An architecture and method for accelerating neural network computing based on distributed weight storage
A distributed storage and neural acceleration technology, applied in the field of neural network computing, can solve problems affecting computing efficiency, data access delay, etc., to achieve the effect of improving computing efficiency, reducing high latency, and having complete functions
Active Publication Date: 2022-07-15
INSPUR SUZHOU INTELLIGENT TECH CO LTD
View PDF0 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
[0005] The present invention provides a framework and method for accelerating neural network computing based on distributed weight storage, which is used to solve the problem of data access delay and affecting computing efficiency in existing neural network computing
Method used
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View moreImage
Smart Image Click on the blue labels to locate them in the text.
Smart ImageViewing Examples
Examples
Experimental program
Comparison scheme
Effect test
Embodiment Construction
[0033] In order to clearly illustrate the technical features of the solution, the present invention will be described in detail below through specific embodiments and in conjunction with the accompanying drawings. The following disclosure provides many different embodiments or examples for implementing different structures of the invention. In order to simplify the disclosure of the present invention, the components and arrangements of specific examples are described below. Furthermore, the present invention may repeat reference numerals and / or letters in different instances. This repetition is for the purpose of simplicity and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed. It should be noted that the components illustrated in the figures are not necessarily drawn to scale. Descriptions of well-known components and processing techniques and processes are omitted from the present invention to avoid unneces...
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More PUM
Login to View More Abstract
The invention provides an architecture and method for accelerating neural network computing based on distributed weight storage. The architecture includes a main processor system, a distributed weight storage calculation module and a target acceleration neural network; the main processor system is used for running system programs , and load the driver of the distributed weight storage computing array; the distributed weight storage computing module divides the weight storage space according to the array to form a distributed storage computing unit array, and configures processing for each distributed storage computing unit A MEM-PE unit is formed; the MEM-PE unit is used to load the weights locally for calculation according to the calculation instruction, and perform distributed storage for the intermediate calculation results. The MEM-PE unit formed by the present invention has distributed memory and instruction memory, which can perform computation locally, and store intermediate results in other distributed memories locally or in the array, thereby reducing weight access in traditional neural network computing. high latency and improve computational efficiency.
Description
technical field [0001] The invention relates to the technical field of neural network computing, in particular to an architecture and method for accelerating neural network computing based on distributed weight storage. Background technique [0002] With the advent of the era of artificial intelligence, big data has shown an exponential growth, and various deep learning algorithms represented by neural networks have been widely used. The neural network algorithm is currently mainly based on multi-layer neural networks, also known as "deep belief networks". The algorithm first finds a value close to the optimal solution for the weights of the neural network through pre-training, and then performs fine-tuning. (fine-tuning), which optimizes the weights of the entire network. No matter what kind of neural network it is, a lot of convolution calculations are required in the process of reasoning, and the amount of calculation increases with the number of layers and neurons, whic...
Claims
the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More Application Information
Patent Timeline
Login to View More Patent Type & Authority Patents(China)
IPC IPC(8): G06N3/063
CPCG06N3/063
Inventor 邹晓峰李拓刘同强周玉龙王朝辉李仁刚
Owner INSPUR SUZHOU INTELLIGENT TECH CO LTD