Project 07
A compact cooling concept that routes liquid between thermoelectric legs instead of attaching an external cold plate. A COMSOL multiphysics comparison found up to 57.9% higher simulated COP than a conventional configuration under favorable operating conditions. Research accepted for IEEE IECON 2026.
Project overview
Compact processors and power devices need localized cooling without the bulk of a separate heat sink. This research routes coolant through all the spaces between thermoelectric legs inside a sealed module, allowing convection alongside Peltier cooling. A conventional thermoelectric cooler with an external straight-fin microchannel plate served as the numerical baseline.
01 / Architecture
The proposed ceramic-enclosed assembly places the coolant path between the thermoelectric legs rather than in a separate hot-side cold plate; liquid enters one face and leaves the opposite face. The paper describes 40 bismuth-telluride legs per unit (20 p–n pairs), with 1.5 mm square legs and 0.72 mm inter-leg channels. Ten units form an approximately 46 × 46 × 4 mm assembly; this remains a proposed design, not a fabricated prototype.
02 / Numerical method
Coupled laminar flow, conjugate heat transfer and thermoelectric effects in steady-state COMSOL Multiphysics models of the integrated (G1) and conventional (G2) layouts. Ran 80 simulated cases across two cooling-surface temperatures (293.15 and 313.15 K), four voltages (0.05–0.5 V) and five Reynolds numbers (100–2000), under matched conditions. Checked mesh independence to below 1% change in cooling flux and compared the model with published numerical results from Ruan et al.; cooling power and COP differed by 2.2% and 2.9%, respectively.
| Validation measure | Published reference | Present model | Deviation |
|---|---|---|---|
| Cooling power | 0.45 W | 0.44 W | 2.2% |
| COP | 0.60 | 0.58 | 2.9% |
Table 02-01 — Numerical model comparison against Ruan et al. as reported in Table IV of the paper; not experimental validation of the proposed hardware.
03 / Findings
At 0.05 V and Re = 2000, simulated COP reached 151.42 for the integrated layout versus 95.87 for the conventional layout: a 57.9% improvement at that operating point. At 0.5 V and Re = 2000, simulated electrical input was 0.551 W versus 0.697 W (20.9% less); Joule-heating losses were around 22–26% lower over moderate-to-high cooling loads. The paper estimates more than 50% less overall cooling-system form factor by eliminating the separate cold plate. This is a packaging estimate, not a prototype measurement.
The trade-off matters: at 0.05 V and Re = 1000, the conventional design absorbed 1.411 W at the cooling surface versus 0.760 W for the integrated design; higher COP does not always mean more instantaneous cooling.
04 / Research status
Results are numerical, not measurements from a fabricated cooler. The intended implementation calls for a dielectric-compatible fluid; the comparison simulations used a published-reference graphene-nanoplatelet nanofluid instead. Sealing, fluid compatibility and performance require experimental verification. At high voltage and low flow, Joule heating can overwhelm cooling; the COP advantage is strongest at low voltage and sufficiently high flow.
The cover image is a concept illustration, not a photograph or an exact representation of the simulated geometry. The diagrams and charts above are reproduced from the supplied research paper.