Affine Correction for Conductance Drift in Neuromorphic Crossbar Arrays
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Neuromorphic systems based on crossbar array structures, particularly those using phase-change memory (PCM) devices, face errors due to temporal conductance variations, which affect the accuracy of deep neural network (DNN) inference and training.
Innovation Solution
A method involving a correction unit that applies an affine transformation to output currents from a crossbar array structure to compensate for temporal conductance variations, using programmable parameters such as a multiplicative coefficient and an additive parameter, integrated within the neuromorphic system to maintain accuracy over time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If PCM devices are used in crossbar array structures for neuromorphic computing, then computational memory functionality is achieved, but temporal conductance variations cause errors in computations
Solution Approach 1:
The patent implements feedback mechanisms where the system monitors temporal conductance variations in PCM devices and dynamically adjusts correction parameters. The correction unit receives feedback about drift magnitude and direction, then applies appropriate affine transformation parameters to compensate for the variations, creating a closed-loop system that maintains computation accuracy despite device degradation over time
Solution Approach 2:
The patent changes the parameters of the correction unit (affine transformation parameters) to adapt to temporal conductance variations. By adjusting these parameters based on observed drift characteristics, the system compensates for conductance changes without requiring physical replacement of PCM devices or fundamental redesign of the crossbar architecture
2Reliability
If correction units are added to compensate for conductance drift, then computation accuracy is improved, but device complexity increases
Solution Approach 1:
The patent segments the correction function into modular correction units that can be independently implemented and configured. Each correction unit handles specific aspects of drift compensation, allowing the system to add only the necessary correction capability without redesigning the entire neuromorphic system. This modular approach limits the increase in overall system complexity
Solution Approach 2:
The correction unit acts as an intermediary component between the crossbar array and the computational logic. It receives output currents from the crossbar, applies affine transformations to compensate for drift, and delivers corrected signals to the computational units. This intermediary approach isolates the complexity of drift compensation from both the memory array and the computational logic
Data Source
AI summary
A method of operating a neuromorphic system is provided. The method includes applying voltage signals across input lines of a crossbar array structure, the crossbar array structure including rows and columns interconnected at junctions via programmable electronic devices, the rows including the input lines for applying voltage signals across the electronic devices and the columns including output lines for outputting currents. The method also includes correcting, via a correction unit connected to the output lines, each of the output currents obtained at the output lines according to an affine transformation to compensate for temporal conductance variations in the electronic devices.