Nano-thermoelectric infrared bolometers

The infrared (IR) part of the electromagnetic spectrum, where the photon wavelength is longer than that of visible light, is relevant for numerous applications ranging from thermal imaging for night vision to remote temperature measurements and chemical analysis using infrared spectroscopy.^1–3 IR imaging and spectroscopy can be utilized, for example, in the detection of cancerous cells,⁴ in thermography in medicine, biology, and sports^5,6 as well as in industrial applications such as bioprocess monitoring,⁷ and in dynamic material studies.⁸ The thermal emission of most objects in our ambience is strongest in the long-wave infrared range (LWIR), typically at wavelengths of 8 µm–15 µm, thereby making this range ideal for IR applications as fully passive operation can be achieved.

In the IR range, the most common detector types are quantum detectors based on electron–hole pair generation and thermal detectors (bolometers) based on radiation-induced heating of an absorber. Quantum IR detectors, either photovoltaic or photoconductive, typically have higher sensitivity than bolometers but require cooling from room temperature down to cryogenic temperatures to reach the maximum performance, especially in the LWIR range where photon energies are relatively small. The quantum detectors for the LWIR range require also relatively expensive and toxic materials, such as HgCdTe and InAsSb. As quantum detectors transduce the IR signal directly into an electrical signal, they are typically faster than the state-of-art thermal detectors. A high-speed detector not only provides fast data acquisition but also enables imaging and sensing of fast processes and environments, which is a necessity, for example, in vehicular applications.

The state-of-the-art bolometers can reach relatively high sensitivities without cooling. The bolometers are essentially thermometers, which detect the temperature change in the absorber. Widely used resistive bolometers^1,9,10 are based on temperature-dependent resistors. Diode and transistor based bolometers^11–13 are similar to resistive bolometers, but they use semiconductor devices as sensitive thermometers. In thermoelectric bolometers,^1,9,10 often also referred to as thermopiles or thermocouples, the thermoelectric transduction of thermal signals into electrical signals has, in principle, low noise as the main noise sources are the fundamental thermal fluctuation noise and the Johnson–Nyquist noise.^9,10,14 Furthermore, the thermoelectric transduction does not require any external power in signal generation, thus enabling low-power operation. Although here we focus on detectors operating at room temperature, we note that interesting thermoelectric detector concepts operating at cryogenic temperatures have also been proposed.^15,16

Traditional thermoelectrics relies on the bulk material properties, whereas nano-thermoelectrics capitalizes on different combinations of nano-fabrication and synthesis techniques to engineer the electro-thermal performance, i.e., charge carrier and phonon transport.^17–20 Nano-thermoelectrics has been mainly driven by applications in thermal energy harvesting^21–24 with very little attention to detectors until recently when it was postulated that nano-thermoelectrics can provide an attractive low noise transduction method for bolometers.^14,25

Silicon is an attractive detector material as it is widely used in the semiconductor industry, cost-efficient, non-toxic, and, in general, abundantly available. Shaping silicon to nano-scale membranes^26,27 or wires²⁸ causes its thermal conductance to collapse due to increased phonon scattering while keeping the Seebeck coefficient and electrical conductivity virtually unaltered. This increases the thermoelectric figure of merit significantly and, together with the maturity of silicon technology, makes Si-based nanostructures extremely attractive transducer materials for thermoelectric detectors.

For efficient radiation-to-electricity conversion, bolometers require an efficient radiation absorber. To enhance the performance, it is essential that the absorber is supported solely by the thermoelectric legs to reduce the thermal coupling to the surroundings. Various optical absorbers based on thin layers have been proposed and demonstrated for bolometers.²⁹ The thermal mass, or heat capacity, of the absorber together with the thermal conductance to the surroundings defines the speed of the bolometer. Therefore, for example, for imaging applications, high absorption efficiency should be obtained with minimal absorber heat capacity.

