/DRAM power measurements and energy model calibration

DRAM power measurements and energy model calibration

Master projects/internships - Leuven | More than two weeks ago

As part of this memory exploration simulator framework, we are calibrating a detailed, highly parametrized model for DRAM memory energy usage in real servers, in cooperation with TU Dresden. 
 

Computer energy consumption is forecasted to exceed global energy production in 2040 (Semiconductor Ind. Assoc., 2015). A large factor (~20% and increasing) of this power is attributed to the main memory in those computers, using DRAM technology. At imec, we are developing a cross-layer simulator framework intended to explore novel memory technologies and architectures, to evaluate their effect on the performance and energy usage of future computer systems and applications. These new memory technologies, such as alternative DRAM with high retention time, have the potential to save a significant amount of energy in datacenters around the globe.

As part of this simulator framework, we are calibrating a detailed, highly parametrized model for DRAM memory energy usage in real servers. DRAM energy usage heavily depends on the application behavior, and we want to capture this correlation in a power model that can accurately predict the DRAM memory energy consumption, based on an application profile and a memory specification.
To improve the existing baseline power model, we want to calibrate the model using power measurements on real hardware. For this, we are collaborating with the Technische Universität Dresden, working on state-of-the-art instrumented server setups which allow accurate, non-intrusive, high-speed measurements of DRAM power while running real applications.

The MsC thesis internship will involve instrumenting and profiling applications, writing microbenchmarks, performing measurements, parsing measurements results, charting the CPU and DRAM microarchitecture, simulating these architectures, ... to eventually improve the energy model’s accuracy. 

The goal of this 9-month internship is not only a successful thesis, but also a presentation in front of our industry partners (semiconductor companies from around the world), and a paper in an
academic workshop.

To fulfill these goals, we are looking for a highly motivated computer science/computer engineering student with (at least some) experience in C++ and python, and a high interest in computer architecture, memory
organization and performance analysis.

Reference literature: https://tu-dresden.de/zih/forschung/ressourcen/dateien/projekte/hdeem/HDEEM_2014_E2SC_accepted.pdf?lang=en

Type of project: Combination of internship and thesis 

Master's degree: Master of Engineering Science; Master of Engineering Technology 

Duration: 6-9 months 

Master program: Electrotechnics/Electrical Engineering 

KU Leuven supervisor: Francky Catthoor 

Supervising scientist: For further information or for application, please contact Francky Catthoor (Francky.Catthoor@imec.be).

Imec allowance will be provided for students studying at a non-Belgian university.