Neuronal computing uses living neural networks as part of a computational system. The most direct implementations grow dissociated neurons on multielectrode arrays so that software can stimulate the culture and record its electrical activity in real time.
Because the neurons sit close to the electrodes, two-dimensional cultures offer relatively direct input and output compared with three-dimensional organoids. That makes them attractive for closed-loop experiments in which neural activity changes the environment and the new environment changes the next stimulation.
From DishBrain to CL1
DishBrain demonstrated a closed-loop Pong environment in 2022. Cortical Labs subsequently productised the neuron-on-silicon approach as CL1 and made remote access available through Cortical Cloud.
FinalSpark takes a related but three-dimensional route, maintaining neural organoids on electrode systems and exposing them remotely to researchers.
What neurons might be good at
The plausible advantages are adaptive behaviour, continual learning, low-data learning and complex nonlinear dynamics. None of those automatically creates a useful computer. Biological variability, limited I/O, culture lifetime and the electronic infrastructure around the cells can erase theoretical advantages.
See how biocomputers work for the signal path and the database for evidence-labelled systems.