A dosa and a motorcycle radiator are the same kind of object: a thin, wide sheet whose job is to move heat across its thickness and out of a large surface. On a 175 °C tawa, about 98% of the heat that cooks a dosa arrives by conduction through a 2 to 4 mm layer of batter, and the time it takes grows with the square of that thickness. A radiator fin works the same way in reverse, and its effectiveness depends on conductivity times thickness.
- 98% of the heat that cooks a dosa arrives by conduction from the tawa.
- Cooking time grows with thickness squared: about 19 s for 2 mm of batter, 76 s for 4 mm.
- Dosa conducts at 0.42 W/m·K (Venkateshmurthy and Raghavarao, 2015), about 500 times worse than aluminium.
Calculated in this post from 0.42 W/m·K for dosa (that study) and 205 W/m·K for aluminium (Wikipedia list of thermal conductivities, which gives 204 to 250 W/m·K for pure aluminium): 205 ÷ 0.42 ≈ 490.
- Fin quality depends on one product, k × thickness: a 3 mm dosa matches a 0.1 mm steel fin.
- A 0.1 mm aluminium fin is about 96% efficient; the 3 mm dosa about 62%.
- The link breaks at fin spacing: radiator gaps of 1–2 mm are set by 0.9 mm boundary layers.
By that measure, a 3 mm dosa is about as good a cooling fin as a 0.1 mm sheet of steel, though about 16 times worse than a 0.1 mm aluminium fin.
Calculated in this post from conductivity times thickness: 0.42 W/m·K × 3 mm for the dosa (Venkateshmurthy and Raghavarao, 2015), and 16 W/m·K (stainless steel) or 205 W/m·K (aluminium) × 0.1 mm for the fins (Wikipedia list of thermal conductivities; 0.1 mm is the fin thickness in Oshionebo et al., 2025). The fin thicknesses are example inputs.
This is a tier-4 "collision", where two unrelated things are paired and the real physics linking them is proved, number by number. The dosa side comes from a study that measured an industrial dosa machine; the radiator side from fin theory and a 2025 radiator study. The verdict at the end says where the link holds and where it breaks.
How does heat actually cook a dosa?
Almost entirely by conduction from the hot plate up through the batter. K. Venkateshmurthy and K. S. M. S. Raghavarao at CSIR-CFTRI in Mysuru measured the heat flows in an automatic dosa machine and found:
- Conduction from the plate: about 98% of the heat.
- Radiation from the hood: about 1.5%.
- Convection: negligible.
The machine made 360 dosas an hour on a 1.2 m rotating plate. Each came from 100 g of fermented rice and urad dal batter spread to 20–23 cm across and 2–4 mm thick, baked for about 120 s on a plate preheated to 443–453 K (170–180 °C).
Using a commercial dosa machine with a 1.2 m, 0.12 m thick rotating hot plate, the authors isolated each heat transfer mode with insulating covers. They measured dosa's thermal conductivity at 0.42 W/m·K (three runs of 0.40–0.44), its emissivity at 0.31 and its specific heat at 1.83 kJ/kg·K; moisture fell from about 65% in the batter to about 35% in the dosa, and the machine's thermal efficiency was about 51%. Conduction carried about 98% of the heat. The limit: it is one machine with one batter recipe, and the measured temperature profiles ran below theory because evaporation cools the batter.
Read the paper (free)The batter is mostly water, and most of the heat goes into boiling it off rather than warming it. A 100 g dosa leaving the plate at 72–76 g has lost about 24–28 g, mostly as steam. Evaporating that much water takes \( 0.026 \times 2.26\times10^6 \approx 59 \) kJ, which over a 120 s bake is close to 490 W. The study's own estimate of latent heat use was about 532 W, so the arithmetic agrees.
Why does a thick dosa take so much longer?
Because heat diffusing through a slab takes time proportional to the thickness squared. The governing law is Fourier's heat equation, and its characteristic time is:
Here α is the thermal diffusivity, how fast temperature changes spread. The study gives k and cp; it does not give the batter's density, so I assume 1,100 kg/m³ for a wet rice-and-dal batter (an estimate):
- Diffusivity. \( \alpha = 0.42 / (1100 \times 1830) = 2.1\times10^{-7} \) m²/s.
- 2 mm dosa. \( t = (0.002)^2 / 2.1\times10^{-7} = 19 \) s.
- 3 mm dosa. \( t = 9\times10^{-6} / 2.1\times10^{-7} = 43 \) s.
