3D Vertical Cross-Point Synaptic Weight Device
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current technologies face challenges in fabricating multi-level synaptic weight devices with a 16 or greater bit-width required for neural networks, as they struggle to find materials with resistance values that change accordingly, limiting the implementation of such devices in AI systems like image recognition systems.
Innovation Solution
A multi-level synaptic weight device with a 3D vertical cross-point structure is developed, incorporating a control peripheral circuit, a cross-point block, and resistive switching layers, where the conductance is adjusted by selector switches and resistors, allowing for a rewritable conductive path that responds to voltage, utilizing materials like NiOx and SiO2 for resistive switching, enabling a sufficient bit-width for synaptic weights.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a resistive switching material with multiple resistance values is used to fabricate a multi-level synaptic weight device with 16 or greater bit-width, then the synaptic weight precision is improved, but the ease of manufacture deteriorates due to limitations and difficulties in finding such materials
Solution Approach 1:
The synaptic weight device is segmented into multiple conductive paths with different resistance values. Instead of relying on a single material with continuous resistance modulation, the invention divides the resistance control into discrete segments (first, second, third conductive paths) that can be independently controlled by selector switches, achieving 16-bit precision through combination of segmented elements
Solution Approach 2:
The invention transitions from a planar structure to a three-dimensional vertical cross-point structure. Multiple horizontal conductive lines are stacked vertically with vertical conductive lines intersecting them, creating multiple cross-point nodes in the vertical dimension. This 3D architecture enables more conductive paths to be packed into a smaller footprint, facilitating higher bit-width implementation
2Measurement precision
If a 3D vertical cross-point structure is implemented to achieve sufficient bit-width for synaptic weights, then the synaptic weight precision is improved, but the device complexity increases
Solution Approach 1:
The selector switches in the invention serve multiple functions: they act as selection elements for choosing specific conductive paths, as control elements for enabling/disabling current flow, and as part of the overall weight adjustment mechanism. This multi-functionality reduces the need for separate dedicated components, thereby managing device complexity while achieving high precision
Solution Approach 2:
The 3D vertical cross-point structure employs a nested arrangement where horizontal conductive lines are stacked in layers, with each layer containing multiple horizontal lines. Vertical conductive lines penetrate through multiple horizontal layers, creating a nested configuration where conductive paths are embedded within the 3D structure. This nesting enables compact integration of multiple conductive paths
3Adaptability or versatility
If multiple conductive paths with different resistance values are created using selector switches and resistors, then the conductance adjustment capability is improved, but the manufacturing precision requirements increase
Solution Approach 1:
The invention achieves conductance adjustment by changing the configuration of conductive paths rather than relying on precise control of material resistance parameters. By using selector switches to enable/disable specific paths and by combining multiple paths with different fixed resistance values, the system achieves fine-grained conductance control through topological reconfiguration rather than material parameter tuning
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 solution allows for a scalable and efficient synaptic weight device capable of adjusting conductance levels, overcoming the limitations of existing technologies by providing a sufficient bit-width for neural networks, enhancing the performance of AI systems like image recognition.
Implementation Method 1
first and second resistive switching layers interposed between the horizontal stack structure and the vertical array structure, respectively, each comprising a plurality of cross-point nodes, wherein each of the plurality of cross-point nodes is a conductive path depending on a weight assigned to the horizontal conductive line
Data Source
AI summary
A multi-level synaptic weight device having a 3D vertical cross-point structure according to an embodiment includes a multi-level conductance structure configured to couple any one of multiple word lines to any one of multiple bit lines. The conductance structure may include a multiplexer, configured to include multiple selector switches coupled in parallel to the word line and to select any one of the multiple parallel-coupled selector switches in response to an externally applied selection signal; a fixed resistor block including multiple fixed resistors coupled to the bit line; a cross-point block configured such that the multiple selector switches in the multiplexer and the multiple fixed resistors in the fixed resistor block intersect in a matrix form so as to be coupled to each other and each of the selector switches in the multiplexer has a unique number of cross-points; and a conductive plate on the fixed resistor block.