In short

AI technologies become increasingly complex and need increasingly large amounts of computational power. As a result, energy consumption of AI is a major issue. In contrast, computation in biological brain tissue consumes six orders of magnitude less power than silicon-based systems. Motivated by this fact, this project aims to develop the world’s first hybrid neuro-in-silico Artificial Intelligence (AI) computer, introducing a fundamentally new paradigm for AI computing. The team aims develop a hybrid AI computer, composed of an in-silico Bayesian control agent (BCA) that communicates with synthetic neural tissue hosted on a microfluidic Brain-on-Chip (BoC). This hybrid AI computer will be trained to solve a real-world AI problem for much less power consumption than a full silicon-based AI computer.


PhD students, main role: research

  • Sepideh Adamiat, PhD student (EE)
  • Focusses on in-silico AI agent, supervised by Wouter, Robert and Bert
  • Guillem Monso, PhD student (ME)
  • Focusses on brain-on-chip hardware, supervised by Burcu, Regina and Wouter

Junior faculty staff, main role: daily supervisors

  • Dr. Burcu Gumuscu Sefunc, assistant professor (BMT)
  • pioneer of microfabrication, hydrogel micro-patterning and single-cell analysis via protein-barcoded microparticles
  • Dr. Wouter Kouw, assistant professor (EE)
  • Background in both neuroscience (MSc) and engineering AI (PhD); specialist in development of free energy minimizing algorithmic intelligent systems.
  • Dr. Robert Peharz, assistant professor (MCS)
  • Specialist in probabilistic AI. Currently at TU Graz, but still part-time assist. Prof. at MSC and remains fully involved with BayesBrain.

Senior faculty, main role: supervision, management, and communication

  • Dr. Regina Luttge, associate professor (ME-Microsystems)
  • chair Neuro-Nanoscale Engineering at Microsystems design and demonstration of platform technologies for (neuro)bio-hybrid systems based on microchip technology.
  • Bert de Vries, Professor (EE)
  • directs the research group BIASlab that focusses on Nature-inspired synthetic Intelligent Agents that learn purposeful behavior solely by minimizing free energy.


With the global use of complex AI tools, the energy consumption of AI also exploded. As these models run on huge data sets, stored in enormous (cooled) data centers, needing more and more computing power, we need urgently more energy-efficient AI computing methods. This project is unique as it implements the physics of the brain (the most successful known ultra-low-power intelligent system), both in a silicon-based and a cultured neural system. In nature, the driving force of movements, including movements of beliefs and information processing in the brain, is that energy differences of any kind are minimized in least time. We will use this method to develop novel synthetic AI on silicon and seamlessly integrate these computations with ultra-low-power computations in cultured brain tissue. As a result, our hybrid AI computer will be as programmable as a current silicon-based AI system, and as energy-conserving as biological brain tissue. The project may also have impact on other disciplines providing new insights in how the brain works, how neurodegenerative diseases develop, and advancing Brain-Computer communication to treat diseases.

The dramatic progress in AI and Machine Learning technology comes with substantial increases in power consumption. The team aims to lay the foundations of a radically different AI concept: instead of large-scale server farms, AI computation could be realized as distributed ultra-low power systems. While several strategies for alternative computational infrastructures exist (e.g., neuromorphic hardware, photonic integrated circuits, organic transistors), this project launches topologically arranged brain-function mimicry in cultured neurons as central computational units.

The proposed method requires a multidisciplinary team of researcher that bring together knowledge of mathematics, biology, electrical-, mechanical- and biomedical engineering, computer sciences and AI. All these expertises are present in the consortium in a balanced manner with different levels of mono- to inter-disciplinary scientists, which is key to making the BayesBrain project successful