- 4 mm dosa. \( t = 1.6\times10^{-5} / 2.1\times10^{-7} = 76 \) s.
- Uttapam at 8 mm. \( t = 6.4\times10^{-5} / 2.1\times10^{-7} = 305 \) s, which is why it needs a lid and a flip.
The ratio between the 2 mm and 4 mm cases is exactly 4, the square law at work. A dosa master who spreads thinner is buying time quadratically. Spreading also raises the surface area for the steam to escape, which keeps the bottom from scorching before the top sets.
Car engineers met the dosa's lesson early. In March 1900, Emil Jellinek, a wealthy businessman in Nice, raced two Daimler cars at Nice–La Turbie; after a fatal crash, he pushed Daimler for a longer, wider car with a lower centre of gravity, and ordered 36 of them for 500,000 marks. The result, the 1901 Mercedes 35 hp, carried Wilhelm Maybach's honeycomb radiator: 8,070 square tubes, each 6 mm across, holding 9 litres of pumped water with a fan behind.
Thousands of thin passages instead of a few fat ones: the same move as spreading batter thin. At Nice–La Turbie in March 1901 the Mercedes cars set a record average of 51.4 km/h, against 31.3 km/h before, and the layout is close to the radiators still used today.
Wikipedia: Mercedes 35 hpSee the data as a table
| t = L²/α (s) | |
|---|---|
| 1 mm | 4.8 |
| 2 mm | 19 |
| 3 mm | 43 |
| 4 mm | 76 |
| 5 mm | 119 |
| 6 mm | 171 |
| 8 mm | 305 |
Source: Calculated with k = 0.42 W/m·K and cp = 1.83 kJ/kg·K from Venkateshmurthy and Raghavarao (2015); batter density 1,100 kg/m³ assumed · Order-of-magnitude diffusion times. Evaporation and drying change the properties as the dosa cooks.

How does a radiator do the same job in reverse?
A radiator moves heat from hot coolant inside flat tubes, through a metal wall, into thin fins, and out into the air blowing past. The fins exist for one reason: air is a poor heat carrier, so you need a huge surface to dump heat into it. The rate at which a surface sheds heat to air is Newton's law of cooling:
The heat transfer coefficient h for air flowing through a radiator core is low, of order 50–100 W/m²·K (an estimate for forced air at road speeds), compared with thousands for the coolant side. Since h is small, A must be large, and the cheapest way to add area is thin fins packed closely.
That is the dosa's trick too: a 100 g lump of batter in a ball would cook from outside in over many minutes, while spread into a thin, wide sheet it cooks in two.
A modelling study of a car radiator with copper fins 0.1 mm thick on brass tubes (644 tubes in six rows, 3.51 mm hydraulic diameter, core 460 × 410 × 84 mm), simulated at a highway air speed of about 24 m/s. Eight nanoparticle-laden coolants raised fluid conductivity by up to 6.6% and the peak Nusselt number by 3.5% over water, at the cost of 6–15% more pressure drop and pumping power.
The limit: it is a simulation, and some figures differ between the abstract and conclusions (for example the size of the coolant temperature drop), so treat its headline gains cautiously. Its geometry is a useful real-world reference for fin thickness.
Read the paper (free)What makes a good fin?
A fin is good if heat can travel along it faster than the air pulls heat off its surface, so that its tip is nearly as hot as its root. The measure is fin efficiency, from the classic one-dimensional fin equation with an insulated tip:
Here L is the fin length from the hot tube to the middle of the gap, t its thickness and k its conductivity. For a thin plate fin, the perimeter-to-area ratio P/Ac is 2/t, which gives the second form. Notice that k and t appear only as the product k·t: the fin's ability to conduct sideways. That product is the whole story of this post.
Work it with these example inputs, h = 80 W/m²·K and L = 4 mm (a fin 8 mm tall between two tubes):
- Aluminium fin, 0.1 mm. \( k t = 205 \times 0.0001 = 0.0205 \) W/K. \( m = \sqrt{160/0.0205} = 88 \) m⁻¹, \( mL = 0.35 \), \( \eta = \tanh(0.35)/0.35 = 0.96 \).
Calculated in this post from k = 205 W/m·K for aluminium (Wikipedia list of thermal conductivities) and a 0.1 mm fin thickness (as in Oshionebo et al., 2025), with these example inputs.
- Steel fin, 0.1 mm. \( k t = 16 \times 0.0001 = 0.0016 \) W/K. \( m = 316 \) m⁻¹, \( mL = 1.26 \), \( \eta = 0.67 \).
