In our seminars, students learn about cutting-edge research in the research fields presented below. Students are offered topics by the supervisors, but also can suggest their own topics in these fields. The seminar is offered in both English and German.
Modern workloads such as machine learning, multimedia processing, data mining show an intrinsic resilience to errors. Their characteristics such as redundancies in their input data and robust-to-noise algorithms or existence of multiple acceptable results allow them to produce outputs of acceptable quality, despite an approximation in some of their computations.
Approximate computing is an emerging design paradigm that leverages applications error resilience by relaxing traditional computation to a limited extent towards improving energy efficiency and performance. Techniques such as skipping non-critical computations at software level, or reducing circuit complexity and lowering the operation voltage at hardware level significantly improve performance in terms of execution time, area, and power/energy. In the era of lower power requirements and increasing computational complexity, they offer promising answers to both low power and high performance systems.
Technology scaling is approaching its end in which breakthroughs become inevitable. This holds even more after starting the massive production of commercial processors at the 7nm technology nodes, where transistors are already fabricated at the atomic level.
In this seminar, we will explore together the new trends in computer architecture such as in-memory computing, near-memory computing, and neuromorphic computing, which promise a significant improvement in computing efficiency. We will also explore how emerging technologies can reshape the future of computing. In addition, we will also focus on learning all the key challenges that will face software and hardware developers in the future while designing reliable, yet energy-efficient processors. As an example of such challenges is circuits aging, thermal and power management of many-core systems, hardware/software security and side channel attacks. The latter impose serious threat for all future processors ranging from small Internet-of-Thing (IoT) devices all the way up to high-end processors in data centers.
Future Computer Architectures
Technology scaling is closing in on the limit of what is physically possible - there are just a few atoms left at 7nm. Without further technology scaling, classical computer architecture concepts will reach their limits. Therefore, new concepts have to be explored to enable further improvements in performance, energy efficiency, and capabilities. The concept of in- and near-memory computing moves processing capabilities closer to the data or even integrates them directly. Neuromorphic computer architectures simulate brains enabling "learning machines" far more energy efficient and sophisticated than a software-based neural network. In this seminar we will discuss those and other future computer architectures. We do offer a range of topics, but also invite you to propose a topic of your own.
Non-Volatile Memory Architectures
The focus of the seminar is on non-volatile emergeing memory technologies like like STT-MRAM, FeRAM or Intel's 3D XPoint.
The era of DRAM and Flash memory is soon coming to an end. Emerging technologies like STT-MRAM, FeRAM or Intel's 3D XPoint are making constant progress. They combine main memory characteristics like durability and speed with storage class's non-volatility and high density. This will revolutionize the memory hierarchy and enable new ways to handle data, from a small IoT device all the way up to large high performance cloud systems.
In this seminar, we will analyze numerous challenges on the way to a new memory architecture. Will there still be a memory hierarchy? If so, what will it look like? How can different technologies complement each other? What will a file system for a one-memory-architecture look like? Which new applications are possible? How does it affect security?
Dependability for Reconfigurable Architectures
Dependability has become one of the prime concerns in recent nano-era. Reliability (the ability of system to deliver services as specified) and Security (the ability of the system to protect itself against deliberate or accidental intrusion) are the two crucial attributes of dependable systems. Among the other reliability threats due to physical limits of CMOS technology, radiation induced soft-errors or transient faults are also the most challenging threat to be handled. During this seminar, we will explore state-of-the-art for the power-efficient Soft-error reliability and study different research solutions to improve soft-error resiliency in power efficient manner leveraging power-performance-reliability trade-offs. During this seminar, the students will also be able to understand hardware security in reconfigurable architectures, and learn the ways to insert Trojan in an FPGA design/IP and explore different techniques to detect such stealthy Trojans, such as bitstream reverse engineering using Open Source Tool flow.
Thermal and Power Aware Embedded Systems
Power densities are continuously increasing along with technology scaling and the integration of more transistors into smaller areas, potentially resulting in thermal emergencies on the chip. To mitigate such emergencies, power and thermal management techniques are employed. The state-of-the-art power and thermal management techniques can be classified into several categories, such as reactive and proactive techniques, centralized and distributed ones. Recently, machine learning algorithms are employed in power and thermal management techniques to make them more proactive and adaptive. Those various categories of the state-of-the-art techniques need to be reviewed in this seminar to demonstrate the advantage and disadvantage of each of them.
Reconfigurable Embedded Systems
Various types of (re)configurable systems have emerged in recent years. The spectrum ranges from one-time configurable systems that are programmed at the design time for product-specific requirements, to reconfigurable systems that can also be adapted after commissioning, to dynamically reconfigurable systems whose configuration can be changed at runtime and their ability to dynamic reconfiguration is an important part of their system functionality.
Security in Resource Management
Efficient resource management in many-core systems (i.e. systems with more than 100 cores, not only a dozen) has become a research challenge in the last years. As complexity and the demand for scalability increase, this new paradigm should also consider security features to avoid or mitigate the effects of malicious applications on the system performance.
In this seminar, we will focus on the state-of-the-art of resource exhaustion, denial of service (DoS) and other similar resource-intensive attacks and their effects on other critical applications running on many-core embedded systems. During this seminar, student will dive into the security aspects of resource management, while investigating answers to the following research questions:
Please register in ILIAS for participation