Talks and presentations

Here you can find the slides of most of the invited talks that I have given.

Kako radi računar?

January 24, 2025

Blokadno predavanje, PMF u blokadi, Novi Sad, Serbia

Apstrakt Bez obzira na struku kojoj pripadamo ili naučnu oblast kojom se bavimo, danas svi koristimo računar; ako ništa drugo, onda bar da bismo došli do informacija poput najave ovog predavanja. No, da li vas je ikad zanimalo kako računar zapravo radi? Polazeći od jednog matematičkog problema koji je 1769. formulisao Leonard Ojler, a koji je rešen tek 1966, upotrebom računara, na ovom predavanju ćemo uz minimalne pretpostavke zajedno konstruisati fon Nojmanov računar koji predstavlja osnovu gotovo svih savremenih računara. Pored interaktivnog dela predavanja u kom ćemo naučiti kako računar zaista radi, osvrnućemo se i na neke zanimljive trenutke iz istorije razvoja računarstva u svetu i kod nas. Ako ste se pitali šta će budućim hemičarima, fizičarima i matematičarima informatika, verovatno će vam biti zanimljivo da čujete kako su upravo hemičari, fizičari i matematičari pomogli da računarstvo uopšte postane naučna oblast. Usput ćemo saznati i kakva je veza između jednog sela iz doline Jadra i prvog računara u Jugoslaviji, a nema nikakve veze sa litijumom.

Securing a Bright Future for FPGAs

November 28, 2024

CROSSING Research Seminar Talk, Technische Universität Darmstadt, Darmstadt, Germany

Abstract Four years ago, when the news of AMD acquiring Xilinx first became public, FPGAs were a rather hot topic. New reconfigurable chips being rolled out in the back-then latest fabrication technologies were promising such things as dramatic performance boosts in AI inference and fundamentally more secure hardware. Yet, in the second quarter of 2024, AMD reported a 41% drop in year over year revenue of its embedded division into which Xilinx was incorporated. This is in stark contrast with GPUs which were part of similar AI-related promises. Two major reasons for deflated expectations from FPGAs are 1) that FPGAs failed to transition to newer technology nodes in the four years that have passed, thus losing much of the competitive edge that they used to have for most of the XXI century and 2) that attempting to use an FPGA can still be a frustrating experience for domain experts.

Algorithms for Automated FPGA Interconnect Design

February 17, 2023

Talk, CUNY Graduate Center, New York, USA

Abstract Field-Programmable Gate Arrays (FPGAs) are the most popular platform for reconfigurable computing. As such, they provide an appealing way to implement custom hardware, especially for acceleraing rapidly evolving tasks like those found in machine learning inference. Despite being around for almost four decades, there are still many unknowns in designing the FPGA architectures themselves; this is paricularly tue for their programmable interconnect.