3D Oscillator Structure for Neural Network Density
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Oscillator neural networks (ONNs) face challenges due to the large footprint and complex wiring requirements of CMOS oscillators, which necessitate a high number of oscillators and complicate interconnection, making them unsuitable for compact and efficient neural network implementations.
Innovation Solution
The use of three-dimensional structures with oxide-based oscillators, including nonlinear hysteretic elements and resistors coupled in series, allows for a high density of oscillators in a small area, facilitating easier interconnection and reducing wire length, thereby improving circuit performance and energy efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If CMOS oscillators are used in ONN, then the oscillators can be integrated with standard CMOS technology, but the footprint becomes very large due to the need for hundreds of oscillators and multiple interconnecting wires
Solution Approach 1:
The patent transitions from two-dimensional planar oscillators to three-dimensional vertically-coupled oscillators. Multiple oscillators are stacked along the vertical dimension and coupled through vertical interconnects, enabling hundreds of oscillators to be packed into a compact footprint while maintaining CMOS compatibility
Solution Approach 2:
The patent implements a hierarchical structure where oscillators are nested in vertical stacks, with each oscillator containing hysteretic elements and resonant circuits. These nested oscillators are further organized into arrays coupled through vertical interconnects, achieving high density integration
2Productivity
If hundreds of CMOS oscillators are used, then the neural network can achieve sufficient computational capability, but the wiring complexity increases significantly making routing difficult
Solution Approach 1:
The patent introduces vertical coupling between oscillators using vertical interconnects, transforming the wiring architecture from planar to three-dimensional. This reduces the number of horizontal wire crossings and simplifies routing while maintaining connections between hundreds of oscillators
Solution Approach 2:
The patent divides the large array of oscillators into smaller vertical stacks or groups that are coupled through shared vertical interconnects. This segmentation reduces the overall wiring complexity by localizing connections within each stack while using vertical interconnects to bridge between stacks
3Quantity of substance
If oxide-based oscillators with nonlinear hysteretic elements are used, then the oscillator density increases in a small area, but the device structure becomes more complex
Solution Approach 1:
The patent utilizes materials exhibiting insulator-metal transition (IMT) with sharp resistance changes at transition temperatures. By controlling the resistance state of hysteretic elements through voltage or temperature, the oscillators achieve high density packing while the nonlinear characteristics enable compact resonant circuit design
Solution Approach 2:
The patent employs composite structures combining oxide-based hysteretic elements with resonant circuits and vertical interconnects. The hysteretic elements use materials like VO2 or other oxides that exhibit nonlinear resistance characteristics, integrated with standard CMOS components to achieve both high density and functionality
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 approach enables the creation of smaller, more efficient ONNs with shorter wires and improved delay and energy use, suitable for dense neural network implementations.
Implementation Method 1
an insulator-metal transition (IMT) structure electrically coupled with the first column, wherein the capacitor structure is electrically positioned between the IMT structure and the first column
Implementation Method 2
The oscillator may include a nonlinear hysteretic element and a resistor coupled together in series
Data Source
AI summary
Embodiments may relate to a structure to be used in a neural network. A first column and a second column, both of which are to couple with a substrate. A capacitor structure may be electrically coupled with the first column. An insulator-metal transition (IMT) structure may be coupled with the first column such that the capacitor structure is electrically positioned between the IMT structure and the first column. A resistor structure may further be electrically coupled with the IMT structure and the second column such that the resistor structure is electrically positioned between the second column and the IMT structure. Other embodiments may be described or claimed.