Burkhard Ringlein

Burkhard Ringlein is a Postdoctoral Researcher in the Hybrid Cloud department of the IBM Research Zurich Laboratory. He is an accomplished AI systems researcher and designs, builds, debugs, and optimizes practical systems for low-latency, high-throughput machine learning applications. Currently, he improves the platform- and performance-portability of generative AI applications by enabling the automated optimization deployment of AI pipelines and implementing domain-specific compilers to support machine learning models on new hardware. His research enables AI inference frameworks like vLLM or watsonx.ai to adapt itself automatically to run faster, more efficient, and cheaper on a larger number of platforms.

His broader research interests are accelerated and energy-efficient computing (distributed, network-attached FPGAs), domain-specific and reconfigurable architectures (customized AI engines, dataflow processors), custom compiler stacks for heterogeneous computing platforms (MLIR, TVM, FPGA compilers and high-level synthesis transpilation), and infrastructure automation (automated configuration search in very large search spaces for analytics and AI). He received his PhD in 2022 from the Faculty of Engineering of the Friedrich-Alexander University Erlangen-Nürnberg, Germany with the thesis topic "Mapping of a Machine Learning Algorithm Representation to Distributed Disaggregated FPGAs".

In the last five years, he coauthored to over 15 scientific articles and 5 patents. His research accomplishments were recognized by the German Informatics Society, the leading professional computer science association in Germany, by appointing Dr. Burkhard Ringlein as one of three Junior-Fellow in October 2023. Prior to joining IBM Research Zurich, he had positions ad IBM Security in Germany (Kassel) and US (Atlanta), the Fraunhofer Institute for Integrated Circuits (Erlangen) as well as Nokia (Nürnberg).

Projects

Burkhard Ringlein actively engages in the open-source community and contributed (to) the following libraries and projects:

• triton-dejavu: A Framework to reduce the autotuning overhead of triton (a domain- specific language) to zero and therefore enable high-performance and yet flexible de- ployments in production: github.com/IBM/triton-dejavu
• DOSA: An customizable compiler to map Deep Neuronal Networks to distributed FPGAs using the operation set architecture concept: github.com/cloudFPGA/DOSA
• cloudFPGA Development Kit: A development kit with for the IBM cloudFPGA project, including multiple Shells, a network stack, partial reconfiguration controller, and build scripts to establish a community-based unified and portable platform for FPGAs in the Cloud: github.com/cloudFPGA/cFDK