Arbeitgeber: Technische Universität Dresden

TUD Dresden University of Technology, as a University of Excellence, is one of the leading and most dynamic research institutions in the country. For TUD diversity is an essential feature and a quality criterion of an excellent university. Accordingly, we welcome all applicants who would like to commit themselves, their achievements and productivity to the success of the whole institution. At the Faculty of Electrical and Computer Engineering, Institute of Circuits and Systems, the Chair of Highly-Parallel VLSI Systems and Neuro-Microelectronics offers a position as Research Associate / PhD Student (m/f/x) (subject to personal qualification, employees are remunerated according to salary group E 13 TV-L) starting as soon as possible. The position is limited until May 31, 2029. The period of employment is governed by the Fixed Term Research Contracts Act (Wissenschaftszeitvertragsgesetz - WissZeitVG). The position offers the chance to obtain further academic qualification (usually PhD). Balancing family and career is an important issue. The position is generally suitable for candidates seeking part-time employment. Please indicate the request in your application. State-of-the-art AI systems depend heavily on models, providers and hardware from the U.S. and China, which represents a big challenge of Europe’s sovereignty for AI model development and deployment (e.g., trustworthiness, dependability, performance, modularity). The Horizon Europe project OptimAIse (Optimising Performance and Trust for Integrity-driven Modular genAI Software Engineering) addresses these challenges by delivering a scalable, modular, and interoperable reference architecture that leverages European hardware to enable efficient and simplified large language model deployments. The SpiNNaker2 hardware, developed by TU Dresden and commercialized by SpiNNcloud, is one of Europe’s most promising alternatives for the energy-efficient serving of LLMs. SpiNNaker2 is a massively parallel architecture with locally dense compute and globally sparse and low-latency communication, ideal to realize efficient AI models by leveraging sparse and event-based computing. The candidate will develop and adopt LLMs for the SpiNNaker2 hardware. The models shall be implemented on the hardware using an existing software stack and optimized with ML compilers such as MLIR. Besides applying known approaches such as Mixture-of-Experts, the candidate shall follow the state of the art of efficient language models and try novel approaches on SpiNNaker2. In addition, the work will derive requirements and recommendation for next-generation AI hardware such as Spinnaker3, thus guiding the future of efficient AI systems. scientific research in efficient language models and their hardware deployment development and training of sparse and communication avoiding GenAI models optimized for SpiNNaker2 hardware implementation of GenAI models (resp. their layers) on SpiNNaker2 using ML compilers (e.g., MLIR) presentation and publication of research results in top-tier conferences/journals collaboration in European project OptimAIse, integrating GenAI models on SpiNNaker2 for use-case demonstrators university degree (Master’s or equivalent) in computer science, electrical engineering, machine learning or related fields of expertise good understanding of LLMs and how they are processed on AI hardware for inference very good programming skills (e.g., C++, Python) good written and spoken English skills high motivation and ability to work independently and in teams excellent skills and practical experience in one or more of the following research areas is beneficial: compiler frameworks (LLVM, MLIR) embedded software development computer and accelerator architectures parallel and distributed computing the opportunity to collaborate within a diverse team of multi-domain experts at HPSN chair access to the world’s-largest brain-inspired supercomputer SpiNNcloud acce...