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

A storage module, module and data processing method based on neural network

A storage module and neural network technology, applied in biological neural network models, physical implementation, etc., can solve the problems of increasing computing processing time, large processing time, data access cost, etc., and achieve the effect of improving the scale of matrix operations and processing speed.

Active Publication Date: 2022-01-07
上海闪易半导体有限公司
View PDF6 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

In the process of neural network algorithm processing, a large number of matrix operations are included. At present, the calculation process is mainly realized through CPU (central processing unit) and memory. The CPU mainly performs core calculations, and accesses data from the memory in each operation. A large number of multiple data accesses greatly increase the time of the entire calculation process
And other parallel computing processor devices are also applied to neural networks, such as GPU (image processor) and TPU (tensor processor), etc. These devices can perform parallel computing, but data access still consumes a lot of processing time

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 storage module, module and data processing method based on neural network
  • A storage module, module and data processing method based on neural network
  • A storage module, module and data processing method based on neural network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0076] refer to Figure 6 As shown, in this embodiment, each storage unit includes a nonvolatile memory, the first direction is the row direction, and the second direction is the column direction. In the memory array, the first source and drain of each row of memory are Connected to the first electrical connection AL, the second source and drain of each column of memory are connected to the second electrical connection BL, the gate of each row of memory is connected to the third electrical connection CL, of course, the connection of the gate The manner may also be that the gate of each column of memory is connected to one electrical connection.

[0077] In the forward propagation, the first electrical connection AL can be used as the input terminal, and the second electrical connection BL can be used as the output terminal. Before matrix operation and propagation, data has been written in each memory and is in the first data state. When performing operations, the input signal...

Embodiment 2

[0081] refer to image 3 As shown, in this embodiment, each storage unit includes a non-volatile memory and a MOS device, the two devices are connected in series, the first source and drain of the memory are connected to the second source and drain of the MOS device, and the first The direction is the row direction, and the second direction is the column direction. In the storage array, the first source and drain electrodes of each row of MOS devices are connected to the first electrical connection line AL, and the second source and drain electrodes of each row of memory devices are connected to the second The electric connection line BL, the gate of each row of memory is connected to the third electric connection line CL, and the gate of each column of MOS devices is electrically connected to the fourth electric connection line DL. Of course, the connection mode of the gate can also be as follows: The gates of a column of memory are connected to an electrical connection, and ...

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 invention provides a memory module based on a neural network. A memory array is formed by memory cells including non-volatile memories. In the memory array, a source drain of a memory in one direction is electrically connected to a first electrical connection line, and the other side The other source and drain of the upward memory are electrically connected to the second electrical connection, and one of the first electrical connection and the second electrical connection is used to load the input signal, and the input signal can be used to represent the input vector of the layer in the neural network Due to the storage characteristics of non-volatile memory, the stored data is the conductance value between the two sources and drains of the memory, which can be used to represent the connection weight value. Then, the output signal output on the other electrical connection is The value after matrix operation. In this way, the processing, transmission and storage of data signals in the neural network can be completed through the storage array without consuming data access time, and at the same time, the matrix operation scale and processing speed in the neural network can be effectively improved.

Description

technical field [0001] The invention relates to the field of neural network integrated circuit design, in particular to a neural network-based storage module, module and data processing method. Background technique [0002] Neural network (Neuron Network, NN) is an algorithmic mathematical model that imitates the behavior characteristics of animal neural networks and performs distributed parallel information processing. This algorithmic model is widely used in artificial intelligence fields such as speech recognition, image recognition, and automatic driving. [0003] In the neural network, depending on the complexity of the system, the purpose of processing information is achieved by adjusting the interconnection relationship between a large number of internal nodes. In the process of neural network algorithm processing, a large number of matrix operations are included. At present, the calculation process is mainly realized through CPU (central processing unit) and memory. ...

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): G06N3/063
CPCG06N3/063
Inventor 张睿
Owner 上海闪易半导体有限公司