Staggered 3D Crossbar Synapse Circuits for Spike-Timing Dependent Plasticity

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current neuromorphic and synaptronic systems fail to effectively mimic the spike-timing dependent plasticity (STDP) mechanism found in biological brains, limiting their ability to efficiently process and integrate spatiotemporal patterns in sensory inputs.

Innovation Solution

A neuromorphic and synaptronic system is designed with a crossbar array network of electronic neurons, featuring a staggered layout of synapse devices that implement STDP through a method where spiking signals from neurons trigger changes in synaptic conductance based on the relative timing of pre- and post-synaptic neuron firings, utilizing variable state resistors or memory elements at cross-point junctions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If conventional synaptic computing circuits are used, then the system can perform basic neuromorphic computation, but the system fails to effectively mimic STDP mechanism and efficiently process spatiotemporal patterns

Engineering Contradiction:
ImproveSTDP mechanism implementationVSAvoidspatiotemporal pattern processing efficiency
Core Design Contradiction:
Adaptability or versatilityVSProductivity

Solution Approach 1:

The patent transitions from conventional planar crossbar architectures to a three-dimensional stacked crossbar architecture with staggered patterns. Multiple crossbar arrays are stacked vertically with offset positioning, enabling signals to propagate through multiple layers and实现 complex temporal sequences and spike timing dependencies that are essential for STDP mechanisms

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

Solution Approach 2:

The system divides the neural network into multiple segmented crossbar arrays stacked in three dimensions. Each crossbar array represents a distinct computational layer with specific neurons and synapses, allowing independent optimization of STDP learning rules in each layer while maintaining overall system functionality

Inventive Principle:
Principle #1Segmentation

2Ease of operation

If traditional digital models are used for neuromorphic computing, then sequential processing of 0s and 1s is achieved, but parallel and distributed processing analogous to biological brains is lost

Engineering Contradiction:
Improveparallel distributed processing capabilityVSAvoidconnection architecture complexity
Core Design Contradiction:
Ease of operationVSDevice complexity

Solution Approach 1:

The patent merges multiple crossbar arrays into a unified three-dimensional structure where axons and dendrites from different layers interact through staggered positioning. This integration enables parallel signal propagation across multiple layers simultaneously, achieving biologically-inspired distributed processing while maintaining a structured connection architecture

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

By stacking crossbar arrays in the vertical dimension with staggered horizontal positioning, the system creates a three-dimensional connection topology that naturally supports parallel processing. Multiple neural pathways can operate simultaneously across different layers without interfering with each other, enabling efficient parallel distributed computation

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

3Adaptability or versatility

If synapse devices are arranged in a dense grid pattern, then high connectivity is achieved, but power and space requirements increase

Engineering Contradiction:
Improveneural network connectivityVSAvoidpower consumption
Core Design Contradiction:
Adaptability or versatilityVSUse of energy by stationary object

Solution Approach 1:

The staggered three-dimensional arrangement allows synapse devices to be distributed across multiple vertical layers with offset horizontal positions. This spatial distribution reduces the density of any single layer, lowering local power consumption and heat generation while maintaining high overall connectivity through inter-layer connections

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

Solution Approach 2:

The neural network connectivity is segmented across multiple stacked crossbar arrays rather than concentrated in a single dense grid. Each layer handles a portion of the total connectivity requirements, distributing the power and space demands across multiple physical locations and reducing the burden on any single device

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP2616996B1Compact cognitive synaptic computing circuits
Publication Date: 2017.09.27 INTERNATIONAL BUSINESS MACHINE CORPORATION
  • EP2616996B1 patent drawingFigure 1A
  • EP2616996B1 patent drawingFigure 1B
  • EP2616996B1 patent drawingFigure 2A

AI summary

Embodiments of the invention relate to producing spike-timing dependent plasticity using electronic neurons interconnected in a crossbar array network. The crossbar array network comprises a plurality of crossbar arrays. Each crossbar array comprises a plurality of axons and a plurality of dendrites such that the axons and dendrites are transverse to one another, and multiple synapse devices, wherein each synapse device is at a cross-point junction of the crossbar array coupled between a dendrite and an axon. The crossbar arrays are spatially in a staggered pattern providing a staggered crossbar layout of the synapse devices.