Four-Terminal RPU Crossbar Array for Linear Programming
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current cross-bar array systems with two-terminal resistive processing units (RPU) cells suffer from leakage drift, limited array size, and high power consumption, which affect the accuracy and performance of artificial neural network (ANN) applications, particularly over time, and are plagued by non-linearity issues during programming.
Innovation Solution
The implementation of a cross-bar array system using multi-terminal RPU cells, where each RPU cell is configured as a four-terminal device with multiple rails for improved linearity and reduced drift, allowing for enhanced data storage and processing capabilities, including matrix inversion and decomposition, by incorporating separate write and read lines made of materials with different resistivity levels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If two-terminal resistive processing units are used in cross-bar arrays, then device simplicity and manufacturing ease are improved, but leakage drift occurs and programming linearity deteriorates
Solution Approach 1:
The patent divides the traditional two-terminal RPU into four separate terminals (row write, row read, column write, column read), allowing independent control and measurement of current and voltage at each terminal. This segmentation enables separate optimization of write and read paths, eliminating cross-talk and improving programming linearity while maintaining manufacturing feasibility through standard cross-bar fabrication processes.
Solution Approach 2:
The patent introduces separate read lines as intermediary elements that do not participate in the write operation. These read lines serve as mediators to measure voltage or current without interfering with the write process, enabling accurate measurement and linear programming control while isolating the write path from read disturbances.
2Device complexity
If two-terminal RPU cells are used, then device complexity is reduced, but cross-talk between lines increases and programming accuracy deteriorates
Solution Approach 1:
By segmenting the terminals into independent row write, row read, column write, and column read lines, the patent eliminates electrical cross-talk between write and read paths. Each terminal can be independently controlled and measured, achieving precise programming accuracy while the overall device architecture remains relatively simple through systematic terminal division.
Solution Approach 2:
Separate read lines act as intermediary elements that measure electrical parameters without affecting the write operation. This intermediary approach isolates measurement functions from actuation functions, preventing cross-talk and improving programming accuracy without significantly increasing device complexity.
3Area of stationary object
If conventional cross-bar arrays are used, then array size is limited, but scaling to larger arrays increases leakage drift and power consumption
Solution Approach 1:
The patent segments the array into modular units with independent four-terminal RPU cells, enabling scalable expansion without proportionally increasing leakage drift. Each terminal can be independently controlled, allowing selective activation of array regions and reducing overall power consumption while supporting larger array sizes for complex ANN computations.
Solution Approach 2:
Separate read lines serve as intermediaries that enable efficient readout operations without requiring full array activation. This allows selective reading of specific array elements, reducing the total current drawn from power supply and minimizing power consumption in large-scale arrays while maintaining the ability to process complex computations.
4Manufacturing precision
If multi-terminal RPU cells are implemented, then programming linearity and reduced drift are improved, but device complexity increases
Solution Approach 1:
The patent segments the RPU into four independently controllable terminals, which systematically improves programming linearity by eliminating cross-talk effects. While the terminal count increases from two to four, the segmentation follows a regular pattern that can be systematically implemented in fabrication and controlled in operation, managing the added complexity through structural regularity.
Solution Approach 2:
The four-terminal RPU cell design provides multi-functionality, where the same physical device can perform both write and read operations through different terminal combinations. This universal design allows a single device structure to serve multiple functions (programming, reading, measuring), reducing the need for separate components and managing overall system complexity while achieving improved programming linearity.
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 configuration minimizes cross-talk, enhances programming linearity, and reduces power consumption, leading to improved performance and efficiency in ANN applications, enabling faster training and broader practical applications with lower resource requirements.
Implementation Method 1
RPU cells are considered as a promising technology for electronic synapse devices or memristors for neuromorphic computing... an RPU cell can be used as a connection (synapse) between a pre-neuron and post-neuron, representing the connection weight in the form of device resistance
Data Source
AI summary
A multiterminal non-volatile memory cross-bar array system includes a set of conductive row rails, a set of conductive column rails configured to form a plurality of crosspoints at intersections between the conductive rails and the conductive column rails and a resistive processing unit at each of the crosspoints each representing a neuron in a neural network. At least one given conductive row rail includes first and second row lines is in contact with a given resistive processing unit. At least one given conductive column rail including first and second column lines is in contact with the given resistive processing unit.