In this work, we demonstrate for the first time nano-thermoelectric infrared detector pixel technology. In nano-thermoelectric photodetectors, a high efficiency absorber with low heat capacity is supported solely by the thermoelectric transducer beams (see Fig. 1). The thermoelectric beams are heavily doped poly-Si nanolayers. The thin-film metal grid acts, in addition to being the radiation absorber, as the electric contact between the n- and p-type beams forming the thermocouples. We demonstrate the use of both TiW and TiN as the metal-grid based effective media. These material choices make the detector cost-effective by allowing standard fabrication processes. The technology presented here is easily scalable to various sizes of matrices and single pixel detectors. For our TiW and TiN based devices with ∼25 × 25 µm² absorbers, we obtain very small thermal time constants τ of 66 µs and 190 µs and IR responsivities of 179 V/W and 494 V/W, resulting in specific detectivities of 1.5 × 10⁷ cm Hz^1/2/W and 8.7 × 10⁷ cm Hz^1/2/W. For our larger device with a 40 × 40 µm² absorber, we obtain larger sensitivity, as demonstrated by the respective figures of 2930 V/W and 3.1 × 10⁸ cm Hz^1/2/W, at a speed corresponding to τ = 3.6 ms. These results pave the way for scalable nano-thermoelectrics based IR detection technology for different applications from chemical sensing to thermal imaging.

Thermoelectric bolometers can be described by a model based on the lumped thermal RC circuit.^10,14,25,30 The speed of the bolometer is characterized by the thermal time constant τ = C_th/G_th, where C_th is the heat capacity of the absorber and G_th is the thermal conductance from the absorber to the environment, which is also one of the determining factors of the bolometer sensitivity. G_th is a sum of the conductance stemming from two different heat transfer mechanisms, conduction and radiation, where in the case of the uniform absorber, the latter can be expressed as G_R = 4A_absϵσT³, where A_abs is the area of the absorber, ϵ is the emissivity, σ is the Stefan–Boltzmann constant, and T is the absolute temperature. The thermoelectric transducer of the bolometer transforms the temperature gradient between the absorber and the bath into voltage and is characterized by the total Seebeck coefficient S = dV/dT = S_p − S_n, where S_p and S_n are the Seebeck coefficients of the p- and n-type thermoelectric elements, respectively.

The optical performance of the bolometer is determined by the wavelength-dependent optical (spectral) efficiency $ηλ$ of the absorber. As the optical efficiencies of absorbers generally depend on the wavelength, the total optical efficiency η_tot of a bolometer is specific to the spectrum of the optical power. This optical efficiency can be written in terms of the absorbed optical power P_abs and the total optical power P incident on the absorber as $ηtot=PabsP=∫ηλPλλdλ∫Pλλdλ$⁠, where $Pλλ$ is the spectral incident optical power at wavelength λ. The optical modeling of the absorber is described below.

The absorption of infrared radiation in the present nanobolometers is based on an electrically conducting absorber film (either TiW or TiN), which is patterned into a grid to form an effective medium. This absorber grid is coupled to an optical cavity with a back reflector at the bottom of the cavity (see the description in Sec. III).

Scanning electron microscope images of the fabricated nano-thermoelectric bolometers are shown in Fig. 1 together with diagonal schematic cross sections and a summary of device characteristics. We demonstrate the realization of the same concept using two sets of materials. The devices consist of an optical absorber supported by n- and p-doped 80 nm (device A) or 70 nm (devices B and C) thick poly-Si beams forming two thermocouples and a frame controlling the strain in the beams.³⁷ In device A, this frame is silicon nitride located below the poly-Si, while in devices B and C, the frame is atomic layer deposited Al₂O₃ on the poly-Si. The thermocouple pairs are electrically connected by a 30 nm (device A) or 50 nm (devices B and C) thick layer of metal, which at the same time acts as the absorbing element. In device A, the absorber metal is TiW, and in devices B and C, it is TiN. This metal-poly-Si stack absorber is patterned into a grid to control the optical impedance of the absorber and for easier release of the suspended parts of the device during fabrication. Under the absorber grid, an optical cavity is formed between the absorber and the substrate, which acts as the back reflector. The optical cavity depth of 2.5 µm was selected to maximize the detector output for room-temperature thermal radiation with the wavelength maximum around 10 µm. Optically the absorber, the optical cavity, and the back reflector form a well-known quarter-wave resistive absorber, also known in radio-frequency engineering as the Salisbury screen, where the sheet resistance of the absorber is matched to the vacuum impedance.³⁸ This kind of structure is also known as an antiresonant interference structure.²⁹ 

