Scientists are on the verge of simulating a human brain using the world's most powerful supercomputers, aiming to unlock secrets of brain function. Led by researchers at Germany's Jülich Research Centre, the project leverages the JUPITER supercomputer to model 20 billion neurons. This breakthrough could enable testing of theories on memory and drug effects that smaller models cannot achieve.
Advancements in computing power are enabling researchers to simulate the human brain at an unprecedented scale. Today's supercomputers, approaching exascale performance with a billion billion operations per second, can handle simulations of billions of neurons, according to the Top500 list, which identifies only four such machines worldwide.
Markus Diesmann at the Jülich Research Centre in Germany explained the shift: “We have never been able to bring them all together into one place, into one larger brain model where we can check whether these ideas are at all consistent. This is now changing.” His team plans to use JUPITER, the Joint Undertaking Pioneer for Innovative and Transformative Exascale Research, based in Germany. Last month, they demonstrated that a spiking neural network—a simple model of neurons and synapses—could scale to run on JUPITER's thousands of graphical processing units, reaching 20 billion neurons and 100 trillion connections. This matches the size of the human cerebral cortex, the hub of higher brain functions.
Diesmann emphasized the value of scale: “We know now that large networks can do qualitatively different things than small ones. It’s clear the large networks are different.” Previous simulations, like those of a fruit fly brain, lack features emerging only in larger systems, similar to how large language models outperform smaller ones.
Thomas Nowotny at the University of Sussex in the UK stressed the need for full-scale efforts: “Downscaling is not just simplifying it a little bit, or making it a bit coarser, it means actually giving up certain properties altogether. It’s really important that eventually we can do full-scale [simulations], because otherwise we’re never going to get the real thing.”
The model draws on real data from human brain experiments, including synapse counts and activity levels, as noted by collaborator Johanna Senk at the University of Sussex. Diesmann added: “We now have these anatomical data as constraints, but also the computer power.”
Such simulations could test theories on memory formation by inputting images and observing reactions, or evaluate drugs for conditions like epilepsy, characterized by abnormal brain activity bursts. Enhanced power allows faster runs to study slow processes like learning and incorporates detailed neuron behaviors.
However, challenges remain. Nowotny cautioned that even brain-sized simulations lack real-world inputs and cannot fully replicate animal behavior. “We can’t actually build brains. Even if we can make simulations of the size of a brain, we can’t make simulations of the brain.”
