Reservoir Element Using Spin-Orbit Torque for Stable Neuromorphic Oscillation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neuromorphic elements using STO elements face issues with resonance frequency variations due to manufacturing errors, leading to improper interaction between chips and potential element failure from high-frequency current application, which increases power consumption and circuit design burden.
Innovation Solution
A reservoir element comprising ferromagnetic layers separated by a spin-orbit torque wiring and a spin transport layer, with a gate electrode, that stabilizes operation by synchronizing magnetization oscillations and reducing power consumption through spin current interference.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If STO elements are used as chips in neuromorphic elements, then the element can perform neuromorphic computing functions, but resonance frequency variations due to manufacturing errors cause improper interaction between chips
Solution Approach 1:
The patent changes the operating principle from resonance-based interaction to spin-orbit torque-based interaction. By applying spin-orbit torque through current flow in the spin-orbit torque wiring layer, the magnetization state of ferromagnetic layers can be controlled without relying on precise resonance frequency matching, thus resolving the manufacturing precision issue while maintaining neuromorphic computing functionality
Solution Approach 2:
The patent replaces the mechanical vibration-based resonance interaction with an electromagnetic spin-orbit torque mechanism. Instead of using high-frequency mechanical oscillation that requires precise frequency alignment, the system uses spin-polarized current to directly manipulate magnetization states, eliminating the resonance frequency alignment problem
2Speed
If high-frequency current is applied to STO elements including insulating layer, then the element can oscillate magnetization, but element failure occurs after long-time operation
Solution Approach 1:
The patent introduces a spin-orbit torque wiring layer as an intermediary between the current source and the ferromagnetic layers. This intermediary layer converts electrical current into spin-orbit torque, which then acts on the magnetization. This indirect mechanism allows magnetization oscillation at high frequencies without requiring direct high-frequency current application to the insulating layer, thus improving reliability
Solution Approach 2:
The patent substitutes the direct high-frequency electrical oscillation mechanism with a spin-orbit torque-mediated mechanism. The spin-orbit torque wiring layer acts as a transducer that converts electrical energy into magnetic energy through spin transfer torque, enabling high-frequency magnetization oscillation without the harmful effects of direct high-frequency current application to insulating layers
3Productivity
If learning is performed in each hierarchy of neuromorphic element, then the element can process information effectively, but circuit design becomes complex and power consumption increases
Solution Approach 1:
The patent merges the learning function into a single output part rather than distributing learning across multiple hierarchical levels. The reservoir element performs fixed-weight transformation of input signals, while the output part performs all learning operations. This consolidation simplifies circuit design and reduces the number of learnable parameters while maintaining information processing capability
Solution Approach 2:
The patent segments the neuromorphic system into two distinct functional parts: a reservoir element with fixed weights that performs parallel signal transformation, and an output part with learnable weights that performs adaptive learning. This segmentation allows the complex learning task to be concentrated in a smaller, more manageable component while the reservoir handles parallel processing, reducing overall system complexity
4Productivity
If the number of chips (neurons) is increased in hierarchical type element, then the processing power increases, but power consumption of the neuromorphic element increases
Solution Approach 1:
The patent merges all learning operations into a single output part, eliminating the need for multiple learning operations across hierarchical levels. This consolidation reduces the total number of active learning circuits and associated power consumption, while the reservoir element uses passive fixed-weight transformations that consume minimal energy, thus maintaining processing power with lower overall power consumption
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
The proposed design enables stable operation of neuromorphic elements by synchronizing magnetization oscillations and reducing power consumption, enhancing data compression and processing efficiency.
Implementation Method 1
at least one spin-orbit torque wiring that faces at least one of the plurality of ferromagnetic layers
Implementation Method 2
a spin transport layer that faces the plurality of ferromagnetic layers, connects at least the two ferromagnetic layers closest to each other among the plurality of ferromagnetic layers and transports spins
Implementation Method 3
The resonance frequency of an STO element varies due to manufacturing errors or the like, and STO elements may not interact with each other properly
Data Source
AI summary
A reservoir element according to an aspect of the present invention includes a plurality of ferromagnetic layers laminated in a first direction and separated from each other, at least one spin-orbit torque wiring that faces at least one of the plurality of ferromagnetic layers, and a spin transport layer that faces the plurality of ferromagnetic layers, connects at least the two ferromagnetic layers closest to each other among the plurality of ferromagnetic layers and transports spins.