Fin Heat Sink

Why fins work: area, not magic

Heat leaves a hot surface into surrounding air by convection, and the rate is governed by Newton's law of cooling:

Q = h · A · ΔT

where:
  Q  = heat rejected            (W)
  h  = heat transfer coefficient (W/m²·K)
  A  = surface area in contact with air (m²)
  ΔT = surface temperature minus air temperature (K)

For a given heat source you can't easily change ΔT (the chip wants to stay near 70 °C) and you can only raise h so far before a fan gets loud or huge. The cheap, linear knob is A — surface area. A bare 40 mm × 40 mm die exposes about 16 cm² to air. Take a 40 mm aluminum block of the same footprint and 25 mm tall, slice it into forty thin fins, and the exposed area balloons past 2,000 cm² — more than a hundredfold. The same convection coefficient now removes a hundred times the heat at the same temperature rise. That is the entire idea of a fin heat sink: pack a huge convective surface into a small volume.

The reason engineers speak of "extended surfaces" rather than just "more metal" is that the metal does double duty. Each fin must conduct heat from the hot base out toward its tip, and then convect it off its two large faces. Those two processes compete, and that competition is what the fin equation captures.

The fin equation

Model a single straight fin of thickness t, width W, length L, conductivity k, attached to a base at temperature T_b, surrounded by air at T_∞. Write θ(x) = T(x) − T_∞ for the temperature excess at distance x along the fin. A 1D energy balance — conduction in, conduction out, convection off the sides — gives:

d²θ/dx² − m²·θ = 0,    m = √(h·P / k·A_c)

For a thin rectangular fin (P ≈ 2W, A_c = W·t):
  m = √(2h / k·t)

Solution (adiabatic tip):
  θ(x) / θ_b = cosh[m(L − x)] / cosh(mL)

P is the fin's perimeter, A_c its cross-section. The dimensionless group mL decides everything. When mL is small (short, thick, high-k fin), θ stays close to θ_b along the whole fin — the tip is nearly as hot as the base, and almost all the added area carries heat. When mL is large (long, thin, low-k fin), θ decays fast and the tip sits near ambient, doing nothing. The temperature you see fading from base to tip in the animation above is exactly this cosh decay.

Fin efficiency and effectiveness

Two ratios make the trade-off concrete. Fin efficiency compares the real fin's heat to that of an ideal isothermal fin held everywhere at T_b:

η_f = Q_fin / Q_ideal = tanh(mL) / (mL)

A fin with mL = 1 runs at tanh(1)/1 ≈ 0.76, so 76% efficient. Push mL to 3 and efficiency falls to tanh(3)/3 ≈ 0.33 — two-thirds of the metal is dead weight. There is little point making a fin much longer than mL ≈ 1 to 1.5; you add cost and mass for diminishing return. Fin effectiveness ε_f answers a different question — is the fin worth adding at all?

ε_f = Q_with_fin / Q_bare_base = η_f · A_fin / A_base

A fin is only worthwhile if ε_f > 1, and you generally want ε_f > 2. This is why fins live on the air side of a heat exchanger (low h, so fins help enormously) and almost never on the liquid side (high h, where the bare surface already convects well).

Aluminum vs copper, natural vs forced

	Aluminum 6063 fins	Copper C110 fins	Al base + Cu heat pipe
Thermal conductivity k	~237 W/m·K	~400 W/m·K	pipe effective > 5,000 W/m·K
Density	2,700 kg/m³	8,960 kg/m³	mixed (light fins)
Relative cost	1×	3–4×	2–3×
Fin efficiency (same geometry)	lower	higher (fins stay hot longer)	high — pipe spreads to base
Manufacturing	extruded, cheap, mass-made	skived or bonded, harder	brazed assembly
Typical use	most consumer coolers, LED, power supplies	premium CPU/GPU, lasers	high-end CPU towers, laptops

Regime	h (W/m²·K)	Optimal fin gap	Best fin shape	R achievable
Natural convection (no fan)	5–25	6–12 mm	tall, widely spaced, vertical	3–15 K/W
Forced convection (fan)	25–250	1.5–3 mm	dense, thin, low pin/plate	0.2–1 K/W
Liquid cold plate	500–10,000+	microchannels < 1 mm	micro-fins, jet impingement	0.02–0.1 K/W

