Electronic Synapse Using Spin-Orbit Torque for Neuromorphic Computing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current neuromorphic computing systems face inefficiencies in simulating the human brain's ultra-high density and low energy consumption due to the mismatch between CMOS transistors and biological synapses, making it challenging to implement ultra-high density (1011 synapses per cm−2) and low energy consumption (˜1 pJ per synaptic event) as seen in biological synapses.
Innovation Solution
A device structure featuring a domain-wall magnet with spin-orbit torque induced spike-timing dependent plasticity, comprising a heavy metal layer, a domain-wall magnet layer with perpendicular magnetic anisotropy, a pinned layer, and an oxide tunnel barrier, decoupling spike transmission and learning current paths to generate STDP, allowing for efficient synaptic learning and spike transmission.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If CMOS transistors are used to emulate synaptic functionality, then the system can be manufactured with existing technology, but the area overhead and power consumption become quite large
Solution Approach 1:
The patent replaces the mechanical/electronic CMOS transistor system with a magnetic domain wall-based system. The domain wall magnet layer with perpendicular magnetic anisotropy uses spin-orbit torque to control domain wall motion, which directly modulates conductance without requiring complex CMOS circuitry. This substitution eliminates the need for voltage-controlled switches and reduces power consumption while maintaining manufacturability through magnetic thin-film deposition techniques.
Solution Approach 2:
The patent changes the fundamental operating parameter from voltage control in CMOS to magnetic domain wall position control. By varying the domain wall position within the magnetic layer through spin-orbit torque, the conductance is continuously modulated. This parameter change enables synaptic functionality with lower energy consumption since magnetic domain wall motion requires significantly less energy than CMOS switching operations.
2Ease of manufacture
If CMOS transistors are used to emulate synaptic functionality, then the system can be manufactured with existing technology, but the area overhead becomes quite large
Solution Approach 1:
The patent replaces the mechanical/electronic CMOS transistor system with a magnetic domain wall-based system. The domain wall magnet layer with perpendicular magnetic anisotropy uses spin-orbit torque to control domain wall motion, which directly modulates conductance without requiring complex CMOS circuitry. This substitution eliminates the need for voltage-controlled switches and reduces power consumption while maintaining manufacturability through magnetic thin-film deposition techniques.
Solution Approach 2:
The patent merges multiple functions into a single magnetic tunnel junction structure. The domain wall magnet layer serves simultaneously as the conductive element, the synaptic weight storage medium, and the learning mechanism. By combining these functions that would otherwise require separate CMOS components into a single magnetic device, the area overhead is dramatically reduced while achieving ultra-high density synaptic emulation.
3Quantity of substance
If nanoscale devices are designed to achieve ultra-high density, then the synaptic density increases, but the energy consumption per synaptic event must be reduced to match biological efficiency
Solution Approach 1:
The patent changes the fundamental operating parameter from voltage control in CMOS to magnetic domain wall position control. By varying the domain wall position within the magnetic layer through spin-orbit torque, the conductance is continuously modulated. This parameter change enables synaptic functionality with lower energy consumption since magnetic domain wall motion requires significantly less energy than CMOS switching operations.
Solution Approach 2:
The patent implements self-service through the intrinsic spin-orbit coupling in the heavy metal layer. The spin-orbit torque is generated automatically by the current flowing through the heavy metal layer adjacent to the domain wall magnet, eliminating the need for external magnetic field generation or additional control circuitry. This self-generated torque mechanism reduces energy consumption while enabling precise control of domain wall position for synaptic weight modulation.
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 solution achieves efficient synaptic learning and spike transmission with decoupled current paths, enabling the implementation of STDP and reducing energy consumption, thereby addressing the inefficiencies in existing neuromorphic computing systems.
Implementation Method 1
Spin-orbit torque generated by the magnetic heterostructure generates STDP
Implementation Method 2
a heavy metal layer having a high spin orbit coupling
Implementation Method 3
the pinned layer, the oxide tunnel barrier, and the free layer form a magnetic tunnel junction
Implementation Method 4
the domain wall magnet layer having a perpendicular magnetic anisotropy
Data Source
AI summary
An electronic synapse is disclosed, comprising a heavy metal layer having a high spin orbit coupling, a domain wall magnet layer having a bottom surface adjacent to a top surface of the heavy metal layer, the domain wall magnet layer having a perpendicular magnetic anisotropy, the domain wall magnet layer having a domain wall, the domain wall running parallel to a longitudinal axis of the domain wall magnet layer, a pinned layer having perpendicular magnetic anisotropy, and an oxide tunnel barrier connected between the domain wall magnet layer and the pinned layer, wherein the pinned layer, the oxide tunnel barrier, and the free layer form a magnetic tunnel junction.