Ternary Hybrid Nanofluids Show Superior Cooling Power for AI Chips in Simulation

As artificial intelligence workloads push processors and data centers to generate ever greater amounts of heat within shrinking footprints, researchers are turning to engineered fluids that can carry heat away faster than conventional coolants. A new numerical study published in Discover Nano examines a ternary hybrid nanofluid made of aluminum oxide, copper oxide, and graphene nanoplatelets dispersed in ethylene glycol, and reports that this three-particle blend outperforms both single-particle and two-particle nanofluids in simulated cooling of a heated flat surface, a configuration the authors use as a simplified stand-in for the near-wall environment of AI-chip microchannel coolers.

The study, led by A. O. Akindele with colleagues including A. W. Ogunsola, O. A. Oladapo, and A. D. Ohaegbue, models the steady, laminar, two-dimensional flow of an incompressible power-law fluid over a non-linearly stretching sheet. The working fluid is treated as a ternary hybrid nanofluid, and the model incorporates a non-uniform magnetic field applied normal to the sheet, temperature-dependent viscosity and thermal conductivity, thermal radiation treated with the Rosseland approximation, viscous dissipation, non-uniform heat generation, porous-medium resistance, and slip boundary conditions. The authors position this combined set of physical effects as a more comprehensive representation than earlier ternary hybrid nanofluid studies that assumed constant thermophysical properties and narrower sets of effects.

To solve the governing nonlinear partial differential equations, the team first converted them into ordinary differential equations using similarity transformations. They then solved the resulting system with two independent numerical techniques implemented in Mathematica 11.3: the Collocation Chebyshev Method, which expands the solution in shifted Chebyshev basis functions over the boundary-layer domain and determines unknown coefficients with the Newton-Raphson method at Gauss-Lobatto collocation points, and the Galerkin-weighted residual method, which uses exponential trial functions and integrates weighted residuals to zero. The authors report that the two methods agreed closely, with differences of less than one percent in the computed Nusselt numbers, and that residual norms fell below prescribed tolerances as collocation points increased, with relative changes in skin friction and Nusselt values becoming negligible beyond fifty collocation points.

The comparison covered three fluid configurations: a mono nanofluid containing only aluminum oxide in ethylene glycol, a binary hybrid nanofluid combining aluminum oxide and copper oxide, and the ternary hybrid nanofluid adding graphene nanoplatelets. According to the authors, the ternary blend benefits from the complementary roles of its constituents: aluminum oxide contributes dispersion stability, copper oxide offers high thermal conductivity, and graphene nanoplatelets provide exceptional heat transport efficiency and a large specific surface area. In the simulations, the ternary hybrid nanofluid achieved higher Nusselt numbers, improved heat dispersion, and lower thermal resistance than the other two fluids, and it also produced the highest velocity and temperature gradients within the boundary layer.

The parametric analysis explored how nine dimensionless parameters shape the velocity and temperature fields: the power-law index, magnetic parameter, suction parameter, viscosity parameter, radiation parameter, variable thermal conductivity parameter, space-dependent and temperature-dependent heat generation parameters, and the Eckert number. Several trends emerged. Increasing the power-law index raised the velocity near the surface, reflecting stronger momentum diffusion in shear-thickening regimes. Stronger magnetic fields suppressed velocity across all three fluids because the Lorentz force opposes fluid motion, yet the ternary hybrid nanofluid retained a comparatively higher velocity even under elevated magnetic influence, a property the authors link to the electromagnetic environments that AI chips generate during high-speed operation.

Suction, which represents fluid withdrawal through the surface, compressed the boundary layer and lowered the velocity profile for all fluids, though the ternary blend showed the least pronounced reduction, suggesting better momentum retention under high suction. On the thermal side, the temperature profile rose with increasing magnetic parameter, radiation parameter, and Eckert number, the last reflecting the conversion of kinetic energy into internal energy through viscous dissipation. In contrast, higher suction, increased variable thermal conductivity, and stronger space-dependent and temperature-dependent heat generation parameters each lowered the temperature profile in the simulations. Across all of these cases, the ternary hybrid nanofluid exhibited the highest temperature distributions, which the authors interpret as evidence of its superior heat transmission efficiency rather than a cooling deficiency, since higher convective transport capacity accompanies the elevated thermal gradients.

