Neural Network Processor Cluster Task Descriptor Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional multi-core neural network processors face ineffective coordination control between single-core processors, leading to repeated, missing, and non-uniform task configurations, which hinders the full utilization of their computational performance.
Innovation Solution
A neural network processor cluster with a main control unit and DMA module analyzes task descriptors to distribute subtasks effectively among single-core processors, using a peripheral module that includes a cluster interface, mailbox, bus forwarding module, program memory, and data memory to manage task configuration and parameter sharing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If multiple single-core neural network processors are used to form a multi-core processor, then computational power is improved, but coordination control between processors deteriorates
Solution Approach 1:
The patent introduces a main control unit as an intermediary component that coordinates between multiple single-core neural network processors. This mediator receives task descriptors, analyzes them, and distributes subtasks to appropriate processors, thereby managing the complexity of multi-core coordination without reducing computational power.
Solution Approach 2:
The patent segments a complex task into multiple subtasks that can be distributed across different single-core processors. By dividing the overall task and its descriptor into manageable portions, the system can effectively manage multiple processors while maintaining coordination through the main control unit.
2Productivity
If conventional single-core processors operate independently in a multi-core system, then processing parallelism is improved, but task configuration uniformity deteriorates
Solution Approach 1:
The main control unit implements a feedback mechanism where it receives task descriptors, analyzes the task requirements, and adjusts the distribution of subtasks to ensure uniform configuration across all processors. This closed-loop control ensures that parallel processing maintains configuration consistency.
Solution Approach 2:
The main control unit serves multiple functions: receiving task descriptors, analyzing task requirements, determining subtask distribution, modifying descriptors, and configuring processors. This multi-functional component ensures uniform task configuration across all processors while enabling parallel processing.
3Speed
If task descriptors are directly distributed to multiple processors, then configuration speed is improved, but configuration accuracy deteriorates
Solution Approach 1:
The main control unit performs preliminary analysis of task descriptors before distribution. By pre-processing and analyzing the task requirements upfront, the system can quickly and accurately determine how to divide and distribute subtasks to processors, ensuring both speed and accuracy in configuration.
Solution Approach 2:
The main control unit creates modified copies of the original task descriptor for each processor, tailoring each copy to the specific subtask and target processor. This copying and adaptation process ensures accurate configuration for each processor while maintaining the overall task integrity.
Data Source
AI summary
A neural network processor and a control method are provided. The neural network processor includes a neural network processor cluster formed by multiple single-core neural network processors and a peripheral module. The peripheral module includes a main control unit and a DMA module. The DMA module is used to convey a first task descriptor to the main control unit. The main control unit is used to: analyze the first task descriptor, determine, according to an analysis result, a subtask to be distributed to each selected processor; modify the first task descriptor to acquire a second task descriptor respectively corresponding to each selected processor; and distribute each second task descriptor to each corresponding selected processor, and activate each selected processor to process the corresponding subtask. The main control unit schedules and manages all of the single-core neural network processors, thereby leveraging operational performance of the neural network processor.