南方科技大学知识苑(SUSTech KC): 适用于多精度神经网络的精度无损存算一体装置及方法

Title	适用于多精度神经网络的精度无损存算一体装置及方法
Alternative Title	Precision lossless storage and calculation integrated device and method suitable for multi-precision neural network
Author	周浩翔; 刘定邦 ; 刘俊; 吴秋平; 韩宇亮 ; 罗少波; 毛伟; 余浩
First Inventor	周浩翔
Original applicant	南方科技大学
First applicant	南方科技大学
Address of First applicant	518055 广东省深圳市南山区学苑大道1088号
Current applicant	南方科技大学
Address of Current applicant	518055 广东省深圳市南山区学苑大道1088号 (广东,深圳,南山区)
First Current Applicant	南方科技大学
Address of First Current Applicant	518055 广东省深圳市南山区学苑大道1088号 (广东,深圳,南山区)
Application Number	CN202210227427.X
Application Date	2022-03-08
Open (Notice) Number	CN114707647A
Date Available	2022-07-05
Status of Patent	实质审查
Legal Date	2022-07-22
Subtype	发明申请
SUSTech Authorship	First
Abstract	本发明公开了适用于多精度神经网络的精度无损存算一体装置及方法，所述方法包括：获取多精度神经网络的输入数据，并将所述输入数据按位拆分后进行数模转换，得到若干模拟信号；基于选择器和处理元件，将若干所述模拟信号与预设的权重采用时空复用的方式进行乘累加运算和多精度的重组运算，得到多精度神经网络的输出数据。本发明实施例通过多精度神经网络的输入数据与预设的权重采用时空复用的方式进行乘累加运算和多精度的重组运算，使得多精度神经网络的存算一体支持混合精度的神经网络计算，避免精度损失从而提升计算准确性，并且对比传统片上系统架构可以大幅度地提高计算能效。
Other Abstract	The invention discloses a precision lossless storage and calculation integrated device and method suitable for a multi-precision neural network, and the method comprises the steps: obtaining the input data of the multi-precision neural network, splitting the input data according to bits, and carrying out the digital-to-analog conversion, and obtaining a plurality of analog signals; and based on a selector and a processing element, carrying out multiply-accumulate operation and multi-precision recombination operation on the plurality of analog signals and a preset weight by adopting a space-time multiplexing mode to obtain output data of the multi-precision neural network. According to the embodiment of the invention, the multiply-accumulate operation and the multi-precision recombination operation are carried out through the input data of the multi-precision neural network and the preset weight in a space-time multiplexing manner, so that the storage and calculation integration of the multi-precision neural network supports the neural network calculation of mixed precision, and the precision loss is avoided, thereby improving the calculation accuracy. And compared with the traditional system-on-chip architecture, the computing energy efficiency can be greatly improved.
CPC Classification Number	G06N3/063 ; G06N3/045
IPC Classification Number	G06N3/063 ; G06N3/04
INPADOC Legal Status	（ENTRY INTO FORCE OF REQUEST FOR SUBSTANTIVE EXAMINATION）[2022-07-22][CN]
INPADOC Patent Family Count	1
Extended Patent Family Count	1
Priority date	2022-03-08
Patent Agent	朱阳波
Agency	深圳市君胜知识产权代理事务所(普通合伙)
URL	[Source Record]
Data Source	PatSnap
Document Type	Patent
Identifier	http://kc.sustech.edu.cn/handle/2SGJ60CL/533249
Department	Preparatory Office of SUSTech Institute of Microelectronics, SUSTech and HKUST 工学院_深港微电子学院
Recommended Citation GB/T 7714	周浩翔,刘定邦,刘俊,等. 适用于多精度神经网络的精度无损存算一体装置及方法.

Files in This Item:
There are no files associated with this item.

If you have any objections to this item, please fill out the form below and the administrator will contact you as soon as possible.
Content:
Email：	*
Affiliation Sequential No:
Verification Code:	Refresh

Any comments and suggestions are welcomed.
Title:	*
Content:
Email：	*
Verification Code:	Refresh