Memory Hub Chiplet Architecture with Interposer D2D Links

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Integrating multiple functions into a single chip increases wire count, power consumption, heat generation, manufacturing costs, and complexity, limiting bandwidth and causing inefficiencies in semiconductor devices.

Innovation Solution

A semiconductor device with a chiplet structure using a die-to-die (D2D) interface and micro bumps on an interposer to connect individual dies, reducing space and latency, and enabling high-bandwidth data transmission.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If multiple functions are integrated into a single chip, then functionality is improved, but chip area and wire count increase

Engineering Contradiction:

ImprovefunctionalityVSAvoidchip area

Core Design Contradiction:

Adaptability or versatilityVSArea of stationary object

Solution Approach 1:

The patent divides the semiconductor device into multiple independent chiplets (processor chiplet, memory chiplet, I/O chiplet) that are fabricated separately on different substrates and then interconnected through an interposer. This segmentation allows each chiplet to be optimized for its specific function without requiring a large single chip area, while still achieving high-level integration and functionality.

Inventive Principle:

Principle #1Segmentation

2Adaptability or versatility

If multiple functions are integrated into a single chip, then functionality is improved, but power consumption increases

Engineering Contradiction:

ImprovefunctionalityVSAvoidpower consumption

Core Design Contradiction:

Adaptability or versatilityVSUse of energy by moving object

Solution Approach 1:

By segmenting the device into separate chiplets with dedicated functions (processor, memory, I/O), each chiplet can be independently optimized for power efficiency. The interposer enables direct electrical connections that reduce transmission losses compared to traditional PCB-based interconnections, thereby managing power consumption more effectively despite the multi-functional integration.

Inventive Principle:

Principle #1Segmentation

3Adaptability or versatility

If multiple functions are integrated into a single chip, then functionality is improved, but heat generation increases

Engineering Contradiction:

ImprovefunctionalityVSAvoidheat generation

Core Design Contradiction:

Adaptability or versatilityVSTemperature

Solution Approach 1:

The patent distributes heat-generating functions across multiple separate chiplets mounted on an interposer, allowing for better thermal management. Each chiplet can have its own thermal pathways and cooling strategies, preventing the concentration of heat in a single large chip. The interposer structure facilitates heat dissipation through its substrate material and mounting configuration.

Inventive Principle:

Principle #1Segmentation

4Device complexity

If a single chip with memory and interface is fabricated, then integration is improved, but manufacturing cost increases

Engineering Contradiction:

ImproveintegrationVSAvoidmanufacturing cost

Core Design Contradiction:

Device complexityVSEase of manufacture

Solution Approach 1:

The patent fabricates different chiplets using optimized process nodes suitable for each specific function (e.g., advanced nodes for processor, different nodes for memory and I/O). This allows each chiplet to be manufactured at its optimal cost point rather than requiring a single complex process for all functions. The interposer provides a standardized platform for integrating these separately-fabricated chiplets, simplifying the overall manufacturing workflow.

Inventive Principle:

Principle #1Segmentation

5Reliability

If conventional bumping process is used, then connection is achieved, but bandwidth per unit area is limited

Engineering Contradiction:

ImproveconnectionVSAvoidbandwidth per unit area

Core Design Contradiction:

ReliabilityVSArea of stationary object

Solution Approach 1:

The patent introduces an interposer as an intermediary substrate between the chiplets and the external environment. The interposer contains a high-density interconnect structure with numerous fine-pitch ball grid arrays (BGAs) that provide multiple parallel data pathways. This intermediary structure enables significantly higher bandwidth per unit area compared to conventional direct bumping, as the interposer can accommodate many more interconnection points in a compact area.

Inventive Principle:

Principle #24Intermediary (Mediator)

6Reliability

If data path length increases, then connection is achieved, but transmission speed decreases

Engineering Contradiction:

ImproveconnectionVSAvoidtransmission speed

Core Design Contradiction:

ReliabilityVSSpeed

Solution Approach 1:

The interposer serves as a high-performance intermediary that provides short, controlled-impedance transmission paths between chiplets. By using the interposer's integrated interconnect structure rather than external PCB traces, the patent minimizes signal path length and reduces transmission delays. The interposer's construction enables high-speed differential signaling with minimal signal degradation, maintaining high transmission speeds despite the distributed chiplet architecture.

Inventive Principle:

Principle #24Intermediary (Mediator)

Data Source

PatentUS12418986B1Semiconductor device including a memory hub with chiplet structure

Publication Date: 2025.09.16 COSIGNON

AI summary

Disclosed is a semiconductor device including a memory hub with a chiplet structure. The semiconductor device includes an interposer, a processor unit placed on the interposer, at least one memory hub placed on the interposer and provided to be physically separated from the processor unit, and at least one memory provided to be physically separated from the memory hub.