Worked example: sizing an aluminum fin

An aluminum fin (k = 237 W/m·K) is 1 mm thick (t = 0.001 m) and 25 mm long (L = 0.025 m), cooled by a fan giving h = 50 W/m²·K. Is it well-proportioned, and how efficient is it?

m = √(2h / k·t)
  = √( (2 × 50) / (237 × 0.001) )
  = √( 100 / 0.237 )
  = √422
  = 20.5  m⁻¹

mL = 20.5 × 0.025 = 0.51

η_f = tanh(0.51) / 0.51
    = 0.470 / 0.51
    = 0.92   → 92% efficient

At mL ≈ 0.5 the fin is 92% efficient — nearly all of its area is pulling its weight, and lengthening it a little more would still pay off. If we instead used a thinner, taller fin (t = 0.3 mm, L = 50 mm), m climbs to ~37 m⁻¹ and mL ≈ 1.85, dropping η_f to about 0.51 — half the fin's far end is loafing. The fan-cooled forced-convection h of 50 is what keeps these fins efficient; in still air at h = 8, m falls and the same fin becomes more efficient per fin but the whole sink rejects far less heat because h itself is small.

The thermal resistance network

Designers rarely compute fin profiles by hand on the job — they collapse the whole sink into a single number, thermal resistance R in K/W, and chain resistances like resistors:

ΔT = Q · R_total

R_total = R_junction-case + R_interface + R_spreading + R_sink

  R_interface  ≈ thickness / (k_paste · A)   (TIM layer)
  R_sink       = 1 / (h · A_eff),  A_eff = A_base + η_f · A_fins

Suppose a 150 W CPU must stay below 95 °C in a 35 °C case. The allowed rise is 60 K, so R_total must be below 60/150 = 0.40 K/W. Junction-to-case might eat 0.20 K/W, the thermal paste 0.05 K/W — leaving only 0.15 K/W for the sink-plus-fan. That is an aggressive target reachable only with a dense copper-base tower and a good fan; it is why high-TDP parts ship with elaborate coolers, and why a smear of dried-out paste (raising R_interface to 0.3 K/W) alone can throttle a chip.

Fin spacing: the hidden optimum

• Too few fins: plenty of airflow but not enough area. Total area A_eff is small, so R_sink stays high.
• Too many fins: huge area, but the boundary layers growing on adjacent fins merge and choke the channel. The local h collapses and air simply detours around the block. Past the optimum, adding fins can reduce total heat rejection.
• Natural convection needs wide gaps (6–12 mm) so warm air can rise as a chimney between fins. Cram them and the buoyant flow stalls.
• Forced convection tolerates tight gaps (1.5–3 mm) because the fan pushes air through regardless, but pressure drop and fan power rise sharply as gaps shrink.
• Pin-fin vs plate-fin: pin fins (cylindrical posts) handle airflow from any direction and suit impingement fans; plate (straight) fins are cheapest to extrude and best for one dominant flow direction.

Failure modes and trade-offs

• Interface starvation. The single biggest real-world killer is not the sink but the joint to it. An air gap or pump-out of thermal paste adds a high R_interface that no fin geometry can recover. Flatness, mounting pressure and a thin, fresh TIM matter more than fin count.
• Airflow bypass. In a real chassis, air takes the path of least resistance and skirts the fin stack unless ducted. Measured h can be a fraction of the textbook value; shrouds and ducts often beat bigger sinks.
• Dust fouling. Tight forced-convection fins clog with dust, raising effective gap resistance and starving the channel — a sink that worked at install can throttle a year later.
• Over-long fins. Marketing favors tall fins, but past mL ≈ 1.5 the tips run near ambient and add mass, cost and tip-over risk for negligible heat rejection.
• Spreading resistance. When a tiny die feeds a large base, heat can't fan out fast enough — the base hot-spots directly under the die while the fin edges stay cool. Copper bases, vapor chambers and heat pipes exist to beat this spreading resistance, not to add fin area.
• Acoustic and power limits. Raising h with a faster fan trades heat for noise and electrical power. Beyond a point, liquid cooling or a phase-change loop is cheaper than ever-louder air.