Thermosensitive Neuronal Networks: Theory & Applications

• Thermosensitive neuronal networks are neural systems whose dynamics and computation are modulated by temperature and thermal noise, linking biophysical and theoretical models.
• They integrate mechanisms from channel kinetics, photo-thermal stimulation, and neuromorphic device physics to achieve controlled stochasticity and adaptive homeostasis.
• Applications include neuromorphic hardware with thermal memristors and quantum thermometry platforms that offer precise temperature monitoring and enhanced computational capabilities.

Thermosensitive neuronal networks are neural systems, either biological or artificial, whose structure, dynamics, or computational properties are fundamentally modulated by temperature or by thermal noise. The temperature sensitivity manifests through physical, biophysical, or circuit-level mechanisms and has been formalized both experimentally and theoretically, with explicit connections to statistical mechanics, biophysical models, and neuromorphic hardware.

1. Thermodynamic Foundations and Temperature Mapping

The analogy between neural network dynamics and thermodynamics is operationalized via mappings between neural variables and thermodynamic quantities. In two-state stochastic neural models, the stochasticity is controlled by a “temperature” parameter $T$ appearing in the logistic update rule: $P = \frac{1}{1 + \exp(-e_g/T)}$ where $e_g$ is the driving “bandgap” (distance to threshold). For leaky integrate-and-fire neurons over a fixed window $T_W$, this mapping is extended by introducing stochastic Poisson excitatory and inhibitory spike trains as sources of effective “thermal noise” (Merolla et al., 2010). The effective temperature is analytically given by: $$T = \frac{T'_W \sigma}{2 (\ln 2)^2 \tau_m f((u_t - \mu)/\sigma)}$$ with $f(x) = \sqrt{\pi} \exp(x^2)[1+\mathrm{erf}(x)]$, where $\sigma^2$ is the noise variance induced by Poisson input rates and synaptic weights, $\mu$ is the mean depolarization, and $T'_W$ is an effective integration window.

Temperature controls the stochasticity of the network—higher effective temperature (via increased noise) leads to more probabilistic, less deterministic spiking, and smooths the transfer (activation) curves, mapping biological processes onto statistical mechanics models of neural computation.

2. Biophysical and Functional Mechanisms of Thermosensitivity

Thermal sensitivity of neurons arises both from intrinsic ionic and structural processes and from network-level integration:

• Channel Kinetics and Bifurcation: In biological thermoreceptors (e.g., snake pit organs), many noisy thermal TRP channel events are integrated by voltage dynamics exhibiting a saddle–node bifurcation. Near the bifurcation, the firing rate $f$ varies as $f \sim \delta^{1/2}$, where $\delta$ is the distance from the critical point. Feedback tunes the control parameter (half-activation voltage) to keep the system at this point of maximal sensitivity, allowing milli-Kelvin thermal changes to be amplified into large changes in action potential rates (Graf et al., 2023).
• Photo-Thermal Stimulation: Neural excitation can also be driven by rapid temperature transients, with the induced depolarizing current $I \propto dT/dt$. Short, high-intensity heating (e.g., via laser) is effective, as experimentally validated in both cortical neurons and peripheral nerves (Farah et al., 2012). The scaling law connecting action potential threshold to the maximal rate of temperature rise provides an empirical foundation for photo-thermal neuromodulation.
• Mechanical–Thermal Coupling: In nociceptive systems, local heat bursts generated by mechanical damage (modeled as rapidly diffusing heat) transiently activate thermo-TRP channels (e.g., TRPV1/V3), coupling mechanical insult to thermal pain (Vincent-Dospital et al., 2020).
• Cytoskeletal Response: Neuronal stiffness is strongly and reversibly temperature-sensitive. Lowering ambient temperature from $37^\circ$C to $25^\circ$C produces a dramatic increase in the soma’s elastic modulus and shifts the dominant cytoskeletal component from tubulin to actin, an effect regulated by myosin-II. Stiffness modulation is fully reversible and pharmacologically tunable (Spedden et al., 2013).

3. Thermosensitivity in Artificial and Neuromorphic Circuits

Thermosensitive properties are harnessed for computation and regulation in neuromorphic hardware:

