From communication to computation in neurons-on-a-chip: an in silico study of neurotopomorphic computing

Michael Taynnan Barros

q-bio.NCOct 5, 2026

From communication to computation in neurons-on-a-chip: an in silico study of neurotopomorphic computing

Authors: Michael Taynnan Barros

Organizations: School of Computer Science and Electronic Engineering, University of Essex, Colchester, UK

Abstract

Living neuronal networks transform inputs through recurrent cellular and population dynamics, yet it is unknown which network architecture supports which computation. Neurons-on-a-chip turn this question into a design problem because microchannels guide axonal growth and set the network architecture. We introduce IC $^3$ , an Integrated Characterisation of Communication-Driven Computation, which characterizes network state through neuronal dynamics, functional communication, and structural support. We implemented nine architectures \textit{in silico} as conductance-based spiking networks and tested each on frequency decoding, temporal-order discrimination, and fading memory. Predominantly feedforward circuits decoded best. Sequential Chain and Microchannel Diode had the lowest IC $^3$ and recruited a third of reachable neurons, yet achieved the two highest scores on both classification tasks. Across architectures, higher IC $^3$ went with lower classification scores. We term this new direction \emph{neurotopomorphic computing}, in which the physical organisation of neuronal connectivity is engineered as part of the computing substrate.

Figures & tables

Figure 1: Neurons-on-a-chip architecture links physical circuit design to neuronal dynamics, functional communication, and computational function. a , Conceptual neurons-on-a-chip platform integrating patterned neuronal networks, PDMS microchannels and compartments, extracellular recording and stimulation electrodes, and an MEA substrate. b Nine candidate architectures were evaluated as graph-defined neuronal circuits. Their activities were characterized using IC 3 , which integrates neuronal dynamics, functional communication, and structural support, together with transfer entropy and propagation-related measurements. The resulting network states were then assessed using close-frequency decoding, temporal-order discrimination, and fading-memory tasks.

Figure 2: IC 3 captures architecture-dependent network configurations. a , Mean IC 3 values for the nine neuronal architectures across seven stimulation frequencies. b , Mean target-entropy-normalized, shuffle-corrected transfer entropy (TE) across the same architecture and frequency sweep. c , Joint IC 3 –TE state space for architecture and frequency means. Markers identify architectures, colours indicate stimulation frequency, and error bars show 95% confidence intervals across network instances. d , Standardized values of the eight IC 3 components across architectures, showing how neuronal dynamics, functional communication, and structural support contribute differently to each network configuration. e , Decomposition of mean IC 3 into neuronal dynamics, functional communication, and structural support. Stacked bars show the weighted domain contributions with mean IC 3 with 95% confidence intervals across network instances.

Figure 3: Architecture shapes the conversion of structural connectivity into downstream neuronal responses. a , Mean post-stimulus response probability as a function of network hop distance, with 95% confidence intervals across independently seeded network instances. b , Relationship between post-stimulus first-spike latency and the shortest physical path delay from the stimulated source to the responding neuron. c , Architecture-level comparison between the fraction of downstream neurons that were structurally reachable from the stimulated source and the fraction that responded after at least 50% of stimulus events ( $R\geq 0.5$ ). d , Mean post-stimulus response probability across alternative response-window durations. Architecture-specific differences remained across the tested windows, while response probability approached a plateau at longer windows.

Figure 4: Architecture-dependent communication phenotypes and their functional trade-offs. a , Relationship between upper-tail downstream firing rate and ISI variability across architectures and stimulation frequencies. b , Operating points relating post-stimulus response probability and response latency across stimulation frequencies. c , Model-predicted post-stimulus response probability as a function of out-closeness, evaluated at a hop distance of two and averaged equally across architectures. d , Standardized communication-phenotype profiles for each architecture across the measured communication endpoints.

