Fabian Peddinghaus

I build compilers for AI accelerators. At Axelera AI I joined the ML compilation group as one of its first engineers and have worked on core components of the production compiler across three silicon generations. I primarily work on memory, scheduling, and tiling optimizations for in-memory compute, the compiler’s concurrency and synchronization model, and high-performance kernels for LLM, ViT, VLM, and CNN inference on Axelera silicon. I also designed and open-sourced OmniMalloc, a static memory allocation framework.

Before Axelera, I was a visiting student researcher at Stanford University, where I co-developed MINOTAUR, an edge transformer accelerator with on-chip resistive RAM, presented at the VLSI Symposium. I also wrote my master’s thesis on this work. I hold an M.Sc. in Electrical and Computer Engineering from TU Munich, where I created muRISCV-NN, an open-source AI kernel library for RISC-V, and a B.Sc. in Computer Engineering from TU Berlin, where I worked on flight control systems and wrote my thesis on spectrum sensing for nanosatellites. During my studies I was supported by the German Academic Scholarship Foundation.

Outside of work I play field hockey, ski, run, and scuba dive (certified Rescue Diver), and I like to tinker with electronics, from custom calculators to walking robots. Some impressions are in my photos.