The devices were fabricated in the VTT Micronova cleanroom facilities on a 150 mm standard single-side-polished p-type (1 Ωcm–50 Ωcm) silicon wafer. First, a 2.5 µm thick layer of sacrificial SiO₂ was deposited by the tetraethyl orthosilicate (TEOS) low pressure chemical vapor deposition (LPCVD) process. Next, in the case of device A, a 270-nm-thick stress-compensation SiN_x layer was deposited by LPCVD. This SiN_x layer was patterned by plasma etching to form a strain-compensation frame similar to that in our previous work.^14,25,37 Then, in all devices, 80 nm (device A) or 70 nm (devices B and C) of LPCVD polysilicon was deposited. The poly-Si was doped selectively with boron and phosphorus by ion implantation and annealing at 700 °C for 2 h. Resistivities of 5.3 mΩcm and 3.4 mΩcm, charge-carrier mobilities of 27 cm²/Vs and 15 cm²/Vs, and Hall carrier concentrations of 4.4 × 10¹⁹ cm⁻³ and 1.3 × 10²⁰ cm⁻³ were obtained for the n- and p-type 80 nm thick poly-Si, respectively, using van der Pauw structures. For the 70 nm thick poly-Si, the corresponding values are 4.7 mΩcm and 3.0 mΩcm, 24 cm²/Vs and 14 cm²/Vs, and 5.5 × 10¹⁹ cm⁻³ and 1.4 × 10²⁰ cm⁻³. These results are in line with the data in the literature for poly-Si films.^39,40 The poly-Si layer was patterned by plasma etching to form the thermoelectric beams. After the poly-Si processing, in the case of devices B and C, 40 nm of Al₂O₃ was deposited by the atomic layer deposition (ALD) technique. After that, all the devices followed the same process flow. The absorber and contact metal, 30 nm of TiW (device A) or 50 nm of TiN (devices B and C), and 300 nm Al pad metal were sputtered and patterned by plasma and wet etching steps. Finally, the devices were released with HF vapor etching of the sacrificial oxide layer.

The optical characterization of the fabricated detector was performed in a vacuum chamber with pressure <1 Pa by illuminating the sample using a calibrated cavity blackbody infrared source (model SR-200 33 of CI-Systems, emissivity 0.99 ± 0.01, and temperature accuracy ±2 K) through a ZnSe window of the vacuum chamber lid. The detector was kept at room temperature with a temperature controlled sample holder. The total optical power is controlled by the area of the variable circular output aperture and the temperature of the blackbody IR source. The temperature of the blackbody IR source determines the spectrum of the IR radiation incident on the detector (see Fig. 2 of the supplementary material). In the setup, the spectrum obtained with the blackbody temperature of 200 °C suits rather well for the present detectors as majority of the IR spectrum is within the optimal absorption range of the detectors. Higher blackbody temperatures provide stronger IR signals (and thereby better accuracy) at all wavelengths but shift the radiation peak to lower wavelengths.

To ensure that the measured signal is not caused by photovoltaic effects in silicon, a high-resistivity (>5 kΩcm) double-side-polished silicon wafer was used as an optical filter directly in front of the ZnSe window. As Si is transparent above the wavelength of ∼1.1 µm, the Si filter has only a minor effect on the shape of the mid- and long-wave infrared spectrum used in the experiments. The combination of the ZnSe window and the high resistivity Si filter results in a relatively flat transmission in the wavelength range of 1 µm–20 µm (see the supplementary material). The total IR power incident on the detector absorber through the optical system is estimated using a numerically calculated blackbody spectrum, and the analytical model of the optical setup with a Lambertian blackbody source (cf. Ref. 41) is described in the supplementary material.

