Bioinspired iontronic architectures for neuromorphic intelligence

Dear Editor,

The human brain operates as an embodied ionic system fundamentally distinct from silicon-based neuromorphic approaches, which remain disembodied and energy-intensive due to von Neumann architectures that separate memory and computation and exceed the brain’s ∼20 W power budget by orders of magnitude. By contrast, the brain couples nanoscale ionic transport, chemical reactions, and energy flow to achieve perception, learning, and decision-making. Inspired by this principle, iontronic systems use ions as charge carriers to replicate neural functions: nanoconfined channels emulate Ca2+-gated neurotransmission, asymmetric electrical double layers (EDLs) encode memory and directional signaling, and memristive ionic networks integrate plasticity and feedback. These hierarchical architectures establish a physical basis for embodied intelligence, where computation emerges directly from matter and charge. Iontronics thus offers a path toward energy-efficient, self-organizing, brain-inspired neuromorphic matter.

The Blue Brain Project, launched by EPFL in 2005 and later expanded into the EU’s Human Brain Project, aimed to digitally reconstruct the mammalian brain by integrating neuroscience, computation, and informatics. These efforts relied on electronic architectures that translated neuronal morphologies and synaptic dynamics into circuit equations executed on supercomputers, driving major advances in brain modeling and shaping artificial intelligence (AI). Yet a fundamental gap remains between simulation and embodiment. The biological brain operates within a ∼20 W power budget, whereas simulating even a single neocortical column on IBM’s Blue Gene/Q requires megawatts, an energy difference of over six orders of magnitude. This disparity originates from the von Neumann separation of memory and computation, which imposes repeated data transfer under centralized control, unlike the brain’s asynchronous, event-driven, and locally powered operation. Typical electronic neuromorphic devices based on ferroelectric and two-dimensional materials partially bridge this gap by reproducing synaptic plasticity and in-memory learning while retaining key practical advantages such as mature integration, high speed, and established manufacturing infrastructure. However, electrons remain the sole information carriers, with a centralized energy supply and computation based on charge displacement rather than local electrochemical evolution, resulting in functional mimicry without physical embodiment and an intelligence detached from its material substrate. Recent ionic-electronic hybrid platforms partially address these limitations by using an external bias to regulate interfacial redox reactions and EDL dynamics, enabling ionic processes to modulate electronic transport and co-localize sensing, memory, and computation.1 Building on this progress, achieving adaptive, low-power computation requires a shift toward self-powered ionically coupled systems driven by intrinsic interfacial electrochemical dynamics, in which matter, energy, and information are intrinsically linked.

Unlike disembodied, electron-based intelligence, where information processing is largely abstracted from material dynamics and encoded in charge flow, the biological brain operates as an embodied system in which ions act as the primary information carriers. Beyond charge, ions encode information through their valence states, chemical identities, and concentration gradients, directly coupling physical state evolution with information processing. At neuronal interfaces, ionic motion within nanoscale EDLs couples with local fields and interfacial reactions, enabling signal transmission, memory formation, and energy conversion within the same processes. These processes operate at picojoule-scale energy per neuronal event, providing a physical basis for the brain’s energy efficiency and multidimensional information encoding.

Realizing embodied intelligence requires returning to the brain’s fundamental unit: the synapse (Figure 1). Beyond a communication interface, the synapse functions as a nanoscale reaction field where perception, memory, and computation converge through coupled ionic and electrochemical dynamics. Inspired by this principle, iontronic architectures integrate sensing, memory, and computation into physical units, forming the basis of low-power, matter-based intelligence. At the channel level, nanoconfined ion transport mimics Ca2+-gated influx; at the interface level, asymmetric EDLs generate memristive plasticity; and at the network level, iontronic arrays enable distributed, in-memory processing. Coupling across these levels allows matter, energy, and information to coevolve, establishing a physical substrate for learning and adaptation and enabling a transition from disembodied AI to embodied intelligence. In the following sections, we outline this hierarchy from nanoconfined ion transport to asymmetric EDLs and finally to ionic memristive networks for adaptive neuromorphic learning.