The authors validated their skin friction results against previously published studies and found strong agreement, supporting the reliability of the numerical approach. They also compared the computational characteristics of the two solvers. The Chebyshev collocation method converged rapidly at comparatively low computational cost for the smooth boundary-layer solutions considered here, while the Galerkin-weighted residual method showed greater robustness for strongly nonlinear interactions and varying parameter conditions at a modest increase in computational effort. This division of strengths, they suggest, offers practical guidance: collocation methods suit problems where efficiency matters and solutions remain smooth, whereas weighted residual approaches are preferable when numerical stability under demanding nonlinearity is the priority.

Importantly, the authors are explicit about the limits of what the model can claim. The formulation is an idealized two-dimensional boundary-layer flow over a stretching flat sheet in a semi-infinite domain, intended to represent local near-wall transport behavior relevant to microchannel cooling rather than the complete confined internal flow of an actual microchannel heat sink. Pressure-drop characteristics, channel confinement effects, and detailed wall interactions of practical AI-chip microchannels fall outside the study’s scope. The work is entirely numerical, with no experimental validation, and the thermophysical properties were evaluated using established mixture correlations that may not capture nanoparticle aggregation, Brownian motion, sedimentation, or interfacial thermal resistance at high particle concentrations. Nanoparticle volume fraction was held fixed throughout the investigation so that other parameters could be examined independently, and a detailed optimization of concentration is left for future work.

Within those boundaries, the findings suggest that aluminum oxide, copper oxide, and graphene nanoplatelet ternary hybrid nanofluids could serve as promising advanced coolants for compact thermal management systems, including AI-chip heat sinks, high-performance computing devices, data centers, microelectronic components, and renewable-energy thermal systems. The authors note that the enhanced thermal transport capability may contribute to improved cooling efficiency and more reliable operation of high heat-flux electronics, while cautioning that the hydraulic penalty of increased viscosity and pumping power requirements must be weighed in practical designs. The balance between thermal gain and flow resistance is a recurring theme in nanofluid engineering, and this study reinforces that any deployment decision must account for both sides of the trade-off.

Future directions outlined by the team include experimental measurements of the ternary hybrid nanofluid under realistic electronic-cooling conditions, with wall temperature, inlet-outlet temperature difference, pressure drop, heat transfer coefficient, and Nusselt number benchmarks to test the numerical predictions directly. Three-dimensional simulations of actual microchannel heat sinks with non-uniform heat flux distributions, comprehensive thermo-hydraulic optimization that jointly considers heat transfer enhancement and pumping power, and investigations of nanoparticle stability, particle-size effects, transient operating conditions, and machine-learning-assisted composition optimization are also recommended. For now, the study offers a mathematically cross-validated case that adding a third, highly conductive nanoparticle to a coolant blend can measurably improve simulated heat transfer performance, a result that, if confirmed experimentally in real channel geometries, could inform the next generation of cooling fluids for the machines that train and run artificial intelligence.

Subject of Research: Numerical simulation of ternary hybrid nanofluid flow and heat transfer for AI chip and data center cooling

Article Title: Thermal analysis of ternary hybrid nanofluids for advanced cooling of AI chips and data centers using dual numerical techniques

Article References: Akindele, A. O., Ogunsola, A. W., Oladapo, O. A., & Ohaegbue, A. D. (2026). Thermal analysis of ternary hybrid nanofluids for advanced cooling of AI chips and data centers using dual numerical techniques. Discover Nano, 21(1), Article 553. https://doi.org/10.1186/s11671-026-05051-1

Image Credits: AI Generated

DOI: 10.1186/s11671-026-05051-1

Keywords: ternary hybrid nanofluid, AI chip cooling, data centers, thermal management, graphene nanoplatelets, ethylene glycol, boundary layer flow, Nusselt number, magnetic field, numerical simulation, Chebyshev collocation, Galerkin method