Dynamic Routing Bus Selection for SoC Memory Bandwidth Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In systems-on-chip (SoCs), on-chip data pathways are limited in bandwidth due to propagation times and distance, leading to underutilization of bandwidth and difficulty in achieving quality-of-service guarantees, especially when multiple data sources have varying traffic requirements and ordering restrictions.
Innovation Solution
Implementing a system-on-chip with a memory subsystem, requestors, routers, and memory arbiters, where routers receive feedback on usage capacity from arbiters to dynamically select routing buses, deprioritizing overloaded paths and balancing traffic by using a prioritized arbitration algorithm like round-robin or least-recently-used.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If multiple parallel data pathways are used to increase bandwidth, then memory bandwidth is improved, but bandwidth under-utilization occurs when pathways are not evenly loaded
Solution Approach 1:
The patent implements dynamic route selection where requestors can switch between fixed and dynamic routing modes. In dynamic mode, requestors monitor pathway utilization and adaptively select routes based on current system conditions, allowing the system to respond to changing traffic patterns and balance load across multiple pathways, thereby preventing bandwidth under-utilization while maintaining high memory bandwidth
Solution Approach 2:
The patent employs feedback mechanisms where requestors receive information about pathway utilization and adjust their routing decisions accordingly. This feedback enables the system to identify under-utilized pathways and redirect traffic to balance the load, ensuring that multiple parallel data pathways operate at optimal capacity and preventing waste of available bandwidth
2Productivity
If all sources have access to all routes for optimal bandwidth utilization, then memory bandwidth is improved, but system complexity and cost increase prohibitively
Solution Approach 1:
The patent implements a hybrid routing architecture where different requestors have different routing capabilities based on their specific needs. Some requestors use fixed routing for simple cases, while others can access multiple routes for complex transactions. This localized differentiation allows the system to provide high bandwidth utilization where needed without imposing complex routing logic on all sources, thereby balancing performance with system complexity
Solution Approach 2:
The patent creates a universal routing framework that can accommodate both fixed and dynamic routing modes within the same system. The routing infrastructure is designed to handle multiple routing strategies simultaneously, allowing the system to scale from simple to complex routing needs without requiring separate systems, thus managing complexity while maintaining bandwidth optimization capabilities
3Reliability
If fixed routing is used for some sources to simplify control, then quality-of-service guarantees become easier to achieve, but bandwidth utilization decreases due to inability to adapt to varying traffic requirements
Solution Approach 1:
The patent enables requestors to dynamically switch between fixed and dynamic routing modes based on their current traffic requirements and service level agreements. This dynamic capability allows the system to maintain quality-of-service guarantees through fixed routing when needed while optimizing bandwidth utilization through adaptive routing when conditions permit, thereby resolving the contradiction between reliability and productivity
Data Source
AI summary
A method includes receiving feedback information indicative of an overload condition from an arbiter. The method further includes deprioritizing a routing bus based on the received feedback information and selecting a routing bus to use to send a transaction across a system-on-chip (SOC).