Interconnect Die Stacking for High-Bandwidth Embedded Memory Access

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Accessing memory in computing systems is slow due to the separation of processor and memory dies, which limits computational performance.

Innovation Solution

An interconnect die is introduced between logic dies stacked over a base die with embedded memory, enabling hybrid bonding and direct, high-speed access to the embedded memory through a configuration with tighter metal pitch and more vertical interconnects.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If processor and memory are separated onto different dies, then integration density is improved, but memory access speed deteriorates

Engineering Contradiction:
Improveintegration densityVSAvoidmemory access speed
Core Design Contradiction:
Quantity of substanceVSSpeed

Solution Approach 1:

The patent transitions from planar interconnection to three-dimensional stacked architecture, placing memory and processor on different vertical levels connected through vertical interconnects (TSVs). This dimensional change enables both high integration density through stacking and fast access through direct vertical pathways, resolving the contradiction between density and speed.

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

Solution Approach 2:

The patent introduces an interconnect die with buffer memory as an intermediary component between the processor die and embedded memory. This intermediary enables faster access by providing local caching and reducing the critical path delay, while maintaining the benefits of separated processor and memory dies for high integration density.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Speed

If processor and memory are integrated on the same die, then memory access speed is improved, but integration density deteriorates

Engineering Contradiction:
Improvememory access speedVSAvoidintegration density
Core Design Contradiction:
SpeedVSQuantity of substance

Solution Approach 1:

The patent segments the system into multiple specialized dies (processor die, memory die, interconnect die) stacked vertically. Each die is optimized for its specific function, allowing the memory die to be accessed rapidly through direct vertical interconnects while the processor die maintains high computational density. This segmentation resolves the contradiction by achieving both fast access and high overall integration density through 3D stacking.

Inventive Principle:
Principle #1Segmentation

3Quantity of substance

If hybrid bonding is used to stack dies, then interconnect density is improved, but manufacturing complexity increases

Engineering Contradiction:
Improveinterconnect densityVSAvoidmanufacturing complexity
Core Design Contradiction:
Quantity of substanceVSEase of manufacture

Solution Approach 1:

The patent employs preliminary alignment marks and bonding pads designed specifically for hybrid bonding processes. These pre-configured features enable precise alignment and robust bonding between stacked dies, reducing manufacturing complexity despite the advanced bonding technique required to achieve high interconnect density.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentUS20250309211A1Interconnect die between logic dies stacked over a base die with embedded memory
Publication Date: 2025.10.02 INTEL CORP
  • US20250309211A1 patent drawing
  • US20250309211A1 patent drawing
  • US20250309211A1 patent drawing

AI summary

Examples of integrated circuit (IC) devices including an interconnect die between logic dies stacked over a base die with embedded memory may enable ultra-high bandwidth memory access. In one example, an IC device includes an interposer (e.g., a die or base die) including embedded memory, a first die (for example, a die including home agent logic) over the interposer, where the first die is hybrid bonded with the interposer. The IC device includes a second die (e.g., a chiplet, compute die, etc.) over the first die. The IC device includes a third die (e.g., an interconnect die) between the first die and the second die, where the third die includes a plurality of interconnect layers and is hybrid bonded with the first die and the second die. In one example, devices in device regions of the first die and second die are coupled via conductive interconnects in the interconnect die.