The opto-thermoelectric response of the detector was measured with a lock-in technique using either an optical chopper or a shutter, a SR865A lock-in amplifier, and a SR560 preamplifier connected to the detector in a differential mode. In the measurement, the two thermocouples of the bolometer were connected in parallel, i.e., the n+ ends of the two thermocouples were connected together as well as the p+ ends (see Fig. 1). The optical shutter was used at very low frequencies (5 Hz and below). In high-frequency response measurements (1 kHz and above), an optical chopper wheel with smaller slits, and thus, a small output aperture for the infrared radiation is needed. Therefore, the optical power was maximized by using a high blackbody temperature.

The measured frequency responses of the IR detectors of Fig. 1 are shown in Fig. 2. The model of Eqs. (1) and (2) fits well both to the experimental amplitude and phase data. We can observe that device A is extremely fast: The model fit yields the detector thermal time constant of 66 µs, which is exceptionally small for a thermal detector. The corresponding thermal cut-off frequency ω_c/2π is 2.4 kHz, suggesting that the detector is capable of detecting optical signals with frequencies up to several kHz depending on the power of the measured signal. Device B is also fast with the thermal time constant of 190 µs (see Table I). As the device performance is always a compromise between the speed and responsivity, the responsivities of devices B and C are much larger than those of device A. For example, device C is the slowest and has the highest responsivity due to its longest beams. The largest absorber size of device C reduces its speed as well.

Figure 3(a) shows the response of the detectors to blackbody infrared signals with power levels extending over two orders of magnitude. The responsivity is measured at low frequencies well below the thermal cutoff. The response data show good linearity over the whole power range, thereby demonstrating excellent dynamic range of the detectors. The characteristics of the present bolometers are summarized in Table I. Compared to device A, the higher sensitivities of devices B and C are clearly shown in Fig. 3(a). The spectral efficiencies of the absorbers η(λ), simulated using circularly polarized IR radiation, are shown in Fig. 3(b). The response of these kinds of absorbers is independent of the polarization (see the supplementary material). As all the detectors in Fig. 3(b) exhibit relatively similar spectral efficiencies, the larger responsivities of devices B and C can be explained by their thinner poly-Si, the lower thermal conductivity of which generates a higher thermal signal. In Fig. 3(b), the spectral efficiencies of the absorbers η(λ) peak around the wavelength of 10 µm, thus making the present bolometers ideal for thermal signals from bodies around room temperature. The peak of the blackbody radiation shifts to lower wavelengths at higher blackbody temperatures (e.g., 6.1 µm at 200 °C and 2.7 µm at 800 °C), which results in a slight reduction of the responsivity of the detectors. However, the intensity of the blackbody radiation increases at all wavelengths, and thereby, the total detected signal grows as a function of the temperature of the blackbody regardless of the optimal wavelength range of the detector. Thus, the blackbody temperatures selected for the measurements give a good basis for the performance characterization covering a broad range of wavelengths (see Fig. 2 of the supplementary material) and providing a good accuracy with a relatively simple measurement setup.

The finite element method (FEM) simulations of the poly-Si-metal absorber grid showed that both the TiW and TiN layers govern the optical absorption process in the bolometers, and the poly-Si layer below the absorber metal layers is mostly inactive because its electrical conductivity is too low at IR frequencies. The optical impedance matching of the absorber is determined by the sheet resistance of the absorber metal layer. In all the devices, the effective surface impedances of the absorber are below the optimal 377 Ω (∼287 Ω in device A and ∼132 Ω in devices B and C). The optimal impedance matching can be achieved by tuning the geometry of the absorber grid or the thickness of the absorber metal layer.⁴² Another way to improve the absorptance is to increase the reflectance of the back reflector from the Si substrate value of about 30%, for example, by introducing a metal layer on top of the substrate in the cavity in the beginning of the fabrication process or using a heavily doped substrate in the fabrication. These straightforward improvements would allow the absorptivity to be increased virtually to 100% for a selected wavelength range. For example, in the case of the TiN absorber (employed by devices B and C), high absorption efficiency can be obtained by thinning the TiN layer to 24 nm and introducing a perfect mirror as a back reflector, as shown in Fig. 3(b).

