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

Spiking neuron hardware architecture for AER feed-forward classification systems

A classification system and hardware architecture technology, applied in biological neural network models, physical implementation, etc., can solve the problems of low parallelism, inability to achieve equipment miniaturization, and high cost, and achieve a large volume, which is conducive to miniaturization and real-time performance. Effect

Active Publication Date: 2019-05-10
青岛展诚科技有限公司
View PDF4 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0006] At present, the SNN algorithm mainly relies on software implementation, which is slow, low in parallelism, and cannot be processed in real time.
Moreover, the software needs a large computer support, in addition to the high cost, it is also impossible to realize the miniaturization of the equipment.

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
  • Spiking neuron hardware architecture for AER feed-forward classification systems
  • Spiking neuron hardware architecture for AER feed-forward classification systems
  • Spiking neuron hardware architecture for AER feed-forward classification systems

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0026] The neuron hardware structure of the classification system SNN proposed by the present invention is as follows: figure 2 shown. Spike neurons can be specifically divided into weight storage area, weight read-write gate, membrane potential multiplication accumulator, PSP function generator, trigger judge, control and timing, and weight correction module. There are 7 parts in total. Some functions are as follows:

[0027] (1) Control and timing: In order to meet the internal coordination of neurons and the speed of network processing, two internal clocks, CLK and SLOW_CLK, are required. The CLK clock cycle is much smaller than SLOW_CLK, which is determined by comparing with pulse code sampling and membrane potential speed of network processing. SLOW_CLK is used to coordinate the internal work of neurons, and all input membrane potential accumulation results can be obtained in less than one SLOW_CLK period. In addition, the RST signal is required for reset to zero, 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 relates to the field of AER image sensor image recognition, aims to meet requirements of an event-driven feedforward classification system based on an AER image sensor on instantaneity, parallel processing and miniaturization and provides a SNN pulse neutron hardware realizing structure based on a FPGA (field programmable gate array) platform. A pulse neutron hardware architecture for an AER feedforward classification system is composed of 7 parts including a weight storage area, a weight reading-writing gate, a membrane potential multiplication accumulator, a PSP function generator, a triggering judger, a control and time sequence and a weight correction module. The pulse neuron hardware architecture is mainly applied to image sensor image recognition occasions.

Description

technical field [0001] The invention relates to the field of AER image sensor image recognition, in particular to a neuron hardware implementation used in AER image recognition and classification impulse neural networks. Background technique [0002] The AER (Address-Event Representation, AER, address-event representation) image sensor detects changes in the target scene in real time. Compared with the traditional image sensor based on the "frame scan" imaging method, it can filter out static background pixel data and greatly reduce redundancy. data. AER only outputs an asynchronous digital event stream of relevant information, which enables subsequent processing systems to be designed to be fully event-driven. [0003] The event-driven feed-forward classification system based on AER image sensor, similar to other classification systems, can be divided into two parts: feature extraction and classification. The feature extraction part uses Gabor direction filter and maximiz...

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): G06N3/06
CPCG06N3/063
Inventor 徐江涛周义豪高志远聂凯明高静马建国
Owner 青岛展诚科技有限公司