AI Chip Ring Topology for Contention-Resilient Data Transmission
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional interconnection architectures, such as mesh and ring topological architectures based on buses, face issues like clock domain conversion, device contention, and path contention, which affect data transmission efficiency and reliability, especially in artificial intelligence chips where large data volumes are processed in parallel.
Innovation Solution
A topological circuit with a ring configuration that uses a controller to send control signals, where node devices convert these signals into bus signals with transmission modes and broadcast indicators, allowing efficient point-to-point or broadcast data operations, and incorporates selectors to bypass faulty or shut-down nodes, optimizing paths for improved efficiency and reliability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a mesh topological architecture based on a bus is used, then device interconnection is achieved, but clock domain conversion is required and device contention occurs
Solution Approach 1:
The system segments the interconnection network into multiple independent ring topologies. Each ring operates autonomously with its own clock domain, eliminating the need for clock domain conversion across the entire system. Node devices can be grouped into different rings based on communication patterns, allowing parallel operation without contention.
Solution Approach 2:
The patent introduces ring controllers as intermediary devices that manage communication within each ring. These controllers handle data routing and coordination locally, eliminating the need for centralized bus arbitration and reducing device contention. The ring controller acts as a mediator that simplifies the interconnection logic for node devices.
2Adaptability or versatility
If a mesh topological architecture based on a bus is used, then device interconnection is achieved, but path contention occurs
Solution Approach 1:
The interconnection network is divided into multiple disjoint ring paths. Data can be routed through different rings simultaneously, providing multiple independent paths between source and destination nodes. This segmentation eliminates path contention by distributing traffic across multiple separate communication channels.
Solution Approach 2:
The ring topology supports dynamic path selection and reconfiguration. When contention is detected or a failure occurs, the system can dynamically switch to alternative paths within the ring structure. The ring controllers can reconfigure data flow in real-time to maintain reliable communication without being constrained by fixed bus paths.
3Productivity
If node devices are connected in a ring via a bus, then data transmission efficiency is improved, but the system is affected by faulty or shut-down nodes
Solution Approach 1:
The patent extracts the control functionality from individual node devices and concentrates it in dedicated ring controllers. When a node device fails or shuts down, only its control functions are affected, while data transmission can continue through the ring controllers. This separation allows the system to maintain operational integrity even when individual nodes are unavailable.
Solution Approach 2:
Ring controllers serve as intermediary devices that can bypass faulty node devices. When a node fails, the ring controller can directly route data around the failed node through the ring bus, maintaining communication without requiring the faulty node to participate. This intermediary approach isolates failures and prevents them from propagating through the entire network.
Data Source
AI summary
A topological circuit, an artificial intelligence chip, and a data transmission method, which relate to the field of artificial intelligence, are proposed. The topological circuit includes: a controller that sends a first control signal; N node devices, which are respectively connected to the controller and connected in a ring through a bus to form a loop. The N node devices include: a first node device, configured to convert the first control signal into a bus signal and transmit the bus signal along the loop in response to receiving the first control signal, the bus signal includes a transmission option and a broadcast indicator, the transmission option is a read transmission or a write transmission; N-1 second node devices, each second node device is configured to perform a corresponding operation according to the bus signal in response to receiving the bus signal.