Integrated Neuron Circuit Using VO2 NDR Devices

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

There is a lack of a practical and foundry-compatible integrated neuron circuit structure and manufacturing process for artificial spiking neurons that can be scaled down to small areas while maintaining energy efficiency and biological plausibility, as existing solutions have not demonstrated a detailed IC structure or layout compatible with CMOS fabrication.

Innovation Solution

The integration of passive thin-film resistors, Metal-Insulator-Metal (MIM) capacitors, and active vanadium dioxide (VO2) Negative Differential Resistance (NDR) devices forms a compact, scalable artificial spiking neuron circuit, compatible with BEOL processes, allowing for ultra-dense memristive neuromorphic networks that can be fabricated using a CMOS-compatible process flow, with VO2 being replaced by other thermally-driven insulator-to-metal transition materials.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If discrete devices and breadboard level connections are used to implement neuron circuits, then biological plausibility and voltage-controlled ion channel emulation are achieved, but integration density and manufacturability are poor

Engineering Contradiction:
Improvebiological plausibilityVSAvoidintegration density
Core Design Contradiction:
ReliabilityVSEase of manufacture

Solution Approach 1:

The patent merges multiple discrete components (resistors, capacitors, VO2 devices) into a single integrated neuron circuit structure that can be fabricated using CMOS-compatible processes. The coupled relaxation oscillators are integrated on a single chip, eliminating breadboard connections and achieving both biological plausibility and high integration density.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent transitions from discrete 2D breadboard layouts to 3D integrated circuit architecture with multiple metal layers and vertical stacking. This dimensional transition enables ultra-dense neuromorphic networks while maintaining the biological plausibility of the neuron circuit functionality.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Ease of manufacture

If conventional CMOS processes are used for neuron circuit fabrication, then manufacturing compatibility is achieved, but area efficiency and scalability are limited

Engineering Contradiction:
ImproveCMOS compatibilityVSAvoidcircuit area
Core Design Contradiction:
Ease of manufactureVSArea of moving object

Solution Approach 1:

The patent utilizes vertical stacking with multiple metal layers (M1, M2, M3) and through-silicon vias to achieve 3D integration. This allows the neuron circuit to be fabricated using standard CMOS processes while dramatically reducing the footprint area compared to planar layouts.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Solution Approach 2:

The patent changes the fabrication parameters by integrating VO2 devices and thin-film resistors/capacitors into the CMOS process flow, specifically using BEOL (Back-End-Of-Line) compatible materials and techniques. This enables area-efficient fabrication without requiring changes to the core CMOS manufacturing process.

Inventive Principle:
Principle #35Parameter changes

3Productivity

If transistor-based neuron circuits are used, then computational capability is achieved, but energy consumption and device complexity increase

Engineering Contradiction:
Improvecomputational capabilityVSAvoidenergy efficiency
Core Design Contradiction:
ProductivityVSUse of energy by moving object

Solution Approach 1:

The patent extracts the transistor component from the neuron circuit implementation, replacing it with passive VO2-based relaxation oscillators. This removal of active transistors significantly reduces energy consumption and device complexity while maintaining the computational capability of spiking neuron behavior through the intrinsic dynamics of the coupled oscillators.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The neuron circuit uses self-oscillating VO2 devices that generate spiking behavior intrinsically without requiring external transistor control. The coupled relaxation oscillators self-regulate their firing patterns based on membrane potential dynamics, achieving computational capability with minimal energy input and no active transistor switching.

Inventive Principle:
Principle #25Self-service

4Manufacturing precision

If detailed IC structure and layout are provided for foundry fabrication, then manufacturing precision is improved, but design complexity and process overhead increase

Engineering Contradiction:
Improvefabrication accuracyVSAvoiddesign complexity
Core Design Contradiction:
Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent designs a universal neuron circuit module with standardized layout and interconnection patterns that can be replicated and scaled. The coupled oscillator structure uses identical VO2 devices and passive components arranged in a repeatable pattern, enabling high manufacturing precision through standardization while reducing overall design complexity through modularity.

Inventive Principle:
Principle #6Universality (Multi-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 energy-efficient, transistorless neuromorphic networks that can execute computationally intensive algorithms in size, weight, and power-constrained platforms, such as autonomous vehicles, by achieving a high effective area density and low dynamic spike energy, making them biologically competitive in terms of size and energy efficiency.

Implementation Method 1

active vanadium dioxide (VO2) Negative Differential Resistance (NDR) devices... VO2 being replaced by other thermally-driven insulator-to-metal transition materials

Methodology Applied
Scientific EffectInsulator-to-metal transition: Phase Change

Data Source

PatentEP3574525B1A scalable, stackable, and BEOL-process compatible integrated neuron circuit
Publication Date: 2023.08.09 HRL LAB
  • EP3574525B1 patent drawingFigure 1
  • EP3574525B1 patent drawingFigure 2A~2B
  • EP3574525B1 patent drawingFigure 3

AI summary

An integrated neuron circuit structure comprising at least one thin-film resistor, one Metal Insulator Metal capacitor and one Negative Differential Resistance device.