Personal tools

Thermal Control of Mobile Devices

From iis-projects

Jump to: navigation, search

Short Description

Driven by the technological scaling and the market growth, mobile processors have seen in the latest years an extraordinary performance increase. Unfortunately this has led to an higher power density and consequently to higher working temperatures. In addition, mobile devices due to their reduced form factor have limited cooling capabilities, worsening their thermal profile. Heterogeneous integration such as the use of GPUs as well as different processing engine (i.e. ARM big.LITTLE) produces an heterogeneous thermal profile which is highly dependent on the actual system usage. As a matter of fact today and future mobile devices are thermally limited.

Dynamic thermal management tackle this issue by adapting the core speed and core performance to limit the power under thermal hazard. Model predictive controller is a promising technology as it uses the knowledge of the thermal behaviour of the device to estimate the future temperature and to find the maximum performance that satisfy the thermal constraint. Model predictive controller is foreseen to outperform classical threshold-based and PID thermal controller as it can avoid overshot and thermal cycles. Dynamic thermal model can be directly identified from the target device by mean of system identification and self-calibrating routines.

Current mobile platforms embed thermal sensors and power control knobs which are directly accessible from the software stack. These HW monitors and actuators can are the ground on which designing software-based thermal control policies. In this project the candidate will design on a real mobile platform a model predictive controller for controlling the temperature. This will require to identify the thermal model of the device and to design the software model predictive controller routine.

Status: Available

Looking for Interested Students
Supervisors: Andrea Bartolini


C Language
Control Theory
Interest in advance control and model learning
Interest in Multicore Run-time Developement


20% Theory
60% Implementation
20% Testing


Luca Benini

↑ top

Detailed Task Description


Practical Details