Student offers

Conventional RSFQ (Rapid Single Flux Quantum) circuits achieve ultra-fast operation but still rely on sequential, memory-based data paths, similar to conventional digital systems. This project investigates a temporal pipeline architecture in which information is represented by the timing of SFQ pulses rather than stored voltage states. By passing computation through controlled pulse delays, the pipeline eliminates buffer memory and clocking overhead, enabling memoryless and energy-efficient computation. The project includes the design of a small multi-stage RSFQ-based temporal pipeline, the simulation of its timing behavior and the comparison of its performance and energy efficiency with a conventional RSFQ pipeline. The work contributes to the development of next-generation temporal co-processors for superconducting computing systems. Requirements:

• - Basic understanding of digital logic design and pipeline/data flow architectures.
• - Experience with WRspice or similar circuit simulation tools for superconducting devices.
• - Working knowledge of Python or MATLAB for waveform analysis and visualization.

References: [1] Madhavan, A., Daniels, M. W. and Stiles, M. D. (2021). Temporal State Machines: Using Temporal Memory to Stitch Time-based Graph Computations. ACM Journal on Emerging Technologies in Computing Systems, 17(3), Article 28. doi.org/10.1145/3451214

Contact: uzma.majeed@tu-ilmenau.de

More information

In current data center networks, stream processing has become the dominant paradigm for large-scale analytics, enabling high-throughput, low-latency processing of data as it is generated. Due to digital accelerators, e.g. CPUs, this incurs significant energy costs, while optical computing is an unconventional technique that promises an alternative in terms of energy efficiency for data processing and analytics tasks.

However, it is difficult to decompose a concrete stream processing application into stream primitives, e.g. aggregation, and assign them to optical primitives, e.g. averaging. However, it is challenging to decompose a concrete stream processing application into stream primitives, e.g. aggregation, and map them to optical primitives, e.g. averaging. The aim of the thesis is to create a decomposition model based on concrete streaming tasks, e.g. traffic characterization [1], evaluate the digital and optical implementation for a representative operator, and analyze when optics offers advantages in terms of energy consumption.

Requirements:

• No prior knowledge of optics/physics required; basic knowledge of data stream processing and programming skills (e.g. Python) required

References: [1] Brülhart, Cornelia, et al. "Transparent TSN for Agnostic End-hosts via P4-based Traffic Characterization at Switches." 2024 IEEE 49th Conference on Local Computer Networks (LCN). IEEE, 2024.

Contact: wenfei.huang@tu-ilmenau.de

More information

Modern machines try to understand their environment in order to improve the actions and reactions of their users. In the audio domain, an important task is to identify different sound objects. Just like for human listeners, it is helpful to know the number of sources as well as the direction of their origin.

This project focuses on two well-established tasks in the audio domain: Direction of Arrival Estimation (DOA) [1-3] and Source Counting (SC) [2-5]. Existing data sets are used to train a recurrent neural network, called reservoir computing [6]. The design of the network is inspired and constrained by a nonlinear optical system that can be used as a reservoir for computation [7]. The resulting system will finally be executed on an optical computing device to verify the practical application. In addition to the task-related accuracy, the computing performance in terms of computing time and energy consumption will also be investigated.

The aim of the project is to realize an efficient optical computing system that is capable of counting sound sources and locating them horizontally.

Supervisors: Christian Kehling, Anja Bartelmei Reviewers: Stephan Werner, Stephan Sinzinger

More information

A common challenge in recent acoustics research is the automatic detection and localization of sound events in noisy environments. This enables smart devices to identify sirens in urban environments or faulty machines in production halls. In order to be able to evaluate different approaches uniformly, a controlled room is needed to auralize such scenarios with high spatial resolution and reproducibility. The Institute of Media Technology offers the opportunity to use a virtual reality laboratory with a built-in spatial audio rendering system to create immersive audio scenes. These scenes can be perceived from different positions in space and can be rendered with Ambisonics up to 4th order or WFS-based spatial rendering from Fraunhofer IDMT.

The goal of this project is to use these systems to create multiple scenes that mimic a noisy environment with multiple audio events, annotate the content in an established file format (i.e. .aaf), validate an advanced sound source localization (SSL) and acoustic event detection (AED) system, and compare the different rendering systems in terms of accurate and plausible auralization of the signals. More information