NoC Buffer Sizing via Machine Learning Optimization

The patent applies parameter changes by dynamically adjusting buffer sizes based on traffic conditions and channel characteristics. The buffer depth is optimized as a variable parameter rather than a fixed value, allowing the system to adapt to different communication scenarios and achieve high communication efficiency while avoiding the constant power consumption and area overhead of uniformly large buffers.

The patent implements dynamics by making buffer sizes adaptive rather than static. The buffer configuration changes dynamically based on observed traffic patterns, channel utilization, and performance metrics. This allows the system to maintain high reliability for handling traffic variability while reducing the need for overly complex fixed buffer configurations that would be required to accommodate all possible traffic scenarios simultaneously.

The patent applies preliminary action by pre-training machine learning models offline to determine optimal buffer characteristics for different traffic scenarios. The trained models are then deployed to make rapid predictions during runtime without requiring complex real-time computations. This approach achieves high performance optimization while keeping the runtime computational complexity manageable by shifting the heavy computational burden to the offline training phase.