Approach 2:
The patent dynamically adjusts data transmission parameters through the data stream dam mechanism. When bandwidth mismatch is detected, the dam modifies data flow parameters such as transmission rate and buffering depth to optimize data flow, thereby maintaining high transmission efficiency across modules with varying bandwidth characteristics while preserving the benefits of the data stream network architecture.
Data Source
AI summary
Embodiments of the present disclosure provide an artificial intelligence chip and a data processing method based on an artificial intelligence chip. The artificial intelligence chip includes: a plurality of computing modules; each computing module is configured to process data on the basis of one of operation nodes corresponding to artificial intelligence algorithms, and the plurality of computing modules are connected in turn according to an operation order of the artificial intelligence algorithms; and the data flows, according to a preset data flowing direction, in a data stream network formed by the plurality of computing modules.