Brain-inspired computing at large-scale
Join us for our upcoming Future Computing Seminar Series
Speaker: Dr. Hector Gonzalez (SpiNNcloud Systems)
Date: Monday, Feb 2nd, 2026, 13:30 CET
Where: ETZ E6
Abstract:
Dynamic sparsity is emerging as a dominant trend for the next generation of AI models and systems, driven by the continuous pursuit of computational efficiency in training across both mainstream and frontier architectures. While current production models already exhibit mild to moderate sparsity through techniques such as structured pruning, mixture-of-experts, conditional computation, and activation sparsity, the trajectory of model design is moving toward extreme sparsity, where only a small fraction of parameters and computation are active at any given time. Dynamic sparse computing addresses this full continuum by enabling computation, communication, and memory access to scale with the active subgraph of the model rather than its dense envelope. Through fine-grained, event-driven execution and sparse data movement, it enables native and efficient execution of both slightly sparse and highly sparse models, while simultaneously opening the design space for a new class of hardware-enabled model architectures. As sparsity becomes a first-class design principle in mainstream AI, dynamic sparse computing, in our case rooted in neuromorphic principles, provides a forward-compatible path that aligns system design with the structural and efficiency constraints imposed by emerging sparsity trends.
Speaker:
Hector is the co-founder and CEO of SpiNNcloud Systems, a deep-tech company providing dynamic-sparse microchips and supercomputers for HPC and AI workloads. Hector has helped position SpiNNcloud among the most relevant hardware startups in Germany. Hector is a Fellow of the Konrad Zuse School of Excellence in Embedded Composite Artificial Intelligence (SECAI). He holds a B.Sc. in Electronics Engineering and is a graduate of the MIT & Masdar Institute Cooperative Program in Abu Dhabi, where he earned an M.Sc. in Microsystems after conducting research on AI hardware for biomedical applications. His PhD studies at TU Dresden focused on chip design for AI-enabled Digital Signal Processors (DSPs) for automotive radars. Hector has held senior industrial positions in Instrumentation & Control across various countries and has received numerous international academic honours and awards. He has authored or co-authored more than 23 peer-reviewed articles, is the inventor of patents in the cognitive radar field, and has been part of numerous press communications from prominent sources such as IEEE Spectrum, the BBC, Arm, EE Times, Sandia National Labs, eeNews Europe, Silicon Angle, and the EU flagship Human Brain Project, among others.