Multi-Layer Memory Buffer Synchronization for Latency and Power Reduction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

High-performance computing memory systems face issues with memory access latency and power consumption due to point-to-point network connections in cascaded memory buffer device networks, which degrade signal propagation and increase energy requirements.

Innovation Solution

A memory system architecture with synchronized memory buffer devices in multiple layers, utilizing conductive pathways for memory signals, where each layer has reduced outbound links to subsequent layers, and multiple inbound links to the controller, optimizing signal transmission and reducing power consumption by halving the number of output drivers and link complexity.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If point-to-point network connections are used in cascaded memory buffer device networks, then memory access latency increases, but connection reliability is improved

Engineering Contradiction:
Improveconnection reliabilityVSAvoidmemory access latency
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

Multiple memory buffer devices in the same memory layer are merged into a single synchronized group that receives outbound links from the memory controller simultaneously. This consolidation reduces the number of propagation paths and allows parallel signal distribution to multiple buffers, thereby reducing memory access latency while maintaining connection reliability through synchronization mechanisms.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent introduces a new dimensional organization where memory buffer devices are arranged in multiple layers with synchronized groups receiving signals simultaneously. By adding the layer dimension and using parallel outbound links to multiple buffers in each layer, the system reduces the sequential propagation delay inherent in traditional cascaded point-to-point connections.

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

2Stability of the object's composition

If point-to-point network connections are used in cascaded memory buffer device networks, then connection stability is improved, but power consumption increases

Engineering Contradiction:
Improveconnection stabilityVSAvoidpower consumption
Core Design Contradiction:
Stability of the object's compositionVSUse of energy by moving object

Solution Approach 1:

Multiple output drivers across different memory buffer devices are merged into a single synchronized output stage that drives multiple memory layers simultaneously. This consolidation reduces the total number of output drivers needed, thereby reducing power consumption while maintaining connection stability through precise synchronization of the merged output signals to multiple layers.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

A single synchronized output driver group serves multiple memory layers simultaneously, making the output stage universal rather than dedicated to individual buffers. This multi-functionality reduces the total number of drivers required across the system, lowering overall power consumption while maintaining stable connections to all layers through coordinated signal distribution.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Loss of time

If synchronized memory buffer devices in multiple layers are used, then memory access latency is reduced, but device complexity increases

Engineering Contradiction:
Improvememory access latencyVSAvoiddevice complexity
Core Design Contradiction:
Loss of timeVSDevice complexity

Solution Approach 1:

Memory buffer devices perform preliminary synchronization actions by aligning their signal reception and processing timing before actual memory access operations. This pre-synchronization establishes a coordinated timing framework that enables reduced latency during actual operations without requiring complex real-time coordination mechanisms, thereby managing device complexity while achieving latency reduction.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system uses feedback mechanisms where memory buffer devices monitor and adjust their synchronization status based on signal timing from the memory controller and from each other. This feedback-based synchronization maintains precise timing coordination across multiple layers, enabling reduced memory access latency while managing device complexity through adaptive rather than purely rigid control.

Inventive Principle:
Principle #23Feedback

4Use of energy by moving object

If synchronized memory buffer devices in multiple layers are used, then power consumption is reduced, but link complexity increases

Engineering Contradiction:
Improvepower consumptionVSAvoidlink complexity
Core Design Contradiction:
Use of energy by moving objectVSDevice complexity

Solution Approach 1:

The patent organizes memory buffer devices in a multi-layer dimensional structure where synchronized groups communicate vertically across layers rather than sequentially through cascaded connections. This dimensional reorganization reduces the number of active output drivers needed (lowering power consumption) while concentrating link complexity into standardized inter-layer connections that can be managed through uniform synchronization protocols.

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

Data Source

PatentEP2021933B1Memory systems for automated computing machinery
Publication Date: 2013.09.04 INTERNATIONAL BUSINESS MACHINE CORPORATION
  • EP2021933B1 patent drawingFigure 1
  • EP2021933B1 patent drawingFigure 2
  • EP2021933B1 patent drawingFigure 3

AI summary

Memory systems are disclosed that include a memory controller and an outbound link with the memory controller connected to the outbound link. The outbound link typically includes a number of conductive pathways that conduct memory signals from the memory controller to memory buffer devices in a first memory layer; and at least two memory buffer devices in a first memory layer. Each memory buffer device in the first memory layer typically is connected to the outbound link to receive memory signals from the memory controller.