Three-State Spintronic Memory Cell for Neural Network Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional convolutional neural networks require large memory capacity and high energy consumption for training and recognition tasks, making them inefficient and time-consuming.
Innovation Solution
A three-state spintronic device is developed, comprising a magnetic tunnel junction with local magnetic domain wall pinning centers and nucleation centers, allowing for resistance state switching via spin current manipulation, which can be used in memory cells and read-write circuits to optimize neural network operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional convolutional neural network operations are used, then floating-point and double-precision convolutional operations can be performed, but memory capacity requirements and energy consumption increase significantly
Solution Approach 1:
The patent converts continuous floating-point values into discrete three-state values (-1, 0, 1) by changing the parameter representation. This discretization reduces memory capacity requirements and simplifies computational operations while maintaining sufficient precision for neural network training and recognition tasks
Solution Approach 2:
The patent replaces traditional electronic computing operations with spintronic device operations. By using spin current to manipulate magnetic domain walls in the magnetic tunnel junction, the system performs computational operations through physical magnetic state transitions rather than electronic calculations, significantly reducing energy consumption
2Measurement precision
If traditional convolutional neural network operations are used, then floating-point and double-precision convolutional operations can be performed, but training duration increases to several days or weeks
Solution Approach 1:
By transforming the parameter representation from continuous floating-point to discrete three-state values, the patent enables parallel processing operations that can be executed simultaneously in spintronic devices, dramatically reducing training time from days or weeks to much shorter durations
Solution Approach 2:
The patent segments the computational process into discrete three-state operations that can be independently processed and combined. This segmentation allows for efficient parallel processing in the magnetic tunnel junction array, reducing overall training duration
3Quantity of substance
If binary neural network is used to optimize memory space, then operation complexity increases with Boolean logical and shift operations
Solution Approach 1:
Instead of using binary representation (-1, 1) as in conventional BNN, the patent inverts the approach by using ternary representation (-1, 0, 1). This inversion allows the system to maintain reduced memory capacity while avoiding the complexity of Boolean logical operations, as ternary values can be directly represented by magnetic domain wall positions
4Reliability
If external magnetic fields are used for resistance state switching, then switching can be achieved, but energy consumption increases
Solution Approach 1:
The patent replaces external magnetic field application with spin current injection. By utilizing the spin Hall effect in the spin-orbit coupling layer, spin current is generated that exerts torque on the magnetic domain wall, enabling resistance state switching through electrical current rather than external magnetic fields, thus reducing energy consumption
Solution Approach 2:
The spin-orbit coupling layer acts as an intermediary that converts electrical current into spin current, which then manipulates the magnetic domain wall in the ferromagnetic free layer. This intermediary mechanism enables efficient state switching without requiring external magnetic fields
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 three-state spintronic device enables efficient resistance state switching without external magnetic fields, reducing energy consumption and memory requirements, thereby enhancing the training and recognition efficiency of neural networks.
Implementation Method 1
a current pulse is injected into the spin-orbit coupling layer, and a spin current is generated to drive a magnetic domain wall in the ferromagnetic free layer to move
Implementation Method 2
a spin-orbit coupling layer
Implementation Method 3
drive a magnetic domain wall in the ferromagnetic free layer to move, so that a resistance state switching is performed
Implementation Method 4
a magnetic tunnel junction
Data Source
AI summary
The three-state spintronic device includes: a bottom electrode, a magnetic tunnel junction and a top electrode from bottom to top. The magnetic tunnel junction includes: a spin-orbit coupling layer, a ferromagnetic free layer, a barrier tunneling layer, a ferromagnetic reference layer, three local magnetic domain wall pinning centers and domain wall nucleation centers. An antisymmetric exchange interaction is modulated, and the magnetic domain wall pinning centers are embedded in an interface between a heavy metal and the ferromagnetic free layer. The magnetic domain wall nucleation centers are at two ends of the ferromagnetic free layer. A current pulse flows through the spin-orbit coupling layer to generate a spin current and the spin current is injected into the ferromagnetic free layer. Under a control of all-electrical controlled, an effective field of a spin-orbit torque drives domain wall to move and displace.