Equation (3) allows the estimation of the material parameters of the thermoelectric transducers of the detectors. The literature data on LPCVD polycrystalline silicon with similar resistivities as in this work^39,43 suggest that the Seebeck coefficients of the n- and p-type poly-Si are −0.25 mV/K …−0.30 mV/K and 0.2 mV/K–0.25 mV/K, respectively. Based on these values, the total Seebeck coefficient S of the detectors is 0.45 mV/K–0.55 mV/K, which is consistent with our previous results.²⁵ For the estimations of the 80 nm poly-Si, we can use the data of device A. Using the 200 °C value of R_V = 179 V/W (see Table I) and the simulated total optical efficiency of the absorber in the characterization conditions, η_tot = 0.48, Eq. (3) gives S/G_th = 373 V/W. Furthermore, with the estimated S, we obtain 1.2 µW/K–1.5 µW/K for the thermal conductance G_th of device A. For devices B and C with 70 nm poly-Si and η_tot of 0.46, we obtain G_th values of 0.41 µW/K–0.51 µW/K and 0.07 µW/K–0.09 µW/K, respectively. The thermal conductances G_th of devices B and C are much smaller than that of device A due to the lower thermal conductivity of the thinner poly-Si. In the case of device C, G_th is further reduced due to the longer thermoelectric beams. The thin-film thermal conductivities of the absorber metals (in the metal/poly-Si absorber stack) range typically from 60 W/(mK) of TiW⁴⁴ to 1.2 W/(mK)–4.7 W/(mK) of TiN.^45,46 Here, it should be noted that although there is a thermal boundary between the poly-Si and the absorber metal, the typical Si-metal thermal boundary resistances^45,47,48 are insignificantly small.

By assuming that thermal conductivities of the n- and p-type poly-Si beams are equal in devices A and B, we estimate that the thermal conductivities of 70 nm and 80 nm poly-Si are 7 W/(mK)–8 W/(mK) and 17 W/(mK)–21 W/(mK), respectively. These values are around an order of magnitude smaller than in bulk silicon. A low thermal conductivity is a prerequisite for turning silicon into a material with a high thermoelectric figure of merit. The thermal conductivity estimates are in line with the thermal conductivities of undoped and doped single and polycrystalline silicon membranes with similar thickness ranges.^{14,25–27,40} Thicker poly-Si films with higher doping levels than in the present poly-Si membranes and bulk poly-Si exhibit thermal conductivities in the range of 30 W/(mK)–120 W/(mK).^43,49–51 The thermal conductivity of the present poly-Si membranes suggests that phonon surface scattering reduces the thermal conductivity similarly as in single-crystalline Si nanomembranes.^26,27

The electrical resistance R of the thermoelectric bolometer is crucial to the sensitivity of the device as it determines the intrinsic voltage noise of the detector.¹⁴ The resistance of a real device includes additional series resistance that is not directly originated from the thermoelectric transducer. To clarify this, we describe the total measured resistance of the bolometer as R_tot. In the present bolometers, the resistance of the absorber grid has ideally only little contribution to the total resistance of the devices as the effective sheet resistance of the absorber grid is close to the vacuum impedance (∼377 Ω). In the present type of devices, the additional series resistance originates, for example, from the resistance of the polysilicon–metal contact in the absorber grid. In device A, this poly-Si-TiW contact resistance dominates: the total resistance of the thermoelectric transducer is R_tot = 50 kΩ, and only a small portion of it, namely, R_beams = 2.5 kΩ, arises from the thermoelectric poly-Si beams. In device B, the effect of the contact resistance is much smaller, and in device C, the beam resistance R_beams is dominating R_tot. The effect of the R values on the device performance can be clearly seen in $ZT̃$⁠. For device A, the values of R listed in Table I and the value of G_th estimated above give $ZT̃$ = 1.0–1.2 × 10⁻³ using the measured total bolometer resistance R_tot and $ZT̃$ = 0.020–0.025 using the resistance of the thermoelectric beams, R_beams, only. For devices B and C, the corresponding former values are an order of magnitude larger, $ZT̃$ = 0.010–0.015 and the latter $ZT̃$ = 0.015–0.067.

