Resistive Cross-Point Array Precision via Switchable Path Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Resistive cross point arrays face challenges in representing large numbers of dictionary weights with precision due to intrinsic process variations in resistance values, leading to unacceptable errors in neuro-inspired machine learning algorithms.
Innovation Solution
A resistive memory system with a cross point resistive network and switchable paths, where variable resistive elements and conductive lines form a cross point array, allowing subsets of resistive elements to provide combined variable conductances, averaging out process variations and enhancing precision in representing dictionary weights.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If resistive devices are provided with multi-level resistive states to represent dictionary weights, then the storage capacity is improved, but manufacturing precision deteriorates due to intrinsic process variations in resistance values
Solution Approach 1:
The patent divides the representation of each dictionary weight into multiple resistive elements (e.g., multiple devices in parallel or series combinations). Instead of relying on a single resistive element to precisely represent a weight value, the system segments the representation across multiple elements whose combined resistance provides the desired precision while tolerating individual element variations.
Solution Approach 2:
The patent combines multiple resistive elements to form a single logical weight representation. By merging multiple devices with inherently varying resistance values into a composite structure, the system achieves more stable and precise effective resistance values that are less sensitive to process variations in individual elements.
2Quantity of substance
If large numbers of dictionary weights are represented with a large resistive cross point array, then the storage capacity is improved, but measurement precision deteriorates due to process variations
Solution Approach 1:
The patent segments the large resistive array into multiple smaller sub-arrays or groups, where each group represents a portion of the dictionary weights. This segmentation allows for independent optimization and measurement of each group, reducing the cumulative impact of process variations across the entire large array.
Solution Approach 2:
The patent combines measurement results from multiple smaller resistive groups to achieve the final weight representation. By measuring and aggregating results from multiple smaller units rather than attempting to measure a single large array, the system achieves higher precision through statistical averaging of process variations.
3Device complexity
If intrinsic process variations in resistance values are tolerated, then device complexity is reduced, but reliability deteriorates due to significant loss of precision
Solution Approach 1:
The patent combines multiple resistive elements into composite weight representations, where the collective behavior of the merged elements provides stable and reliable weight values. This merging approach maintains simplicity at the element level while achieving reliability through the statistical properties of the combined structure.
Solution Approach 2:
The patent changes the effective parameters of the resistive network by using combinations of elements rather than individual elements. By altering how resistance values are aggregated (through series/parallel combinations), the system achieves better precision and reliability without fundamentally changing the simple resistive device structure itself.
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 more precise implementation of learning algorithms by reducing the impact of process variations, improving the accuracy of matrix operations and weight updates in neuro-inspired machine learning, while maintaining energy, area, and latency efficiency.
Implementation Method 1
intrinsic process variations in the resistance values of the resistive devices can result in significant loss of precision
Data Source
AI summary
This disclosure relates generally to resistive memory systems. The resistive memory systems may be utilized to implement neuro-inspired learning algorithms with full parallelism. In one embodiment, a resistive memory system includes a cross point resistive network and switchable paths. The cross point resistive network includes variable resistive elements and conductive lines. The conductive lines are coupled to the variable resistive elements such that the conductive lines and the variable resistive elements form the cross point resistive network. The switchable paths are connected to the conductive lines so that the switchable paths are operable to selectively interconnect groups of the conductive lines such that subsets of the variable resistive elements each provide a combined variable conductance. With multiple resistive elements in the subsets, process variations in the conductances of the resistive elements average out. As such, learning algorithms may be implemented with greater precision using the cross point resistive network.