Datum und Uhrzeit:
Ort: Seminar room OMZ (U013, INF 350, floor -1)
Title: From Potential to Practice: Near-Memory and In-Memory Computing Architectures
Abstract: Conventional computing systems face challenges in balancing power efficiency and performance, leading to the rise of specialized domain-specific accelerators. Among them, near- and in-memory computing paradigms have shown significant improvements in performance and energy efficiency, gaining considerable attention, particularly for memory-intensive applications like machine learning and bioinformatics. However, these architectures also bring challenges in system design, reliability, and programmability.
In this talk, we will explore different ways to tackle these challenges. For programmability, we will discuss how high-level compiler frameworks can help make these novel Systems accessible, even to non-experts. On the system design side, we will look at how complex arithmetic operations can be efficiently implemented using the fundamental primitives of these architectures. Regarding reliability, depending on the application, we will see how we can either turn the limitations of the underlying technologies into an advantage (“defect as a feature”) or treat reliability as an important optimization goal. Specifically, we will see how high-level compiler frameworks can be made aware of these properties to simultaneously optimize for multiple metrics, including performance, energy efficiency, and System reliability.
CV: Dr.-Ing. Asif Ali Khan is a post-doctoral researcher at the Chair for Compiler Construction in the Computer Science Department of the TU Dresden, Germany. His Research interests include computer architecture with a focus on near-memory and in-Memory computing, heterogeneous memories, and compiler optimizations for novel architectures. He has published his research findings in the most important conferences and journals in the domain, including HPCA, ASPLOS, DATE, DAC, LCTES, ISLPED, IEEE CAL, IEEE TCAD, IEEE TETC, IEEE TC, IEEE TVLSI, Proceedings of the IEEE, ACM TECS and ACM TACO. Asif is actively collaborating with academic and industrial partners, including joint work with ETH Zurich, Tampere Univ., Univ. of Notre Dame, Univ. of Zaragoza, Pittsburgh Univ., TU Eindhoven, TU Dortmund, TU Twente, IST Islamabad, and MPI Halle.