TSN-Connected Heterogeneous Accelerators for Deterministic Task Timing

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing computing systems face challenges in managing access to accelerators for multiplication and accumulation operations over time sensitive networking buses, leading to delays and variations in data delivery that affect synchronization in applications like audio and video streaming, due to varying latency and energy consumption among heterogeneous accelerators.

Innovation Solution

A manager dynamically allocates virtual channels over a time sensitive networking bus to satisfy timing requirements and optimize energy performance by balancing workloads across heterogeneous accelerators, using techniques such as memristor crossbars and microring resonators, and adjusting virtual channel allocations based on urgency levels and predicted task patterns.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If heterogeneous accelerators are connected via time sensitive networking bus, then computing speed for multiplication and accumulation operations is improved, but latency variation and synchronization difficulty worsen

Engineering Contradiction:

Improvecomputing speedVSAvoidlatency variation

Core Design Contradiction:

ProductivityVSLoss of time

Solution Approach 1:

The system dynamically adjusts virtual channel allocations based on real-time workload characteristics and urgency levels. The manager monitors accelerator utilization, task deadlines, and resource availability to adaptively reconfigure channel assignments, allowing the system to optimize for either speed or latency determinism depending on current conditions

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The system changes operational parameters including virtual channel bandwidth allocations, priority levels, and scheduling policies based on workload type. Different parameter configurations are applied for different task urgencies and accelerator capabilities to balance computing speed with latency requirements

Inventive Principle:

Principle #35Parameter changes

2Reliability

If virtual channels are dynamically allocated to satisfy timing requirements, then synchronization reliability is improved, but system complexity increases

Engineering Contradiction:

Improvesynchronization reliabilityVSAvoidsystem complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

A manager component is introduced as an intermediary between applications and accelerators to handle the complexity of virtual channel allocation, scheduling, and coordination. This centralized controller abstracts the complexity from individual components while ensuring deterministic timing through coordinated resource management

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The manager component performs multiple functions including virtual channel allocation, task scheduling, workload balancing, and timing monitoring within a single system element. This multi-functional approach consolidates complexity into one coordinated unit rather than requiring separate mechanisms for each function

Inventive Principle:

Principle #6Universality (Multi-functionality)

3Use of energy by moving object

If workloads are balanced across heterogeneous accelerators, then energy consumption is optimized, but access coordination complexity increases

Engineering Contradiction:

Improveenergy consumptionVSAvoidaccess coordination complexity

Core Design Contradiction:

Use of energy by moving objectVSDevice complexity

Solution Approach 1:

The manager implements feedback mechanisms that monitor accelerator utilization, energy consumption, and task completion status in real-time. Based on this feedback, the system dynamically adjusts workload distribution to balance energy consumption across accelerators while adapting to changing system conditions and resource availability

Inventive Principle:

Principle #23Feedback

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

Ensures deterministic and timely data delivery while optimizing energy consumption by dynamically reallocating resources to meet the timing requirements of computing tasks, improving overall system performance and reducing delays in multimedia applications.

Implementation Method 1

Time sensitive networking includes techniques for time synchronization among devices involved in communications over a network, techniques for scheduling and traffic shaping, and techniques for selection of communication paths, path reservations and fault-tolerance

Methodology Applied

Scientific EffectTime Sensitive Networking:

Implementation Method 2

photonic accelerators have been developed to use phenomenon in optical domain to obtain computing results corresponding to multiplication and accumulation

Methodology Applied

Scientific EffectPhotonic computing:

Implementation Method 3

a memory sub-system can use a memristor crossbar or array to accelerate multiplication and accumulation operations in electrical domain

Methodology Applied

Scientific EffectMemristor crossbar:

Data Source

PatentUS20250348357A1Heterogeneous Accelerators Connected via a Time Sensitive Networking Bus

Publication Date: 2025.11.13 MICRON TECHNOLOGY INC

AI summary

An apparatus having: a time sensitive networking bus; a plurality of accelerators connected to the time sensitive networking bus to accelerate multiplication and accumulation operations; and a plurality of components connected to the time sensitive networking bus. The components are configured to: run a plurality of applications; generate, in the applications, tasks of multiplication and accumulation operations; assign the tasks to the accelerators; and allocate virtual channels over the time sensitive networking bus from the applications to the accelerators based on timing data of the applications.