Publication date: 15th December 2025
Neuromorphic computing aims to replicate the computational capabilities of the human brain while operating at comparable energy efficiency. In contrast to conventional computing systems, which require large physical resources and high power consumption, the brain performs complex information processing using minimal energy. Reaching this level of efficiency requires a precise understanding and replication of neuronal synapses, the fundamental elements responsible for communication and learning in biological neural networks.
In neuromorphic hardware, a central challenge is to identify the physical mechanisms governing artificial synaptic behavior and to achieve reliable control over them. In our recent work, we demonstrate that the relaxation time between information states in memristive devices is the key parameter underlying synaptic functionality. This relaxation time is strongly voltage-dependent and governs the dynamic transition between resistive states. By tuning this temporal response, gradual conductance modulation can be achieved within specific time windows, enabling a continuous spectrum of intermediate states and fine control over synaptic plasticity.
We combine experimental and modeling approaches to validate this framework. Experimental studies are performed on fluidic memristors, where a quantitative model for relaxation dynamics is developed, and on halide perovskite memristors, where multiple dynamic processes can be identified and directly linked to synaptic behavior. Complementary simulations based on experimentally grounded models provide further insight into the underlying mechanisms and enable predictive control of synaptic responses. [1], [2]
These results establish voltage-dependent relaxation dynamics as the fundamental mechanism governing artificial synapses in memristive devices, offering a robust pathway toward controllable, energy-efficient neuromorphic systems.
This work was funded by the European Research Council (ERC) via Horizon Europe Advanced Grant, grant agreement nº 101097688 (“PeroSpiker”).