Figure 5: Architecture-dependent computational performance and its relationship to network characterization. a , Mean normalized performance of the nine neuronal architectures across the three computational tasks: close-frequency decoding, temporal-order discrimination, and fading memory. b , Spearman correlations across architecture means ( $n=9$ ) relating task performance to IC 3 and transfer entropy (TE). c , Change in frequency- and temporal-order decoding when the readout is expanded from the stimulated neuron(s) to include first-hop neurons or the full network. d , Change in normalized task performance between native and degree-preserving rewired networks for frequency and temporal-order decoding; positive values indicate higher performance in the native architecture. e , Strength of the architecture effect for each task ( $\eta^{2}$ ), the descriptive partial $\eta^{2}$ for the architecture-by-task interaction, and Spearman correlations between architecture rankings across tasks ( $n=9$ ).

Resource	Designation	Source	Identifier
Simulation and analysis code	Neurons-on-a-chip analysis package	This study	DOI: 10.5281/zenodo.22723783
Numerical source data	Nine-architecture simulation dataset and figure-level data	This study	DOI: 10.5281/zenodo.22723783
Python	Analysis environment	Python Software Foundation	v3.9.6
NumPy	Numerical computation	NumPy developers	v1.26.1
pandas	Data handling	pandas developers	v2.3.3
SciPy	Scientific computation	SciPy developers	v1.11.3

Table 1: Key computational resources used in this study.

Explore similar work

Identifying structural design principles shaping the computational abilities of recurrent neural networks
Jun 22, 2026Tom Talpir, Elad SchneidmanNeural Circuits Recurrent Neural Networks
Department of Brain Sciences, Weizmann Institute of Science, Rehovot, Israel
Nanoparticle Networks for Neuromorphic Computing
Jul 30, 2026Jonas Mensing, Wilfred G. van der Wiel, Andreas HeuerNeuromorphic Computing Molecular Property Prediction
Theory of Complex Systems Group, Institute of Physical Chemistry, University of Münster, Münster, Germany · NanoElectronics Group, MESA+ Institute for Nanotechnology, Center for Brain-Inspired Nano Systems (BRAINS), University of Twente, Enschede, Netherlands · Institute of Physics, University of Münster, Münster, Germany
Scaled Hippocampus-inspired Neural Networks on Neuromorphic Memristive Hardware
Sep 14, 2026Joseph A. Kilgore, Jeffrey D. Kopsick, Zahin Ahmed +2Neuromorphic Computing Reram Crossbar
Department of Electrical and Computer Engineering, George Washington University, Washington, 20052, USA · Center for Neural Informatics, Structures, & Plasticity, George Mason University, Fairfax, 22030, USA

Jun 22, 2026, q-bio.NCJ/K move · Enter open · S save

Identifying structural design principles shaping the computational abilities of recurrent neural networks

Tom Talpir, Elad Schneidman

Department of Brain Sciences, Weizmann Institute of Science, Rehovot, Israel

Neural Circuits Recurrent Neural Networks Neural Network Recurrence

Open paper AskPDF arXiv

Understanding how the architecture of neural networks shapes the computations they carry is a central challenge in neuroscience and machine learning. While specific circuit architectures have been linked to particular network computations and theoretical bounds on expressivity of broad classes of networks have been found, we are still missing general principles connecting the structure of finite networks to their computational capabilities. Here, we characterize the computational abilities of recurrent neural networks as a function of their connectivity by training a large collection of different networks to compute a large set of Boolean functions. For small networks, we constructed the complete ``catalogs'' of network-function performance, which revealed that computational capacity varies widely across architectures and that most networks show poor performance, and most functions are hard to compute. However, we show that having local 2- and 3-cycles in a network strongly enhances its computational ability, and networks with such cycles are often the minimal architectures that can solve particular functions. We further show that a small set of structural statistics accurately predict networks' performance. Extending our analysis to large networks showed that typical networks fail even to approximate a randomly selected function. Surprisingly, adding a small number of sparsely connected biologically-inspired interneurons to the network dramatically increases computational capacity. As in small networks, adding short cycles improved networks' capacity, outperforming acyclic or reachability-matched controls. Thus, our results identify local cycles as design principles linking neural connectivity to computational power, and offer a general framework to explore structure-function relations in computing networks.

Abstract

Explore similar work

Identifying structural design principles shaping the computational abilities of recurrent neural networks

Nanoparticle Networks for Neuromorphic Computing

Scaled Hippocampus-inspired Neural Networks on Neuromorphic Memristive Hardware