Thermopile

A thermopile measures light by the heat it deposits. An absorbing layer is attached to the "hot" junctions of a chain of thermocouples, while the "cold" junctions sit on a heat sink. Absorbed power raises the absorber temperature by $\Delta T$, and each junction pair produces a Seebeck voltage $S\Delta T$; $N$ pairs in series add up. Because the response depends only on absorbed energy, it is nearly independent of wavelength wherever the coating is black, from the ultraviolet to the far infrared, which makes the thermopile the standard sensor in broadband optical power meters and in non-contact infrared thermometers. It needs no bias and produces no dark current, and it responds to steady (DC) light. The price is speed and sensitivity: time constants range from about 10 ms for small micromachined sensors to a second or more for laser power meter heads, and the noise floor is far above that of a photodiode.

Responsivity and time constant

In steady state the absorber temperature rise is set by the absorbed power $\eta P$ and the thermal conductance $G$ to the heat sink, and the output voltage is

$$V = N S\,\Delta T = \frac{N S\,\eta P}{G}.$$

The response time is the ratio of heat capacity to conductance, $\tau = C_{th}/G$. As an illustration, 100 junction pairs of 100 µV/K each, an absorptance of 0.9 and $G = 10^{-4}$ W/K give a responsivity of 90 V/W; with a heat capacity of 10⁻⁶ J/K the time constant is 10 ms. Lowering $G$ raises the responsivity and slows the detector in equal proportion, so the responsivity-bandwidth product is fixed by the heat capacity. Micromachined sensors reduce both by suspending the absorber on a thin membrane, often with polysilicon or bismuth-antimony thermocouples, and many are packaged in a sealed can with an infrared window.

Noise and detectivity

With no bias current, the dominant noise is the thermal noise of the thermopile's own resistance, $\sqrt{4kTR}$, with little 1/f noise. For a 50 kΩ element at 296 K that is 28.6 nV/√Hz; divided by 90 V/W it gives a noise equivalent power of about 0.32 nW/√Hz, and for a 1 mm² element a specific detectivity of about 3 × 10⁸ Jones. That is typical of commercial devices and between one and two orders of magnitude below the room-temperature limit for thermal detectors.

Laser power meters

Thermopile heads for laser power use a large absorbing disc with thermocouples arranged radially, so the output depends on the total heat flow and only weakly on where the beam lands. They cover milliwatts to kilowatts, the upper range needing water or forced-air cooling, and survive power densities that would destroy a photodiode. Because the absorber is spectrally flat, one calibration applies across a wide band, with a correction for the coating's residual wavelength dependence. The head reads average power for both CW and pulsed lasers; its slow response integrates over pulses.

Other uses

Infrared thermometers and thermal presence sensors use small thermopiles behind an 8–14 µm window, matching the peak of blackbody emission near room and body temperature. Non-dispersive infrared gas sensors use two thermopiles behind narrowband filters, for carbon dioxide near 4.26 µm and a reference band, since the flat response and DC stability suit a slowly modulated lamp. Small thermopile arrays give low-resolution thermal images.

Pitfalls

Since the thermopile measures a temperature difference, anything that heats the junctions unequally appears as signal: air currents, a hand near the head, or a change in ambient temperature. Power meter heads should be zeroed with the beam blocked after they reach thermal equilibrium, and readings at low power drift for seconds to minutes. Reflection and scattering from the absorber, and damage to the coating from high fluence, change the calibration. For modulated light above the inverse time constant the response falls, so chopped measurements need a chopping frequency well below $1/(2\pi\tau)$ or a pyroelectric detector, which responds to change instead.

Common questions

What is the difference between a thermopile and a thermocouple?

A thermocouple is one junction pair; a thermopile is many in series, multiplying the voltage for the same temperature rise.

Thermopile or photodiode for measuring laser power?

A photodiode is faster and far more sensitive but wavelength-dependent and typically saturates in the milliwatt to tens-of-milliwatts range without an attenuator. A thermopile suits high power, broadband or uncertain-wavelength sources and pulsed lasers where the average power is wanted.

Why is a thermopile slow?

Its signal requires the absorber to heat up, which takes $\tau = C_{th}/G$. Making it faster means a smaller heat capacity, a thinner absorber, or a larger conductance, which lowers the responsivity.

References: E. L. Dereniak, G. D. Boreman, Infrared Detectors and Systems (Wiley, 1996); A. Rogalski, Infrared and Terahertz Detectors, 3rd ed. (CRC Press, 2019); R. W. Boyd, Radiometry and the Detection of Optical Radiation (Wiley, 1983).