Because the estimated $ZT̃$ is well below 1/3, the intrinsic voltage noise of the present bolometers is determined by the Johnson–Nyquist noise; see Eq. (6), Ref. 14, and experimental verification in the supplementary material. This allows us to calculate the optical NEPs and specific detectivities D^* of the detectors using Eq. (5) by applying the total bolometer resistance R_tot. In the ideal case, the thermoelectric beams alone (R_beams) determine the total resistance R_tot of the bolometer, which would decrease the NEP and increase D^* from the measured values. As discussed above, the total optical efficiency η_tot of the bolometer can be increased to 100% by improving the absorber efficiency. These calculated and estimated parameters are listed in Table I. The resulting values are competitive with the state-of-the-art IR detectors (see a detailed comparison below). Furthermore, $ZT̃$ and, thereby, NEP and D^* of the present bolometers can be improved by selecting optimal geometries for the thermoelectric beams based on the resistivities, Seebeck coefficients, and thermal conductivities of the n- and p-type thermoelectric materials (see Ref. 14 and further the discussion on $ZT̃$ below). The performance can be improved even further by geometrical optimization of $ZT̃$⁠.

Since the surface scattering of phonons increases with decreasing nanomembrane thickness, leading to a decrease in thermal conductivity,²⁷ the performance of the thermoelectric transducer of the detector can be improved by thinning the poly-Si membrane further from the demonstrated thicknesses of 70 nm and 80 nm. Even further performance enhancement can be achieved by maximizing the power factor S²/R by optimizing the doping parameters. However, since S decreases with increasing electrical conductivity,^20,52 the optimization is more complicated than just minimizing R by increasing the doping. This dependence of the power factor and ZT on the doping density of single-crystalline silicon has been rather recently studied in experimental nanoscale thermoelectric generators.⁵³ 

The effect of a thinner poly-Si membrane can be estimated using the thermal conductivity of 8 W/(mK) measured for a 9 nm thick single-crystalline silicon layer with native oxide.²⁷ By optimizing the absorber, we can reach unity η (λ), as discussed above. If the grid geometries were fixed, the optimal thicknesses of the absorber metals would be 19 nm for device A (with TiW) and 18 nm for devices B and C (with TiN) according to the electromagnetic simulations. This and the use of thinner poly-Si lead to lower heat capacity and higher speed for the devices (depending on the geometry). Furthermore, we see that the detectors can be optimized so that their resistance is determined by the resistance of the thermoelectric beams R_beams only. This and the optimization of the thermoelectric material would double power factor S²/R, leading to $ZT̃$ = 0.11 for devices A–C, which leads to D^* values well above 10⁹ cmHz^1/2/W for all the devices. The details of these estimations are described in the supplementary material.

