Variable Resistance Storage Elements for AI Weight Coefficient Management
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional neural network computing circuits face inefficiencies in training processes, particularly when updating connection weight coefficients after product shipment, as they either require multiple steps for writing conductance or struggle with initial setting accuracy due to varying write properties.
Innovation Solution
An artificial intelligence processing device employing two types of variable-resistance nonvolatile storage elements with different properties, where the first element sets connection weight coefficients with high accuracy at product shipment and the second element updates coefficients efficiently post-shipment by varying conductance through successive voltage pulses.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If conventional neural network computing circuits use variable-resistance nonvolatile storage elements for updating connection weight coefficients after product shipment, then training capability is enabled, but write operation requires multiple steps and initial setting accuracy deteriorates
Solution Approach 1:
The patent divides the storage elements into two distinct types: first variable-resistance nonvolatile storage elements for initial setting and second variable-resistance nonvolatile storage elements for updating. This segmentation allows each type to be optimized for its specific function, resolving the contradiction between training capability and initial setting accuracy.
Solution Approach 2:
Different properties are assigned to different storage elements based on their specific functions. First storage elements have properties optimized for accurate initial setting, while second storage elements have properties optimized for efficient updating during training. This local differentiation of properties resolves the contradiction.
2Device complexity
If single type of variable-resistance nonvolatile storage element is used for both initial setting and updating, then device complexity is reduced, but operational efficiency deteriorates due to inability to optimize for different operations
Solution Approach 1:
The patent segments the storage elements into two types with different properties, allowing each to be optimized for its specific operation (initial setting vs. updating). This segmentation improves operational efficiency despite the increase in device complexity.
Solution Approach 2:
Different local qualities (properties) are provided to different storage elements based on functional requirements. This allows the system to achieve high operational efficiency by having storage elements tailored to their specific purposes.
3Manufacturing precision
If write operation is performed based on conductance itself, then initial setting can be done accurately, but training efficiency after product shipment deteriorates due to multiple write steps required
Solution Approach 1:
The patent segments write operations into two distinct modes: initial setting using first storage elements and training updates using second storage elements. This segmentation allows each mode to use the most efficient write method for its purpose, resolving the contradiction between initial setting accuracy and training efficiency.
Solution Approach 2:
The system dynamically selects different storage elements and write methods based on the operational phase (initial setting vs. training). This dynamic adaptation allows the system to optimize for different requirements at different times.
4Adaptability or versatility
If multiple write steps are used for updating conductance, then training can be performed, but loss of time increases due to sequential write operations
Solution Approach 1:
The patent segments the write operation into a single-step process using second variable-resistance nonvolatile storage elements specifically designed for efficient updates. This eliminates the need for multiple sequential write steps, reducing time loss while maintaining training capability.
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
Enables both accurate initial setting and efficient updating of connection weight coefficients, enhancing the operational efficiency of AI processing devices by utilizing the distinct properties of the storage elements for precise and rapid conductance adjustments.
Implementation Method 1
variable-resistance nonvolatile storage elements each having a resistance that varies according to a given electrical signal
Implementation Method 2
a first variable-resistance nonvolatile storage element and a second variable-resistance nonvolatile storage element that hold, as conductance, a connection weight coefficient
Implementation Method 3
when successive applications of a voltage pulse with a same polarity and a same voltage are made, a proportion of an amount of change in the conductance caused by a second application of the voltage pulse relative to an amount of change in the conductance caused by a first application of the voltage pulse in the first variable-resistance nonvolatile storage element is less than a proportion of an amount of change in the conductance caused by a second application of the voltage pulse relative to an amount of change in the conductance caused by a first application of the voltage pulse in the second variable-resistance nonvolatile storage element
Data Source
AI summary
An artificial intelligence processing device includes: a first variable-resistance nonvolatile storage element and a second variable-resistance nonvolatile storage element having different properties and provided on a single substrate. When successive applications of a voltage pulse with a same polarity and a same voltage are made, a proportion of an amount of change in conductance caused by a second application of the voltage pulse relative to an amount of change in conductance caused by a first application of the voltage pulse in the first variable-resistance nonvolatile storage element is less than a proportion of an amount of change in conductance caused by a second application of the voltage pulse relative to an amount of change in conductance caused by a first application of the voltage pulse in the second variable-resistance nonvolatile storage element.