A logically correct SoC design isn’t an optimized design

By Rick Bye, Arteris
EDN | October 14, 2025

The shift from manual design to AI-driven, physically aware automation of network-on-chip (NoC) design can be compared to the evolution of navigation technology. Early GPS systems revolutionized road travel by automating route planning. These systems allowed users to specify a starting point and destination, aiming for the shortest travel time or distance, but they had a limited understanding of real-world conditions such as accidents, construction, or congestion.

The result was often a path that was correct, and minimized time or distance under ideal conditions, but not necessarily the most efficient in the real world. Similarly, early NoC design approaches automated connectivity, yet without awareness of physical floorplans or workloads as inputs for topology generation, they usually fell well short of delivering optimal performance.

Modern GPS platforms such as Waze or Google Maps go further by factoring in live traffic data, road closures, and other obstacles to guide travelers along faster, less costly routes. In much the same way, automation in system-on-chip (SoC) interconnects now applies algorithms that minimize wire length, manage pipeline insertion, and optimize switch placement based on a physical awareness of the SoC floorplan. This ensures that designs not only function correctly but are also efficient in terms of power, area, latency, and throughput.

To read the full article, click here