The specific detectivity vs time constant performance of our nano-thermoelectric bolometers is plotted in Fig. 4 together with LWIR detector results reported in the literature. Note that in order to give a comprehensive view of the whole LWIR detector field in addition to thermal detectors, we have also included cooled and uncooled quantum detectors in Fig. 4. Device A is an order of magnitude faster than the fastest state-of-the-art thermoelectric bolometer and three times faster than the fastest state-of-the art resistive bolometers. The sensitivity of the present device is in the same order of magnitude as some thermal detectors and most uncooled quantum detectors. The fill factors of devices A and B, η_FF of 66%–67%, are large compared to the fill factors of many existing thermoelectric bolometers.^54,55 As an example, the detector in Ref. 56 has a fill factor of 1%. Detectors with a large fill factor are attractive in imaging applications where tightly packed detector arrays are needed. As discussed above, we expect that the performance of our present device can be improved remarkably by straightforward material and geometrical optimization. To highlight this, Fig. 4 also shows the performance values of the resistance-optimized nano-thermoelectric bolometers with the total resistance R determined by the beams only and unity absorptance (see Table I) as well as the proposed optimized detectors based on 9 nm poly-Si designed either for speed or sensitivity, as described in the supplementary material. These estimated cases cover thermal time constants over three orders of magnitude. This demonstrates the versatility of this technology and its feasibility to many applications. Note that as a rough rule of thumb for all bolometers, the geography of detectivity vs time-constant plot (Fig. 4) can be interpreted so that the points with smaller (larger) time-constant and detectivity are for imaging array (single/few pixel sensing) applications.

The detector sensitivity can be enhanced further by utilizing high ZT materials,^57–59 which allow NEP to be pushed to the thermal fluctuation limit defined by Eq. (4). It should be noted that the dashed horizontal line in Fig. 4 denoting the 300 K background radiation limit is also a photon thermal fluctuation noise limit, which follows from Eq. (4) in the case where only photons exchange heat between the absorber and the environment. The specific detectivity values of the largest pixels (Fig. 4, estimated points) approach this limit, suggesting that nano-thermoelectric IR detectors could operate at the thermal background limit. The time constants for these detectors are relatively large excluding high speed imaging applications but allowing for single or few pixel chemical sensing applications, for example.

In conclusion, by combining nano-thermoelectrics and nanomembrane photonics, we have demonstrated LWIR bolometers based on CMOS-compatible materials and shown that this technology can be utilized in a wide range of applications by tailoring the speed and sensitivity according to the specific needs. Nano-thermoelectric IR detectors can provide high speed and high sensitivity, low-power operation, and cost-effectiveness within single technology. For the fabricated devices of different sizes, the measured responsivities are in the range of 180 V/W–2900 V/W and specific detectivities are in the range of 1.5 × 10⁷ cm Hz^1/2/W–3.1 × 10⁸ cm Hz^1/2/W. The thermal time constant of the fastest device is as low as 66 µs, and even the device with a larger absorber and highest specific detectivity of 3.1 × 10⁸ cm Hz^1/2/W exhibits a time constant (3.6 ms) comparable with the existing state-of-the-art bolometers. The fast operation speed stems from the low heat capacity multi-functional absorber, where a single metal layer grid absorbs the radiation and, at the same time, connects the poly-silicon n- and p-type nano-thermoelectric support beams, which convert the IR radiation induced temperature rise into voltage. In addition, we have discussed how the specific detectivity of the nano-thermoelectric bolometers can be pushed above 1 × 10⁹ cm Hz^1/2/W without compromising the speed of the detector and benchmarked the performance with the state-of-the art LWIR detectors.

See the supplementary material for supporting details on the IR characterization, the optical modeling of absorbers, the detector noise and performance estimation, and the LWIR detectors of the comparison.

This work was financially supported by the European Union Future and Emerging Technologies (FET) Open under Horizon 2020 programme (Grant Agreement No. 766853, project EFINED), the Business Finland co-innovation project RaPtor (Grant No. 6030/31/2018), and the Academy of Finland projects EXACT, BOLOSE, and LAMARS (Grant Nos. 295329, 314447, and 314809, respectively). The work of J.T. was personally supported by the Academy of Finland through Grant No. 324838. This work is part of the Academy of Finland Flagship Programme, Photonics Research and Innovation (PREIN), decision 320168. We acknowledge gratefully the technical assistance of Teija Häkkinen, Kai Viherkanto, Paula Holmlund, and Tahvo Havia in device fabrication and characterization.