Bonded FPGA-DRAM Structure With 3D Vertical Interconnects

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Field-programmable gate arrays (FPGAs) face limitations in cost and working frequency due to large chip area consumption and signal transfer delays, such as RC delay from metal routing, which restrict their performance in applications like deep neural networks.

Innovation Solution

A semiconductor device with a programmable logic device core and cache integrated on a bonded chip, featuring a large number of short-distance vertical metal interconnects instead of conventional long-distance metal routing, to achieve higher working frequency, wider data bandwidth, and lower power consumption.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Speed

If conventional long-distance metal routing is used in FPGAs, then chip area can be reduced, but signal transfer delay increases and working frequency decreases

Engineering Contradiction:
Improvesignal transfer speedVSAvoidchip area
Core Design Contradiction:
SpeedVSArea of stationary object

Solution Approach 1:

The patent transitions from planar (2D) metal routing to three-dimensional (3D) vertical interconnects through wafer bonding. By stacking multiple semiconductor wafers and creating through-wafer vias, the interconnect structure moves into the third dimension, enabling short-distance vertical connections that replace long-distance lateral routing paths.

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

Solution Approach 2:

The patent divides a single large FPGA chip into multiple smaller semiconductor wafers that are bonded together. Each wafer contains specific logic functions or memory blocks, and the segmentation allows vertical interconnection between these segments, reducing the routing distance within each segment while maintaining overall system functionality.

Inventive Principle:
Principle #1Segmentation

2Loss of time

If FPGAs use large chip area for routing, then manufacturing cost increases, but signal transfer delay also increases

Engineering Contradiction:
Improvesignal transfer delayVSAvoidmanufacturing cost
Core Design Contradiction:
Loss of timeVSEase of manufacture

Solution Approach 1:

The patent employs 3D vertical interconnects through wafer bonding to replace extended 2D metal routing. This dimensional transition creates direct vertical pathways between functional blocks on different wafers, dramatically reducing signal transfer delay without requiring additional lateral routing area that would increase manufacturing cost.

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

Solution Approach 2:

The patent merges multiple semiconductor wafers into a single bonded chip structure through wafer bonding technology. This combining approach integrates logic functions and memory blocks in close proximity with direct vertical connections, reducing the need for extensive metal routing and associated manufacturing costs while improving signal transfer performance.

Inventive Principle:
Principle #5Merging (Combining)

3Productivity

If conventional metal routing is used, then device complexity is reduced, but data bandwidth and working frequency are limited

Engineering Contradiction:
Improvedata bandwidthVSAvoidinterconnect structure complexity
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent introduces 3D vertical interconnects that penetrate through multiple wafer layers, creating high-density connection pathways. This dimensional advancement enables parallel data transmission through multiple via layers simultaneously, significantly increasing data bandwidth despite the increased structural complexity of the bonded wafer assembly.

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

Solution Approach 2:

The patent implements a nested interconnect structure where through-wafer vias are embedded within the bonded wafer stack. Multiple layers of interconnects are nested within each other, with lower-layer vias contained within the overall structure of upper-layer vias, enabling high-density routing with organized complexity that supports increased data bandwidth.

Inventive Principle:
Principle #7Nested doll (Nesting)

Data Source

PatentEP3891797B1Bonded semiconductor devices having programmable logic device and dynamic random-access memory and methods for forming the same
Publication Date: 2024.04.10 YANGTZE MEMORY TECH CO LTD
  • EP3891797B1 patent drawingFigure 1A~1B
  • EP3891797B1 patent drawingFigure 2A~2B
  • EP3891797B1 patent drawingFigure 3A~3B

AI summary

Embodiments of semiconductor devices and fabrication methods thereof are disclosed. In an example, a semiconductor device includes a first semiconductor structure including a programmable logic device, an array of static random-access memory (SRAM) cells, and a first bonding layer including a plurality of first bonding contacts. The semiconductor device also includes a second semiconductor structure including an array of dynamic random-access memory (DRAM) cells and a second bonding layer including a plurality of second bonding contacts. The semiconductor device further includes a bonding interface between the first bonding layer and the second bonding layer. The first bonding contacts are in contact with the second bonding contacts at the bonding interface.