Calculated in this post from k = 16 W/m·K for stainless steel (Wikipedia list of thermal conductivities) and a 0.1 mm fin thickness, with these example inputs.
- Dosa "fin", 3 mm. \( k t = 0.42 \times 0.003 = 0.00126 \) W/K. \( m = 356 \) m⁻¹, \( mL = 1.43 \), \( \eta = 0.62 \).
Calculated in this post from k = 0.42 W/m·K (Venkateshmurthy and Raghavarao, 2015) and a 3 mm thickness, inside that study's 2 to 4 mm range, with these example inputs.
- Dosa "fin", 0.1 mm. \( k t = 0.000042 \) W/K. \( m = 1{,}952 \) m⁻¹, \( mL = 7.8 \), \( \eta = 0.13 \). Almost useless.
So a 3 mm dosa conducts sideways about as well as a 0.1 mm steel fin (k·t of 0.00126 against 0.0016), and about 16 times worse than aluminium.
Calculated in this post from the k·t values above: 0.0205 ÷ 0.00126 ≈ 16, using the dosa's measured 0.42 W/m·K (Venkateshmurthy and Raghavarao, 2015) and the metals' conductivities from the Wikipedia list of thermal conductivities.
A dosa thick enough to be a decent fin is too thick to cook quickly, and one thin enough to cook fast is a terrible fin.
That tension is the real link: in both, thickness sets the conduction path, and the material's k decides how thin you can afford to go.
Fin efficiency calculator: metal or batter
Why are radiator fins spaced only a millimetre or two apart?
Because each fin is wrapped in a thin layer of slow air, and if the fins get too close those layers merge and choke the flow. Air touching a surface sticks to it, and a boundary layer of slower air grows along the fin. For laminar flow, Blasius's solution gives its thickness:
- Flow conditions, with these example inputs. Air at about 10 m/s through the core, kinematic viscosity \( \nu \approx 1.7\times10^{-5} \) m²/s for warm air (about 16.9×10⁻⁶ m²/s at 40 °C, per Engineering ToolBox), and a flow depth x = 20 mm.
- Reynolds number. \( Re = 10 \times 0.02 / 1.7\times10^{-5} \approx 11{,}800 \).
- Boundary layer. \( \delta \approx 5 \times 0.02 / \sqrt{11{,}800} = 0.92 \) mm.
- What it means. With fins about 2 mm apart, the layers on facing fins nearly meet by the back of the core. Packing them tighter adds area but each square millimetre works less, and the pressure drop rises.
Radiator makers fight this with louvres, small slats cut into each fin that restart the boundary layer every millimetre or so, keeping it thin.
This is where the dosa analogy runs out. A tawa holds one dosa with open air above it, so there is no spacing problem. A radiator is hundreds of "dosas" stacked a couple of millimetres apart, and spacing becomes a second design decision that the cook never faces.
| Property | Dosa on a tawa | Motorcycle radiator fin |
|---|---|---|
| Heat flows | From 170–180 °C plate into the sheet | From about 90–110 °C coolant out of the sheet |
| Thickness | 2–4 mm | About 0.1 mm |
| Thermal conductivity | 0.42 W/m·K (measured) | About 205 W/m·K (aluminium) |
| k × thickness | 0.0008–0.0017 W/K | About 0.02 W/K |
| Main heat path | Conduction through the thickness (98%) | Conduction along the fin, then convection to air |
| Where the heat ends up | Evaporating water (about 490–530 W) | Moving air |
| Spacing problem | None, one sheet with open air | Yes, fins 1–2 mm apart, boundary layers merge |

How sensitive is the comparison?
- Batter density. If the batter is 1,000 or 1,200 kg/m³ instead of 1,100, the diffusion times move by about 9% either way. The square law does not change.
Calculated in this post from the diffusivity α = k ÷ (density × specific heat) (Wikipedia: thermal diffusivity), so the time scales with density: 1,000 ÷ 1,100 and 1,200 ÷ 1,100 are −9% and +9%. The 1,100 kg/m³ is an assumed example input; the study gives no batter density.
- Heat transfer coefficient. Doubling h to 160 W/m²·K drops the aluminium fin to 92% efficiency and the 3 mm dosa to 48%. The ranking holds.
Calculated in this post from the fin formula (Wikipedia: fin (extended surface)) with these example inputs, using k = 205 W/m·K for aluminium and 0.42 W/m·K for dosa as above.
- Fin length. Taller fins (L = 6 mm) push the aluminium fin to 92% and the steel fin to 50%, which is why steel radiators need shorter, thicker fins and why aluminium won.
