Series-Connected MTJ Stacks for Spiking Neural Network Neuron Circuits
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing structures and methods for forming multiple magnetic-tunnel-junction layer stacks in semiconductor devices, particularly for spiking neural networks, lack efficiency and effectiveness in achieving improved spike rates and control variability.
Innovation Solution
A structure comprising a series connection of multiple magnetic-tunneling-junction layer stacks connected to a pulsed power supply, which enhances the spike rate and control variability by utilizing the additive property of voltage spikes and discrete probability distributions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple magnetic-tunnel-junction layer stacks are connected in series, then the spike rate and control variability are improved, but the device complexity increases
Solution Approach 1:
The patent divides the neural network into multiple discrete magnetic-tunnel-junction layer stacks connected in series, where each stack represents a segmented neuron or synapse. This segmentation allows independent control and optimization of each unit while collectively achieving improved spike rate and variability control through the series connection architecture.
Solution Approach 2:
The series-connected magnetic-tunnel-junction layer stacks serve multiple functions simultaneously: they act as individual neurons, synapses, and collectively as a neural network processing unit. The pulsed power supply enables these stacks to perform both storage and computation functions, reducing the need for separate dedicated components.
2Duration of action of stationary object
If a pulsed power supply is used to increase spike rate, then the endurance is improved, but the energy consumption increases
Solution Approach 1:
The patent employs periodic pulsed power supply to drive the magnetic-tunnel-junction layer stacks, where pulses are applied at specific intervals to induce spin-torque and achieve magnetization switching. This periodic action enables precise control of spike timing and rate while allowing the system to return to a low-power state between pulses, improving endurance through controlled energy delivery.
Solution Approach 2:
The patent utilizes parameter changes in the pulsed power supply, specifically varying pulse width, amplitude, and frequency, to optimize the balance between energy consumption and endurance. By adjusting these parameters, the system can achieve high spike rates when needed while consuming minimal energy during idle periods, thereby improving overall endurance.
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 solution increases the spike rate and improves control over variability in neuron circuits for spiking neural networks, while also increasing endurance through the use of a pulsed power supply.
Implementation Method 1
magnetic-tunnel-junction layer stacks
Implementation Method 2
additive property of voltage spikes
Data Source
AI summary
Structures including multiple magnetic-tunnel-junction layer stacks and methods of forming such structures. The structure comprises a first magnetic-tunneling-junction layer stack, a second magnetic-tunneling-junction layer stack connected in a series connection to the first magnetic-tunneling-junction layer stack, and a pulsed power supply connected to the first and second magnetic-tunneling-junction layer stacks.