Computational Memory Programming Through Iterative Conductance Adaptation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing in-memory computing technologies face challenges in achieving accurate matrix-vector multiplication due to the stochastic switching behavior of memristive devices, leading to computational imprecision.
Innovation Solution
An iterative process is employed to adapt the conductance values of memristive devices in a computational memory based on measured results, using machine learning or multivariate linear regression to minimize errors until an accuracy condition is met, thereby improving the representation of matrix elements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Use of energy by moving object
If memristive devices are used for in-memory computing, then high density and low power consumption are achieved, but computational precision deteriorates due to stochastic switching behavior
Solution Approach 1:
The patent implements an iterative feedback process where the computational memory performs computation tasks using stored elements, measures the results, and adapts the stored elements based on measurement errors. This closed-loop feedback mechanism continuously refines the conductance values to compensate for stochastic behavior, thereby maintaining computational precision while utilizing the energy-efficient memristive devices.
Solution Approach 2:
The patent dynamically adjusts the conductance parameters of memristive devices through iterative adaptation. By changing the physical state (conductance value) of the devices based on measured computational errors, the system compensates for stochastic switching variations and achieves precise matrix-vector multiplication results.
2Measurement precision
If iterative adaptation process is applied to improve accuracy, then computational precision is improved, but processing time increases
Solution Approach 1:
The patent performs partial adaptation by selectively updating only those conductance values that contribute most to computational error, rather than uniformly adjusting all elements. This approach achieves sufficient accuracy with fewer iterative steps, reducing the time penalty associated with iterative refinement.
3Manufacturing precision
If conductance values are adapted iteratively, then representation accuracy of matrix elements is improved, but device complexity increases
Solution Approach 1:
The computational memory performs self-adaptation by using its own computational results to guide the adjustment of its stored elements. The system measures its own computational errors and autonomously adjusts conductance values without requiring external calibration equipment or complex programming infrastructure, thereby reducing overall system complexity.
Data Source
AI summary
Provided is a method, device, and computer program product for programming a set of first elements onto a computational memory. The computational memory allows for performing a computation task from a set of second elements that encode the set of first elements in the computational memory, respectively. The method includes performing the computation task by the computational memory using the set of second elements and adapting at least part of the set of second elements in the computational memory based on a measured result of the computation task, until the measured result of the computation task fulfils an accuracy condition.