Calculated in this post from the fin formula (Wikipedia: fin (extended surface)) with these example inputs, using k = 205 W/m·K for aluminium and 16 W/m·K for stainless steel (Wikipedia list of thermal conductivities).
- Drying. As a dosa dries and fills with tiny steam pockets, its conductivity falls. A crisp dosa is a worse conductor than the batter it came from, which is part of why the top stays soft while the base browns.
So is a dosa really a radiator?
In the way that matters to an engineer, yes. Both are thin sheets designed so that heat crosses a short path quickly and leaves through a large area, and both obey the same diffusion and fin equations, in which conductivity times thickness is the deciding product.
The analogy is honest about one more difference: in a dosa, most of the heat leaves as steam, while in a radiator it leaves as warm air. Evaporation is the better cooler of the two, which is why a radiator boils over when the coolant is low, and why the engine's pressure cap exists. The pressure-cap story is in the pressure cooker and black hole post, and what happens when cooling fails under sustained load is in the redline-for-an-hour what-if.
A dosa and a radiator fin obey the same heat equations, and the same product, conductivity times thickness, decides how well each works: 0.00126 W/K for a 3 mm dosa against 0.0016 for a 0.1 mm steel fin and 0.0205 for aluminium, giving fin efficiencies of 62%, 67% and 96%. Batter thickness and fin thickness really are the same decision, governed by the square law (19 s for 2 mm of batter, 76 s for 4 mm). The link breaks at fin spacing: a dosa has open air above it, while a radiator's 1–2 mm gaps are set by boundary layers about 0.9 mm thick.
For the batter itself, Rasa covers the fermentation science of idli and dosa batter. On the bike side, the difference between air and liquid cooling shows up in single versus twin-cylinder engines, and the full cooling-system checklist for wet weather is in the monsoon maintenance guide.
What's new
- Study models nanofluid coolants in ruffled-fin car radiators Scientific Reports, Dec 2025
- Vertically aligned carbon nanotube heat sink sheds heat by conduction, convection and radiation together Advanced Science, Nov 2026
What if next?
- What a tabla can teach you about engine knock. Another round, thin, stretched sheet, this time ringing at its own frequencies.
- Why your chain and a pearl necklace wear out the same way. Small clearances at many joints, adding up.
- A cricket ball's swing and your helmet's buffeting are one phenomenon. The same boundary layers that crowd radiator fins, on a ball and a visor.
Questions people ask
Why does a thick dosa not cook properly?
Heat diffuses through batter in a time proportional to the thickness squared. Doubling thickness from 2 mm to 4 mm roughly quadruples the time, from about 20 to 76 seconds, so the base can brown before the top sets.
What temperature should a tawa be for dosa?
In the CFTRI study of an automatic dosa machine, the plate was preheated to 443–453 K, which is 170–180 °C, and each dosa baked for about two minutes.
Why are radiator fins so thin?
Air carries heat poorly, so a radiator needs a large surface. Thin aluminium fins add area cheaply while still conducting heat well along their length; at 0.1 mm an aluminium fin of typical height is about 96% efficient.
Why are radiator fins louvred?
Louvres cut into the fins break up and restart the boundary layer of slow air on each fin, keeping it thin. That raises the heat transfer coefficient, at the cost of some extra pressure drop.
Sources
- Venkateshmurthy, K. and Raghavarao, K. S. M. S.: Analysis of modes of heat transfer in baking Indian rice pan cake (Dosa), J Food Sci Technol 52(8) (2015) (source for plate temperature, thickness, 98% conduction, k = 0.42 W/m·K, cp = 1.83 kJ/kg·K, moisture and latent heat)
- Oshionebo, F. D. et al.: Optimized nanofluid coolants enhance thermal performance in ruffled fin automotive radiators, Scientific Reports 15 (2025) (source for 0.1 mm fins and radiator core geometry)
- Wikipedia: Fin (extended surface) (source for the fin equation and efficiency tanh(mL)/mL)
- Wikipedia: Blasius boundary layer (source for δ ≈ 5x/√Re)
- Wikipedia: List of thermal conductivities (source for aluminium, copper and steel conductivities)
- Wikipedia: Mercedes 35 hp (source for Jellinek's order, Maybach's 8,070-tube honeycomb radiator and the 1901 Nice–La Turbie record)
- Park, J. et al.: Multiscale freestanding vertically aligned CNT heat sink enabling multimodal heat dissipation, Advanced Science (2026)

