An organic brain-inspired platform with neurotransmitter closed-loop control, actuation and reinforcement learning

Bruno U., Rana D., Ausilio C., Mariano A., Bettucci O., Musall S., Lubrano C. and Santoro F., Mater. Horiz., 2024, Advance Article, https://doi.org/10.1039/D3MH02202A

Our colleagues have published (30.04.2024) a paper in Materials Horizons, which was selected as a part of the themed collection: Memristors and Neuromorphic Systems . In this manuscript a brain-inspired closed-loop system has been demonstrated for the accomplishment of motor control and actuation tasks, through a learning process mediated by a neurotransmitter. This system integrates well established silicon technologies with organic materials, more suitable for communication with biological neurons. While the intelligence and the decision-making process is completely delocalized, it lies in the local adaptation of a neuromorphic OECT. Such adaptation is closed-loop controlled using a PID control law, and it mimics the neurotransmitter-mediated synaptic plasticity of biological neural networks (BNNs). From the existing field of research, sensing and motion control have been achieved by exploiting organic neuromorphic architectures, with the goal to recapitulate autonomous local learning typical of the human neural processing (Krauhausen, I. et al. Organic neuromorphic electronics for sensorimotor integration and learning in robotics. Science Advances 7, eabl5068 (2021).). In these applications there are two main aspects still missing: the first one is the lack of a neuromorphic control-loop architecture strongly desirable for the adaptive responsiveness to external stimuli of the system; the second one is the employment of biological signalling as responsible of the synaptic plasticity during the learning process, typical of the human brain and useful for the active integration of this technologies in a biological environment. In this new concept dopamine is the signal used for the strengthen of the artificial synapse, integrated in a closed-loop system able of adaptive learning.

Figure 1. Overall neuromorphic closed-loop architecture and open-loop artificial neuron characterization. (a) Schematic of a closed-loop neuromorphic system, coupling a silicon microcontroller to drive external actuation, while delocalizing the decision-making process to the organic synaptic device (left). Comparison between a biological synapse and the organic synaptic device (right). (b) Synaptic potentiation elicited by a train of square voltage pulses applied at the gate terminal in presence of DA. (c) Synaptic potentiation as a function of DA molarity (Numerical values: 3.9 ± 0.3; 4.6 ± 1.9; 6.6 ± 2.2; 10.3 ± 1.3; 19.8 ± 5.5; 21.0 ± 6.7; 18.7 ± 7.1; (N = 3)), along with schematic of the PEDOT:PSS reduction mechanism. (d) Synaptic depression elicited by a train of square voltage pulses applied at the gate terminal in presence of H₂O₂. (e) Synaptic depression as a function of H₂O₂ molarity (Numerical values: 1.2 ± 0.6; 1.8 ± 1.3; 2.6 ± 1.6; 2.4 ± 1.0; (N = 3)), along with a schematic of the PEDOT:PSS oxidation mechanism.

Publication: Bruno U., Rana D., Ausilio C., Mariano A., Bettucci O., Musall S., Lubrano C. and Santoro F., Mater. Horiz., 2024, Advance Article, https://doi.org/10.1039/D3MH02202A