Hardened NoC Mapping for FPGA Interconnect Bottlenecks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing Field-Programmable Gate-Arrays (FPGAs) and programmable logic devices (PLDs) face inefficiencies in interconnect performance due to traditional point-to-point routing methods, leading to resource wastage and suboptimal bandwidth utilization.
Innovation Solution
The integration of a hardened Network-on-Chip (NoC) within FPGAs, which allows for customizable traffic routing and optimized performance by reconfiguring the NoC based on application-specific traffic profiles, thereby enhancing bandwidth utilization and reducing latency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional point-to-point routing methods are used in FPGAs, then the routing is simple and easy to implement, but the interconnect performance is inefficient and bandwidth utilization is suboptimal
Solution Approach 1:
The routing network is segmented into multiple independent virtual channels (VC0, VC1, VC2, VC3) that can be configured separately. Each virtual channel operates independently, allowing efficient packet switching and reducing congestion. This segmentation enables the system to achieve high interconnect performance while maintaining manageable complexity through modular virtual channel management.
Solution Approach 2:
The routing architecture is made dynamic by allowing reconfiguration of virtual channels based on traffic patterns. The system can dynamically allocate and route packets through different virtual channels depending on the application requirements, enabling optimal bandwidth utilization. This dynamic adaptability resolves the contradiction by providing high performance when needed while maintaining simplicity through automated routing decisions.
2Productivity
If a hardened Network-on-Chip (NoC) is integrated into FPGA, then bandwidth utilization and performance are optimized, but the device complexity increases
Solution Approach 1:
The hardened NoC architecture is designed to serve multiple functions within a single integrated structure. The same NoC infrastructure handles different types of traffic (data, control, addressing) through virtual channel multiplexing. This multi-functionality allows the system to achieve optimized bandwidth utilization without proportionally increasing complexity, as one architectural element serves multiple purposes.
Solution Approach 2:
The virtual channels are nested within the physical routing structure, with multiple virtual channels sharing the same physical interconnect resources. This nesting allows efficient space utilization and reduces the overall hardware footprint. The hierarchical organization of virtual channels within the NoC structure enables optimized bandwidth performance while keeping the integrated complexity manageable through compact design.
3Loss of time
If the NoC is reconfigured based on application traffic profiles, then latency is reduced and resource allocation is optimized, but the reconfiguration process becomes more complex
Solution Approach 1:
The system performs preliminary analysis of traffic profiles during the design phase to pre-determine optimal virtual channel assignments and routing paths. This preliminary configuration allows the NoC to be optimized for specific application requirements before execution, reducing runtime latency without requiring complex dynamic reconfiguration. The preliminary action resolves the contradiction by establishing optimal paths in advance while keeping runtime reconfiguration simple.
Solution Approach 2:
The reconfiguration process incorporates feedback mechanisms that monitor traffic patterns and automatically adjust virtual channel assignments. This feedback-driven approach reduces latency by adapting to actual runtime conditions while simplifying the reconfiguration process through automated decision-making. The feedback loop enables the system to optimize resource allocation dynamically without requiring manual intervention or complex reconfiguration algorithms.
4Adaptability or versatility
If multiple virtual channels are implemented in the NoC, then traffic customization and load balancing are improved, but the routing control complexity increases
Solution Approach 1:
The virtual channels are designed to operate autonomously with self-managed routing and error handling. Each virtual channel independently manages its own packet switching and congestion control, reducing the burden on centralized routing control. This self-service capability enables sophisticated traffic customization and load balancing while keeping routing control complexity manageable through distributed intelligence.
Solution Approach 2:
The system achieves traffic customization by dynamically changing routing parameters such as virtual channel assignments and packet priorities rather than requiring complex structural reconfiguration. This parameter-based approach allows flexible traffic management and load balancing across multiple virtual channels while maintaining simple routing control logic. The system adapts to different traffic patterns by adjusting parameters rather than redesigning the routing architecture.
Data Source
AI summary
Methods and example implementations described herein are generally directed to the addition of networks-on-chip (NoC) to FPGAs to customize traffic and optimize performance. An aspect of the present application relates to a Field-Programmable Gate-Array (FPGA) system. The FPGA system can include an FPGA having one or more lookup tables (LUTs) and wires, and a Network-on-Chip (NoC) having a hardened network topology configured to provide connectivity at a higher frequency that the FPGA. The NoC is coupled to the FPGA to receive an profile information associated with an application, retrieve at least a characteristic, selected form any of combination of any or combination of a bandwidth requirement, latency requirement, protocol requirement and transactions, associated with the application from the profile information, generate at least one application traffic graph having mapping information based on the characteristic retrieved, and map the application traffic graph generated with into the FPGA using the hardened NoC.