Resistive Memory Crossbar Calibration for Matrix-Vector Multiplication
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing cognitive computing architectures, such as Von Neumann architecture, are inefficient for high-speed data processing required in cognitive computers, particularly for matrix-vector multiplications due to inherent device conductance variations over time, leading to systematic errors.
Innovation Solution
A memory crossbar array with programmable resistive memory elements that performs a calibration procedure to compensate for conductance variations, applying a constant calibration voltage and reading current values to derive an estimation of conductance variation parameters, enabling accurate matrix-vector multiplications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If resistive memory elements are used for matrix-vector multiplication, then computation speed and energy efficiency are improved, but conductance variations over time cause systematic errors
Solution Approach 1:
The patent applies preliminary action by performing a calibration procedure before the actual matrix-vector multiplication. During calibration, the system measures the conductance of each memory element and stores calibration factors. These pre-measured factors are then used to correct the computation results, thereby compensating for conductance variations before they affect the accuracy of the main computation.
Solution Approach 2:
The patent implements feedback by using the measured conductance values from calibration to adjust and correct the computation results. The system reads back the actual conductance of each memory element, compares it with the expected value, and applies correction factors to the final result, creating a closed-loop system that compensates for device variations.
2Measurement precision
If conductance variations are compensated through calibration, then accuracy is improved, but additional calibration steps increase device complexity
Solution Approach 1:
The patent applies self-service by designing a calibration procedure that uses the memory crossbar array itself to perform the calibration measurements. The system uses standard voltage sources and current measurements that are already available in the memory interface, eliminating the need for external specialized calibration equipment and reducing overall system complexity.
Solution Approach 2:
The patent changes the operating parameters by applying known test voltages to the memory elements and measuring the resulting currents to determine conductance values. By varying the voltage levels in a controlled manner during calibration, the system extracts conductance information without requiring additional hardware, thus managing complexity while achieving precise measurements.
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 allows for fast, low-power, and scalable matrix-vector multiplications with high accuracy, reducing systematic errors and improving performance compared to conventional approaches, and can be generalized for matrix-matrix and vector-vector multiplications.
Implementation Method 1
Each junction comprises a programmable resistive memory element
Implementation Method 2
a readout circuit configured to apply read voltages to the row lines of the memory crossbar array and to read out current values of column lines
Data Source
AI summary
The invention is notably directed to a device for performing a matrix-vector multiplication of a matrix with a vector. The device comprises a memory crossbar array comprising of row lines, of columns lines and of junctions arranged between the row lines and the column lines. Each junction comprises a programmable resistive memory element. The device comprises a signal generator and a readout circuit. The device is configured to perform a calibration procedure to compensate for conductance variations of the resistive memory elements. The calibration procedure is configured to program a calibration subset of the plurality of resistive memory elements to initial conductance values and to apply a constant calibration voltage to the row lines of the calibration subset. The device is configured to read calibration current values of the column lines of the calibration subset and to derive an estimation of a conductance variation parameter from the calibration current values.