A Power-Aware Reinforcement Learning Technique for Memory Allocation in Real-time Embedded Systems

Karimi Fatemi, Masoud

A Power-Aware Reinforcement Learning Technique for Memory Allocation in Real-time Embedded Systems

dc.contributor.advisor	Pahlevani, Majid
dc.contributor.advisor	Far, Behrouz Homayoun
dc.contributor.author	Karimi Fatemi, Masoud
dc.contributor.committeemember	Sanati-Nezhad, Amir
dc.contributor.committeemember	Moshirpour, Mohammad
dc.date	2021-02
dc.date.accessioned	2021-01-06T22:44:33Z
dc.date.available	2021-01-06T22:44:33Z
dc.date.issued	2020-12-22
dc.description.abstract	Embedded systems are ubiquitous in today's world. They are used in a vast number of applications, from medical devices to spacecraft. Two of the main characteristics of such systems are real-time constraints and the lack of reliable energy sources. As cache memories negatively contribute to these two challenges, embedded systems have adopted a new concept called scratch-pad memories (SPMs). To further reduce power consumption, hybrid SPMs composed of Static RAMs (SRAMs) and non-volatile memories (NVMs) have been introduced. Compared to SRAMs, NVMs offer low leakage power and high density while suffering from high energy consumption during write activities. As such, numerous studies have been done on performing memory allocation in hybrid architectures. As memory allocation is an NP-complete problem, traditional approaches fall short due to their timing and memory complexities. Therefore, previous studies have offered alternative techniques to resolve this problem in significantly shorter times. In this work, I propose a reinforcement learning technique to tackle the memory allocation problem in a hybrid architecture. As the results showed, the proposed technique can provide a near-optimal solution - only 5.6% less efficient on average - in a substantially shorter time. To evaluate the algorithm's effectiveness, I compared the results with a greedy technique and a Genetic Algorithm (GA). The results showed that the proposed technique could reduce the average energy consumption by 35.84% and 5.1% compared to the greedy technique and GA, respectively. While the improvements gained by the proposed technique was negligible compared to the GA, it could outperform the GA in simulation time once the learning phase was complete. Particularly, the experiments on 15 soft real-time task frames showed a 35.1% reduction in the average simulation time.	en_US
dc.identifier.citation	Karimi Fatemi, M. (2020). A Power-Aware Reinforcement Learning Technique for Memory Allocation in Real-time Embedded Systems (Master's thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca.	en_US
dc.identifier.doi	http://dx.doi.org/10.11575/PRISM/38530
dc.identifier.uri	http://hdl.handle.net/1880/112938
dc.language.iso	eng	en_US
dc.publisher.faculty	Schulich School of Engineering	en_US
dc.publisher.institution	University of Calgary	en
dc.rights	University of Calgary graduate students retain copyright ownership and moral rights for their thesis. You may use this material in any way that is permitted by the Copyright Act or through licensing that has been assigned to the document. For uses that are not allowable under copyright legislation or licensing, you are required to seek permission.	en_US
dc.subject	Reinforcement Learning	en_US
dc.subject	Embedded Systems	en_US
dc.subject	Memory Allocation	en_US
dc.subject	Low Power	en_US
dc.subject	Real-time	en_US
dc.subject.classification	Artificial Intelligence	en_US
dc.subject.classification	Computer Science	en_US
dc.title	A Power-Aware Reinforcement Learning Technique for Memory Allocation in Real-time Embedded Systems	en_US
dc.type	master thesis	en_US
thesis.degree.discipline	Engineering – Electrical & Computer	en_US
thesis.degree.grantor	University of Calgary	en_US
thesis.degree.name	Master of Science (MSc)	en_US
ucalgary.item.requestcopy	false	en_US