3D Memory Cell Layout for Low-Power Neuromorphic Computing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing neuromorphic devices face challenges in integrating memory cells efficiently while reducing power consumption, as they often require significant current flow and high power usage for operations.

Innovation Solution

The semiconductor device arranges memory cells in a three-dimensional grid, storing weights corresponding to one layer of a neural network in memory cells at the same height, allowing for simultaneous activation of a portion of these cells and deactivation of others, thereby reducing current flow and power consumption.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Use of energy by stationary object

If memory cells are arranged in three-dimensional grid and weights are stored in cells at same height, then integration is improved and power consumption is reduced, but device complexity increases

Engineering Contradiction:
Improvepower consumptionVSAvoiddevice complexity
Core Design Contradiction:
Use of energy by stationary objectVSDevice complexity

Solution Approach 1:

The patent applies three-dimensional stacking of memory cells in the vertical direction (third direction perpendicular to substrate) to increase storage capacity without expanding planar area. Memory cells are arranged in multiple stacks at different heights, enabling spatial optimization and reduced power consumption by activating only relevant stacks during operations.

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

Solution Approach 2:

The patent divides the neural network weights into multiple segments stored in different memory cell stacks at various heights. Each stack can be independently activated or deactivated, allowing selective operation of portions of the neural network to reduce power consumption while maintaining overall functionality.

Inventive Principle:
Principle #1Segmentation

2Area of stationary object

If memory cells are arranged in three-dimensional grid and weights are stored in cells at same height, then integration is improved, but device complexity increases

Engineering Contradiction:
ImproveintegrationVSAvoiddevice complexity
Core Design Contradiction:
Area of stationary objectVSDevice complexity

Solution Approach 1:

The patent utilizes vertical stacking of memory cells in the third direction to achieve three-dimensional integration. Multiple memory cell stacks are positioned at different heights above the substrate, enabling high-density storage without increasing the planar footprint, thus improving integration efficiency.

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

Solution Approach 2:

The patent implements nested arrangement where multiple memory cell stacks are vertically positioned within the same planar footprint. Each stack contains multiple memory cells arranged in series along the vertical direction, creating a compact nested structure that maximizes space utilization.

Inventive Principle:
Principle #7Nested doll (Nesting)

3Use of energy by stationary object

If current flow is minimized by selective activation, then power consumption is reduced, but operation speed may be affected

Engineering Contradiction:
Improvepower consumptionVSAvoidoperation speed
Core Design Contradiction:
Use of energy by stationary objectVSSpeed

Solution Approach 1:

The patent activates only the necessary portion of memory cell stacks corresponding to the current neural network layer being processed, rather than activating all stacks simultaneously. This partial action reduces power consumption by minimizing current flow through inactive stacks while maintaining operation speed for the active portion.

Inventive Principle:
Principle #16Partial or excessive action

Solution Approach 2:

The patent employs periodic activation of different memory cell stacks corresponding to different neural network layers. During inference operations, stacks are activated in sequence or in parallel groups, allowing power consumption to be reduced by deactivating unused stacks while maintaining overall processing throughput.

Inventive Principle:
Principle #19Periodic action

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 configuration enhances integration and reduces power consumption by minimizing current flow through bit and source lines, enabling efficient neuromorphic computing operations.

Implementation Method 1

a ferroelectric layer and a gate electrode layer surrounding the channel layer

Methodology Applied
Scientific EffectFerroelectricity:

Data Source

PatentUS20250299731A1Semiconductor device
Publication Date: 2025.09.25 SAMSUNG ELECTRONICS CO LTD
  • US20250299731A1 patent drawing
  • US20250299731A1 patent drawing
  • US20250299731A1 patent drawing

AI summary

A semiconductor device may include: a cell area in which a plurality of memory cells arranged in a first direction, a second direction and a third direction are disposed, the first direction and the second direction being parallel to an upper surface of a substrate and intersecting, and the third direction being perpendicular to the upper surface of the substrate; and a peripheral circuit area in which a word line driver connected to the plurality of memory cells through a plurality of word lines, a sense amplifier circuit connected to the plurality of memory cells through a plurality of bit lines, and a source line driver connected to the plurality of memory cells through a plurality of source lines, are disposed.