Publication Type : Conference Paper
Publisher : IEEE
Source : 2026 International Conference on Smart Futuristic Technology
Url : https://doi.org/10.1109/icsft66733.2026.11507605
Campus : Amritapuri
Center : Humanitarian Technology (HuT) Labs
Year : 2026
Abstract : Pipeline hazards such as data, control, and structural conflicts are still major issues in current highly efficient 5-stage RISC pipelines (IF, ID, EX, MEM, WB). Common Methods Used for hazard mitigation strategies, such as stalling, forwarding, and static branch prediction, are usually firm, effectively applicable to all workloads, independent of their heterogeneous characteristics. This may result in inefficient resource use, higher stalls, and wasteful energy use. Through this work, a workload-aware adaptive hazard mitigation solution is introduced based on machine learning that dynamically optimizes pipeline performance. The paper implements a Python-based cycle-accurate 5-stage RISC pipeline simulator that tells an informative set of instruction trace features encoding workload behavior. The adaptive system switches between policies such as stalling, data forwarding, and branch prediction schemes specific to the workload characteristics observed, decreasing pipeline stalls and minimizing cycles per instruction (CPI), and achieved a reduction in stall count by 36% and CPI by 15.2%. By integrating workload awareness with machine-learning-driven pipeline control, the approach advances beyond traditional static hazard-avoidance methods.
Cite this Research Publication : Betha Jyothi Swaroop, Sreedharan Babu, Meka Nitin Srinivas, Rajesh Kannan Megalingam, Python-Based Simulator for Machine Learning-Driven Adaptive Hazard Mitigation in 5-Stage RISC Pipelines, 2026 International Conference on Smart Futuristic Technology, IEEE, 2026, https://doi.org/10.1109/icsft66733.2026.11507605