Unlock instant, AI-driven research and patent intelligence for your innovation.

A matrix convolution optimization operation method and circuit

A technology of matrix operation and operation circuit, which is applied in the field of neural network circuit and can solve the problem of large amount of operation of neural network.

Active Publication Date: 2022-03-29
FUZHOU ROCKCHIP SEMICON
View PDF6 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0003] For this reason, it is necessary to provide a technical solution for PCBA board testing to solve the problem that there is no existing circuit structure to realize the Winograd algorithm, which leads to a large amount of calculation of the neural network.

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 more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • A matrix convolution optimization operation method and circuit
  • A matrix convolution optimization operation method and circuit
  • A matrix convolution optimization operation method and circuit

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0043] In order to explain in detail the technical content, structural features, achieved goals and effects of the technical solution, the following will be described in detail in conjunction with specific embodiments and accompanying drawings.

[0044] see figure 1 , is a schematic diagram of the principle of the Winograd algorithm involved in an embodiment of the present invention. In simple terms, the Winograd algorithm uses more addition calculations to reduce multiplication calculations. Therefore, a premise is that in the processor, the number of clock cycles for multiplication is greater than the number of clock cycles for addition. Winograd calculates the number of multiplications that need to be done for convolution:

[0045] μ(F(m×n,r×s))=(m+n-1)×(n+s-1)

[0046] r×s represents the size of the convolution kernel, and m×n represents the output size.

[0047] So, do a simple comparison calculation: 3×3 convolution kernel, the output is 2×2, then the number of multip...

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

No PUM Login to View More

Abstract

The present invention provides a matrix convolution optimization operation method and circuit, wherein the circuit includes a multiplication operation unit, a matrix operation unit, and a data buffer unit; the matrix operation unit includes a first matrix operation unit and a second matrix operation unit; The multiplication operation unit includes a plurality of multipliers; the first matrix operation unit is used to receive the first operation result of the multiplier connected to it, perform the first matrix operation, obtain the first matrix operation result, and transfer the first matrix operation result A matrix operation result is stored in the corresponding data cache unit; the second matrix operation unit is used to obtain the first matrix operation result of the data cache unit, perform a second matrix operation, obtain and output the second matrix operation result. The circuit designed by the above scheme can realize the relevant functions of the Winograd algorithm and reduce the amount of multiplication in the convolution process.

Description

technical field [0001] The invention relates to the field of neural network circuits, in particular to a matrix convolution optimization operation method and circuit. Background technique [0002] With the rapid development of the artificial intelligence industry, users have higher and higher requirements for the computing speed of neural networks. Convolution operation is an important part of the neural network operation process. Due to its huge amount of calculation, it has become the bottleneck of neural network acceleration. The Winograd algorithm is an optimization algorithm for matrix multiplication, which can effectively reduce the amount of multiplication in convolution, but there is no circuit structure that can realize this algorithm, resulting in complex convolution operations and affecting the overall neural network operation. speed. Contents of the invention [0003] Therefore, it is necessary to provide a technical solution for PCBA board testing, which is ...

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
no application Login to View More
Patent Type & Authority Patents(China)
IPC IPC(8): G06N3/04G06N3/063
CPCG06N3/063G06N3/045
Inventor 廖裕民张越
Owner FUZHOU ROCKCHIP SEMICON