3D Embedded Ferroelectric Memory Stacks for High-Bandwidth ICs

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional embedded DRAM and FeDRAM architectures face challenges in increasing on-chip bit-cell count and fabrication complexity, particularly in integrating multiple levels of memory arrays with host logic circuitry, leading to difficulties in achieving high bandwidth and low signal latency communication.

Innovation Solution

The integration of multiple levels of embedded memory arrays with ferroelectric capacitors and thin film access transistors, utilizing monolithic fabrication and direct bonding techniques, allows for the formation of composite IC structures with enhanced memory array levels, including peripheral CMOS circuitry and control circuitry, enabling efficient communication through a chip bus.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If conventional embedded DRAM architecture is used, then fabrication process is simpler, but memory density and bandwidth are limited

Engineering Contradiction:
Improvememory densityVSAvoidfabrication complexity
Core Design Contradiction:
Quantity of substanceVSDevice complexity

Solution Approach 1:

The patent transitions from planar 2D memory architecture to three-dimensional stacked architecture with multiple memory levels (Level 1, Level 2, etc.) vertically stacked above the substrate. This vertical stacking enables significantly higher memory density by utilizing the third dimension (height) rather than only expanding in the planar direction, directly resolving the contradiction between memory density and fabrication complexity.

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

Solution Approach 2:

The patent implements a nested structure where multiple memory levels are stacked vertically, with each level containing bit cells, word lines, and bit lines that are nested within the three-dimensional space. The Level 2 memory is positioned above Level 1 memory, creating a nested hierarchical structure that maximizes space utilization and achieves high density while maintaining organized fabrication processes.

Inventive Principle:
Principle #7Nested doll (Nesting)

2Quantity of substance

If more memory levels are stacked, then memory density increases, but signal latency and communication time increase

Engineering Contradiction:
Improvememory densityVSAvoidsignal latency
Core Design Contradiction:
Quantity of substanceVSLoss of time

Solution Approach 1:

The patent segments the memory system into multiple independent levels (Level 1, Level 2, etc.), each with its own array of bit cells, word lines, and bit lines. This segmentation allows parallel access to different memory levels, reducing the overall access time compared to a single large memory block. The segmented architecture enables simultaneous operations across levels, mitigating the latency increase that would otherwise result from stacking.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

By organizing memory in vertical stacks with multiple levels accessed through three-dimensional interconnects, the patent enables short signal paths between memory cells and control circuitry. The vertical stacking with direct inter-level connections reduces the physical distance signals must travel compared to planar expansion, thereby reducing signal latency despite increased memory density.

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

3Ease of manufacture

If planar expansion is used, then manufacturing is easier, but bandwidth and communication efficiency are limited

Engineering Contradiction:
Improvemanufacturing easeVSAvoidbandwidth
Core Design Contradiction:
Ease of manufactureVSProductivity

Solution Approach 1:

The patent achieves high bandwidth by stacking multiple memory levels vertically and implementing three-dimensional interconnect structures that provide multiple parallel data paths between memory levels and control circuitry. This vertical architecture with multi-layer interconnects enables simultaneous data transfer across multiple levels, dramatically increasing bandwidth compared to planar expansion while using standard semiconductor fabrication processes.

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

Solution Approach 2:

The patent employs universal fabrication processes from the backend-of-line (BEOL) to create both the memory structures and interconnect layers. The same fabrication techniques used for planar interconnects are applied to create vertical vias and stacked structures, making the manufacturing process scalable and compatible with existing semiconductor manufacturing capabilities, thus maintaining ease of manufacture while achieving high bandwidth.

Inventive Principle:
Principle #6Universality (Multi-functionality)

4Quantity of substance

If vertical stacking is implemented, then memory density increases, but fabrication precision requirements increase

Engineering Contradiction:
Improvememory densityVSAvoidalignment precision
Core Design Contradiction:
Quantity of substanceVSManufacturing precision

Solution Approach 1:

The patent implements vertical stacking with precise alignment between memory levels using backend-of-line fabrication processes that create vertically aligned vias and interconnect structures. The fabrication process utilizes sequential deposition and etching steps that inherently maintain vertical alignment, achieving the required manufacturing precision for stacked memory while enabling high density through the third dimension.

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

Solution Approach 2:

The patent employs planarization techniques and conformal deposition processes that create uniform, flat surfaces at each memory level, ensuring equipotential conditions for subsequent fabrication steps. This approach maintains consistent alignment and spacing between stacked levels, reducing the impact of surface variations and achieving the necessary manufacturing precision for vertical stacking.

Inventive Principle:
Principle #12Equipotentiality

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 enhances memory density and performance by allowing higher retention rates and lower leakage currents, facilitating higher bandwidth and lower signal latency communication between memory devices and processors.

Implementation Method 1

each capacitor comprises a ferroelectric material. An embedded memory architecture may then rely on polarization states of a capacitor's ferroelectric material, which can be changed when an electric field applied across the capacitor conductors is of correct polarity and sufficient strength to alter the semi-permanent dipoles within the ferroelectric material.

Methodology Applied
Scientific EffectFerroelectric polarization:

Implementation Method 2

A transistor and a capacitor of each 1T-1C cell may be electrically coupled through one or more metal interconnect layers formed in the back-end-of-line (BEOL) over logic circuitry formed in the front-end-of-line (FEOL).

Methodology Applied
Scientific EffectElectrical conduction: Conduction (electrical)

Data Source

PatentUS12604481B2IC's with multple levels of embedded memory
Publication Date: 2026.04.14 INTEL CORP
  • US12604481B2 patent drawing
  • US12604481B2 patent drawing
  • US12604481B2 patent drawing

AI summary

Integrated circuits with embedded memory having multiple levels. Each memory array level includes ferroelectric capacitors coupled to an array of thin film access transistors according to a 1T-1F or 1T-many F bit-cell architecture. The levels of embedded memory are monolithically fabricated, one over the other, or after monolithically fabricating one level of embedded memory in a host IC structure, a second IC structure with another level of memory array is directly bonded to a front or backside of the host IC structure in a face-to-face or face-to-back orientation. The second IC structure may include additional peripheral CMOS circuitry, such as sense amps or decoders, or not.