Spin Orbit Torque Processing-in-Memory Device for Energy Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional computer structures face energy inefficiency and operational speed constraints due to the memory bottleneck in von-Neumann architecture, particularly for applications like artificial neural networks that require frequent memory access for MAC operations, and existing resistive memory-based processing-in-memory (PIM) devices face issues with area size and power consumption related to analog-to-digital conversion.
Innovation Solution
A spin orbit torque device-based processing-in-memory device that employs digital logic gates using current switching and voltage-controlled magnetic anisotropy, reducing data migration by performing MAC operations within memory through a weight storage array, product operation array, and sum operation array connected via a calculation line, replacing analog MAC operations with digital logic operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If conventional von-Neumann architecture is used, then device complexity is reduced, but energy efficiency deteriorates due to frequent data migration between memory and ALU
Solution Approach 1:
The patent merges the memory function and arithmetic logic function into a single integrated device. The memory cell includes both storage elements (first and second memory cells) and arithmetic logic elements (first and second logic gates) within the same cell structure, enabling simultaneous storage and computation operations to eliminate data migration between separate memory and ALU components.
Solution Approach 2:
The memory cell is designed to perform multiple functions: it can store data in the first and second memory cells, perform logical operations through the first logic gate (AND, OR, NOT operations), perform arithmetic operations through the second logic gate (addition, subtraction, multiplication, division), and output results through first and second output terminals. This multi-functional design eliminates the need for separate memory and ALU units.
2Productivity
If resistive memory-based PIM with analog MAC operation is used, then productivity is improved, but area size and power consumption increase due to required ADC
Solution Approach 1:
The patent extracts and removes the analog-to-digital converter (ADC) component from the system by implementing digital logic operations directly within the memory cell. Instead of performing analog MAC operations that require external ADCs for digital output, the invention uses digital logic gates (AND, OR, NOT) and arithmetic logic gates to perform computations entirely in the digital domain, eliminating the need for separate ADC hardware and reducing overall device area.
3Productivity
If resistive memory-based PIM with analog MAC operation is used, then productivity is improved, but use of energy increases due to required ADC
Solution Approach 1:
The patent extracts and removes the analog-to-digital converter (ADC) component from the system by implementing digital logic operations directly within the memory cell. Instead of performing analog MAC operations that require external ADCs for digital output, the invention uses digital logic gates (AND, OR, NOT) and arithmetic logic gates to perform computations entirely in the digital domain, eliminating the need for separate ADC hardware and reducing overall device area.
4Reliability
If weight storage array with MTJ device is used, then reliability is improved, but device complexity increases due to multiple arrays and calculation lines
Solution Approach 1:
The patent merges the memory function and arithmetic logic function into a single integrated device. The memory cell includes both storage elements (first and second memory cells) and arithmetic logic elements (first and second logic gates) within the same cell structure, enabling simultaneous storage and computation operations to eliminate data migration between separate memory and ALU components.
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 system energy efficiency, reduces area size, and minimizes energy consumption by eliminating the need for an analog-to-digital converter, while increasing power efficiency and area efficiency for computational loads.
Implementation Method 1
a magnetic tunnel junction (MTJ) device having resistance determined depending on a magnetization direction
Implementation Method 2
spin orbit torque device capable of being used as a memory element
Data Source
AI summary
Embodiments of the present disclosure provide a spin orbit torque device-based processing-in-memory device that includes a digital logic gate using a current switching and voltage controlled magnetic anisotropy (VCMA) effect of a spin orbit torque device capable of being used as a memory element, and increases the overall system energy efficiency by designing a memory capable of performing an MAC operation to reduce the number of data migration between a memory and an operator.