• Thermal Memristors and Neuristors: Devices built from metal–insulator transition (MIT) materials (e.g., VO$_2$) exhibit temperature-dependent resistance with hysteresis, enabling both leaky integrate-and-fire functionality and thermal memory (Ben-Abdallah, 2017, Valle et al., 2019, Velichko et al., 2019, Qiu et al., 2023). The “membrane potential” is mapped to local temperature; the neuron’s firing threshold corresponds to a thermal barrier. External voltage biases and the geometry of devices allow for threshold tuning and for exploiting the superposition of thermal pulses in networks.
• Computational Implications: Arrays of coupled thermal neuristors show reconfigurable excitatory–inhibitory dynamics, all-or-nothing spiking, refractory periods, and leaky integration—direct analogues of neural network motifs (Qiu et al., 2023, Zhang et al., 2023). These networks perform reservoir computing with high accuracy in image and time series tasks, and demonstrate that long-range order (from time-nonlocal thermal memory) and synchronization phenomena can arise even in the absence of criticality.
• Homeostatic Regulation: Metal–oxide memristors exhibit programmable temperature sensitivity in their resistance states (Abbey et al., 2021). By designing memristor synapses with adjustable thermal dependence, global homeostatic control over network excitability can be implemented, mimicking slow, broad-brush biological homeostatic processes.
• Bioinspired Thermoregulation Circuits: Spiking electronic circuits based on temperature-sensitive integrate-and-fire neurons, with feedback and feedforward loops, reproduce thermostatic control akin to physiological homeostasis. Feedforward (anticipatory) and feedback (compensatory) signals enable the system to maintain core temperature robustly against changing ambient conditions, a principle distilled into hybrid dynamical system models (Rosito et al., 1 Sep 2025).

4. Collective and Network-Level Phenomena

Thermodynamic paradigms extend to large ensembles:

• Criticality and Phase Transitions: Experimental studies on retinal ganglion cell populations reveal near-criticality: the entropy per neuron closely matches the energy per neuron (S ≈ E), and the heat capacity shows a pronounced peak at the effective temperature (T = 1 in model units) (Tkacik et al., 2014). Theoretical models using Ising-inspired spin networks with a temperature-like control parameter $T$ show sharp transitions in synchrony and sensitivity near a critical $T_c$, with critical fluctuations and long-range correlations—features connected to the “critical brain hypothesis” (Sarmastani et al., 8 Jun 2025).
• Hybrid and Multimodal Coupling: Networks with both electrical and chemical synapses, modeled via hybrid coupling of FitzHugh–Nagumo neurons, can access a panoply of dynamic states—synchrony, incoherence, chimera, and traveling wave states. Temperature (via cell size or noise) and external electric field frequency precisely control the balance of coherence, with low-frequency external fields selectively entraining localized synchronized clusters (chimeras), and chemical synapses supporting traveling chimera states. These phenomena provide a route to targeted neuromodulation (Nguessap et al., 18 Sep 2025).
• Thermodynamic Network Analysis: Viewing the time-varying brain connectome as a thermodynamic system—quantifying spectral core entropy, node and internal energy, and temperature—provides novel metrics for detecting brain state transitions and discriminating control from pathological conditions (e.g., dynamic differences in autism) (Qin et al., 2 Sep 2024).

5. Experimental Platforms and Quantitative Measurement

Various platforms enable quantification and control of thermosensitivity:

• Quantum Thermometry: Nanodiamond quantum sensors using optically detected magnetic resonance (ODMR) precisely track $<0.1$°C temperature changes during neuronal firing. The frequency shift in the NV center splitting directly encodes local temperature, providing a sensitive, multiplexed readout of energy utilization in neural circuits (Petrini et al., 2022).
• Microfluidic Neural Ensemble Engineering: Modular microfluidic culture platforms organize cortical neurons into hierarchical ensembles with tunable intermodular coupling, suppressing global synchrony and fostering diverse neuronal activity patterns. Such devices are amenable to integration with thermal control for dissecting the effects of temperature on ensemble dynamics and plasticity (Murota et al., 29 May 2024).

6. Computational and Theoretical Implications

Thermosensitive neuronal networks provide a unified framework bridging physics, hardware, and neural computation:

• Statistical Mechanics and Information Processing: The mapping between statistical physics (temperature, entropy, energy) and neural dynamics enables analytic approaches to understand collective behavior, noise-induced transitions, and the emergence of computation via probabilistic state updates (e.g., Boltzmann machines, high-dimensional attractor dynamics) (Merolla et al., 2010, Tkacik et al., 2014).
• Principle of Amplification via Bifurcation: Biological sensors integrate signal and noise near a dynamical bifurcation, effectively amplifying weak input by exploiting critical slowing down or heightened susceptibility, a motif extensible to artificial systems (Graf et al., 2023).
• Role in Homeostasis and Adaptivity: Circuit- and device-level thermosensitivity supports slow, self-stabilizing adaptation in hardware and artificial networks, echoing the homeostatic regulation of biological neurons (Abbey et al., 2021, Rosito et al., 1 Sep 2025).

Thermosensitive neuronal networks, across biological, synthetic, and hybrid settings, constitute a class of systems in which thermal fluctuations, phase transitions, and temperature-regulated mechanisms are inseparable from computation, plasticity, and adaptation. Their paper elucidates deep connections among physical law, neural information processing, control, and the engineering of future neuromorphic platforms.