Compensation for Conductance Drift in Analog Memory Crossbar Arrays
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Analog memory-based artificial neural networks face accuracy degradation due to conductance drift in resistive elements over time, leading to unstable resistivity and conductivity.
Innovation Solution
A method and system that compensate for conductance drift by determining a function that maps output activation vectors at a later time to those at an initial time, and applying this function to subsequent output activations, using a crossbar array with memory devices at each crosspoint storing synaptic weights.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If analog memory-based neural networks are used, then computational efficiency and energy savings are improved, but inference accuracy degrades over time due to conductance drift in resistive elements
Solution Approach 1:
The patent implements a feedback mechanism where output activation vectors are periodically read from the crossbar array, a mapping function is determined based on drift observations, and this function is applied to compensate for conductance drift in subsequent computations, thereby maintaining inference accuracy while retaining the energy efficiency of analog memory-based processing
Solution Approach 2:
The patent changes the parameter of output activation vectors by applying a compensation function that maps them from the current time state to the initial time state, effectively adjusting the parameters to counteract conductance drift and maintain accurate inference
2Productivity
If resistive elements are used for storing synaptic weights, then device density and integration are improved, but stability of resistivity and conductivity deteriorates over time
Solution Approach 1:
The system continuously monitors the output activation vectors and uses feedback to determine drift characteristics, allowing the system to adapt and compensate for the unstable resistivity of resistive elements while maintaining high device density
Solution Approach 2:
The patent performs preliminary actions by reading output activation vectors at multiple time points to determine the mapping function before it affects subsequent computations, allowing compensation to be applied proactively to maintain stability
Data Source
AI summary
A system can compensate for activation drift in analog memory-based artificial neural networks. A set of input activation vectors can be input, at a first point in time, to a crossbar array. The first set of output activation vectors can be read from the output lines of the crossbar array. At a second point in time, which is a later time than the first point in time, the input set of activation vectors can be input to the crossbar array. A second set of output activation vectors can be read from the crossbar array. A function that maps the second set of output activation vectors to the first set of output activation vectors can be determined. The function can be applied to subsequent output activation vectors output by the crossbar array. A method thereof, can also be provided.