Resistive Memory Device Group Programming for Weight Storage
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current memory devices with resistive elements face challenges such as high programming power requirements, noise, and resistance drift, which hinder efficient performance in cognitive computing applications like memcomputing and matrix-vector multiplications.
Innovation Solution
A memory device comprising a plurality of resistive elements and a control unit that programs single weights by groups of at least two resistive elements, using a combination of single shot and iterative programming schemes to minimize drift and noise, and perform matrix-vector multiplications with high accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If single resistive elements are programmed individually, then programming precision can be controlled, but programming power consumption increases and programming time extends
Solution Approach 1:
The patent combines multiple resistive elements into groups where each group represents a single weight. By programming groups of resistive elements simultaneously rather than individually, the patent reduces total programming power consumption while maintaining acceptable precision through the collective behavior of the grouped elements.
Solution Approach 2:
The patent segments the weight representation into multiple resistive elements within each group, where each element contributes a portion of the total weight value. This segmentation allows parallel programming of multiple elements to achieve the same weight with reduced power consumption compared to programming a single element to the full weight value.
2Productivity
If groups of resistive elements are used to represent single weights, then programming power is reduced and programming speed increases, but noise and resistance drift increase
Solution Approach 1:
The patent implements iterative programming schemes that use feedback mechanisms to adjust the resistance values of resistive elements in groups. By measuring the actual resistance after programming and comparing it with target values, the system can compensate for noise and drift effects, improving reliability while maintaining the speed benefits of group-based programming.
Solution Approach 2:
The patent uses combinations of single-shot programming (for speed) and iterative programming (for precision) to achieve the desired balance. By applying partial programming actions in sequence rather than requiring full precision in a single step, the system achieves both fast programming and reduced noise/drift effects.
3Speed
If high programming power is applied to achieve fast programming, then programming speed increases, but noise and resistance drift worsen
Solution Approach 1:
The patent employs iterative programming schemes that apply programming signals in periodic cycles rather than as a single high-power pulse. This periodic action allows the system to achieve the desired programming speed through multiple lower-power cycles, reducing the harmful effects of high power application such as noise and resistance drift.
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 enables significant power and speed advantages over conventional Von-Neumann architectures, allowing for efficient inexact fast matrix multiplication with high accuracy, suitable for algorithms that do not require high precision, and reduces noise and drift in resistive memory devices.
Implementation Method 1
PCM is a non-volatile solid-state memory technology that exploits the reversible, thermally-assisted switching of phase-change materials, in particular chalcogenide compounds such as GST (Germanium-Antimony-Tellurium), between states with different electrical resistance
Implementation Method 2
In a write operation, the resulting programming signal causes Joule heating of the phase-change material to an appropriate temperature to induce the desired cell-state on cooling
Data Source
AI summary
A memory device may include a plurality of resistive elements and a control unit for controlling the memory device. The memory device is configured to program single weights of the memory device by groups of at least two resistive elements. A related method and a related computer program product may be also provided.