A neural network synaptic structure based on memristor unit and its adjustment method
A technology of neural network and memristor, which is applied in the field of synaptic structure and regulation of neural network based on memristor unit, which can solve the problems of inability to complete neural network functions, inaccurate adjustment of synaptic weights, etc.
Active Publication Date: 2021-10-26
UNIV OF ELECTRONICS SCI & TECH OF CHINA
View PDF4 Cites 0 Cited by
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
[0005] Using the difference between two memristive units to represent the weight of a synapse, and using the Manhattan weight update rule, can only achieve simple image recognition because it does not fundamentally solve the inaccuracy of synaptic weight adjustment , so it is impossible to complete more complex neural network functions
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
[0026] Below in conjunction with accompanying drawing and embodiment the present invention is described in further detail
[0027] figure 2 The test and control system used in the embodiment, the upper part of the figure is the 3706A system control switch, which selects the memristive unit to be operated by connecting and disconnecting the internal circuit. The 2400 interface is used to connect the Keithley2400Source Meter digital multimeter to read the resistance value of the memristor unit. It is also connected to the function generator interface that generates the excitation voltage, and the pulse voltage generated by the function generator is used to update the resistance value of the memristor unit. . The DUT in the figure is the memristive unit to be controlled, and only one unit is shown in the figure for illustration.
[0028] image 3 The memristive unit array prepared for the embodiment, each cross point is a memristive unit, the two ends are connected to the cir...
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 relates to the technical field of computer and electronic information, in particular to a memristor unit-based neural network synapse structure and an adjustment method thereof. In the present invention, multiple memristor units are cascaded together as an electronic synapse (that is, a cascaded memristor unit group), and a weighted mapping rule is used to map its resistance value to the synaptic weight value in the neural network , and adopt the method of full adjustment or step-by-step adjustment to avoid the influence caused by the inaccurate adjustment of the memristor during the learning process of the neural network, so that all kinds of neural networks can run normally on the memristor-based system . The invention solves the problem of mapping from the resistance value to the synaptic weight value and the uncertainty problem of the change of the resistance value of the memristive unit.
Description
technical field [0001] The invention relates to the technical field of computer and electronic information, in particular to a memristor unit-based neural network synapse structure and an adjustment method thereof. Background technique [0002] Synapse is the intermediate structure connecting different neurons in the neural network, and its weight value is continuously updated through the corresponding neural network algorithm, which is the basis of neural network information processing. As a new type of component with adjustable resistance, memristor is considered to be one of the four basic circuit components alongside resistors, inductors and capacitors. [0003] The researchers found that the change law of synaptic connection strength is similar to the change law of the electrical characteristics of memristor (its conductance value is adjusted by the applied electrical signal), so a single memristor can simulate the function of a synapse. Compared with using electronic ...
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 帅垚乔石珺吴传贵罗文博王韬张万里彭赟潘忻强梁翔
Owner UNIV OF ELECTRONICS SCI & TECH OF CHINA