Memcapacitor Synapse Device for Fast Neural Network Weight Modulation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional neural network devices face challenges with slow operation speed, limited rewriting capability, and difficulty in setting multiple weights with good controllability due to the slow polarization change in dielectric films and hysteresis in phase-change or ferroelectric devices, which limits the number of gradations of coefficients used for integration.
Innovation Solution
A memcapacitor is introduced as a synapse device with a configuration that allows electrostatic capacitance to be changed at multiple levels by controlling the potential difference across its electrodes, enabling the formation of conductive filaments and improving weight modulation speed and controllability, using a non-volatile multi-value memory technology.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If phase-change or ferroelectric devices are used for weight modulation in neural network devices, then non-volatile memory capability is achieved, but the polarization change is slow and hysteresis effects limit the number of gradations of coefficients
Solution Approach 1:
The synapse device is segmented into multiple functional layers: a first dielectric layer for charge accumulation, a variable resistance portion for weight modulation, and a second dielectric layer for isolation. This segmentation allows the device to achieve both non-volatile memory capability through charge trapping in the dielectric layers and fast weight modulation through variable resistance change, resolving the contradiction between reliability and speed.
Solution Approach 2:
The invention changes the operational parameter from polarization change (slow, hysteresis-limited) to variable resistance change (fast, controllable). By applying voltage pulses that modulate the resistance of the variable resistance portion rather than relying on polarization switching, the device achieves rapid weight modulation with multiple distinguishable gradations, overcoming the speed and controllability limitations of phase-change and ferroelectric devices.
2Reliability
If dielectric films are used for weight storage, then non-volatile memory is achieved, but the polarization change is slow limiting arithmetic operation speed
Solution Approach 1:
The synapse device separates the memory function (charge accumulation in dielectric layers) from the computation function (variable resistance modulation). This allows non-volatile weight storage to be maintained while enabling fast arithmetic operations through resistance change, resolving the contradiction between reliability and productivity.
Solution Approach 2:
The invention replaces the slow polarization switching mechanism with a faster variable resistance modulation mechanism. By using voltage-controlled resistance change in the variable resistance portion rather than polarization switching in dielectric films, the device achieves both non-volatile memory and high-speed arithmetic operations, improving productivity without sacrificing reliability.
3Adaptability or versatility
If conventional synapse devices are used, then neural network functionality is achieved, but weight modulation has poor controllability and limited number of gradations
Solution Approach 1:
The invention changes the control parameter from polarization state (hard to control, limited gradations) to variable resistance value (easily controlled, multiple gradations). By modulating the resistance of the variable resistance portion through voltage pulses, the device achieves precise weight modulation with excellent controllability and a large number of distinguishable gradations, improving ease of operation while maintaining neural network functionality.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach enhances the arithmetic operation speed and reliability of neural network devices, allowing for more efficient weight modulation and improved learning performance by enabling the application of multiple weights and faster integration processes.
Implementation Method 1
the variable resistance portion 122 allows diffusion of a metal element constituting the upper electrode 121 to the inside of the variable resistance portion 122
Implementation Method 2
formation of conductive filaments and improving weight modulation speed and controllability
Implementation Method 3
the first dielectric layer 124 serves as a stopper that prevents diffusion of the metal atoms constituting the upper electrode 121 to the inside of the first dielectric layer 124
Implementation Method 4
electrostatic capacitance to be changed at multiple levels by controlling the potential difference across its electrodes
Data Source
AI summary
A memcapacitor according to an embodiment includes a first electrode, a first dielectric layer provided on the first electrode, a plurality of variable resistance portions provided separately from each other on the first dielectric layer, a second dielectric layer provided on the first dielectric layer and between the variable resistance portions, and a second electrode provided on the variable resistance portions and the second dielectric layer. Each of the variable resistance portions is formed of a material that allows diffusion of metal atoms constituting the second electrode to inside of the variable resistance portion, and the second dielectric layer is formed of a material that prevents diffusion of the metal atoms constituting the second electrode to inside of the second dielectric layer.