PHD Researcher
10-2015 - NowInfra / Networks / Telecom Nowadays, Power Consumption is one of the main limitations of embedded systems. Dynamic Thermal and Power Management methods require an auto-adaptive sub-system based on efficient monitoring techniques, valuable decision strategies to finally ensure an adequate decision to avoid the undesired states of the system.
Successfully developed and established a lightweight Power and Temperature Monitoring Sub-System to set up efficient adaptation techniques such as DVFS, task mapping:
• Effectively contributed to the creation of a specific RTL design flow in order to extract a database from the simulation at different levels of abstraction.
• Created Power consumption model at gate level using internal signal activities (VCD) and at system level using the PMU performance events.
• Successfully designed and implemented Ring Oscillators sub-system in VHDL to Monitor/Predict SoCs temperature on FPGAs.
• Proposed thermal sensor’s placement, thermal Hotspot modeling/tracking.
Thermal Monitoring and Adaptation techniques on modern MPSoC ARM-based processors:
• Performed model for the on-chip Temperature and Power Consumption (robust against the external temperature variations).
• Developed prediction technique to predict the future thermal state of the SoCs.
• Applied and compared different online adaptation strategies For FPGAs & SoCs: Monitoring & Prediction, decision-making, acting.
• Proposed Data Mining selection strategies to extract the relevant PMU performance events.
• Comparing a set of data mining supervised techniques to identify relevant information that correlates with the power consumption & temperature.
Data Mining & Machine Learning
• Preprocess: Filtering feature.
• Feature Selection: Searching and Evaluating - identifying relevant informations.
• Classification: Neural Network.
• Clustering: K-mean, Hierarchical, EM.
• Forecasting/prediction: SES, Holt's linear, Exponential and Damped Trend methods.