The mid-infrared (MIR) wavelength region is attractive for many applications where chemical sensing attracts the most attention, because a number of molecules have unique absorption bands in the “molecular fingerprint” regions (2.5–20 µm) [1–4]. Low loss MIR waveguides based on silicon (Si) and germanium (Ge) have been developed [5–8]. On the other hand, amorphous silicon (a-Si), which has advantages of being low-cost and CMOS compatible has been investigated in the last several years. It also has potential to be applied in MIR [9,10].

Several integrated MIR photon detectors have been demonstrated [11–14]. Apart from the photon detectors, the thermal detectors are in another important class of detectors in the MIR field. Compared to the photon detector, one major advantage of the thermal detector is the wavelength independence when an absorption mechanism is used that is wavelength independent [15]. In the last year, we have developed antenna-assisted integrated waveguide-based bolometers on the SOI platform [16,17]. However, in first-generation bolometers, the high thermal conductivity of the surrounding crystalline silicon (c-Si) resulted in a high heat leak, which strongly limited the responsivity of the bolometer to values at around 1%/mW. Compared with c-Si, a-Si has an advantage of low thermal conductivity. Therefore, it can better confine the heat in a localized region and achieve a higher temperature change. In the current work, we developed the bolometer on the a-Si platform to significantly improve its performance.

The waveguide-based bolometer is based on the a-Si platform. The geometry of the bolometer is shown in Fig. 1. a-Si is a common material for bolometers. It has a high temperature coefficient of resistance (TCR) of 2.5%/K [18] and also a low thermal conductivity of ${2.2}\;{\rm{W/}}({\rm{m}} \cdot {\rm{K}})$ [19]; hence, it can confine generated heat in a localized region. However, the resistivity of intrinsic a-Si is extremely high ($5 \times {{1}}{{{0}}^{10}}\,\Omega {\rm{cm}}$ [20]); hence, a change in resistance is difficult to measure. Therefore, a heavily doped $p$-type a-Si layer is introduced on top of the waveguide in order to reduce the resistance. On the other hand, the doped a-Si would also introduce more absorption loss. To balance the resistivity and the absorption loss, the a-Si platform includes a 400 nm thick intrinsic a-Si layer and a 100 nm thick $p$-type a-Si layer on the top. The heavily doped a-Si layer could be etched after deposition so that it is present only in the bolometer region, but in this demonstration, it is left covering all of the waveguides, to reduce the fabrication complexity.

 

Fig. 1. Diagram of the structure of the designed bolometer based on the a-Si platform.

Download Full Size | PDF

The a-Si bolometer is based on a 1.3 µm wide waveguide core with sub-wavelength grating (SWG) cladding, which is designed for TE only [21]. This kind of waveguide has the potential to be suspended to realize a stronger thermal insulation. In the bolometer, three nanometallic optical rod antennas are fabricated on the surface of the waveguide as the absorber [16]. The dimensions of each antenna are 540 nm length, 160 nm width, and 20 nm thickness, as obtained using e-beam lithography and lift-off. We use gold as the material of choice for its excellent plasmonic properties and chemical inertness. These antennas absorb the incident light and heat up the localized area including the thermometer arm, which is positioned close to these antennas (Fig. 1). The thermometer is a strip that is 1 µm wide and 6.3 µm long, and it is aligned perpendicular to the waveguide between two contact pads (see Fig. 1). In theory, the temperature change in the bolometer is described as [15]

To predict the thermal profile in the bolometer, a simulation model was built using the COMSOL Multiphysics software. Although the absorption rates in these three antennas are slightly different and dependent on the wavelength, in our thermal simulation, the three antennas are considered as a uniform heat source, which provides an accurate description given the proximity of the three sources compared to the thermal diffusion length as discussed below. The background environment temperature is set as 293.15 K, and the convective heat transfer coefficient of air is set as $5\;{\rm{W}}/({{\rm{m}}^2} \cdot {\rm{K}})$ [22]. The properties of the materials are shown in Section B of Supplement 1.

For the heating power of 1 mW in the antennas, the thermal simulation produces the results shown in Fig. 2. From Fig. 2(a), the temperature in the localized region around the antennas is over 600 K (350°C). The temperature decays quickly away from the antennas. In the cross section of the thermometer [Fig. 2(b)], the temperature of the waveguide core is 350 K, and the temperature at the far ends of the bolometer (close to the contacts) is 300 K. Although the antenna region is the heating center with extremely high temperature, there is a 1.5 µm offset between the antennas and the thermometer, which was introduced in order to reduce the light scattering from the thermometers. One advantage of a-Si is the low thermal conductivity. The heat generated in the antennas is confined in a tiny micrometer region. To compare the materials of c-Si and a-Si, a bolometer with a similar structure but based on the c-Si platform with an a-Si thermometer has been simulated. In Fig. 2(c), the temperature in the region immediately around the antennas is only 311 K. Compared with the a-Si waveguide, the c-Si waveguide conducts the heat to the surrounding quickly. In the cross section of the thermometer [Fig. 2(d)], the temperature of the waveguide core is 301 K. It is clear that with the same heating power in the antennas, the c-Si waveguide bolometer has a lower temperature change in the thermometer.

 

Fig. 2. Thermal profile of (a) the whole a-Si bolometer and (b) the thermometer cross section in the simulation. (The temperature of the white region at the outside of the bolometer is lower than 311 K, which is out of the color scale.) Thermal profile of (c) the c-Si bolometer with the a-Si thermometer and (d) the cross section of the thermometer in the simulation. The dashed rectangles are a-Si thermometers.

Download Full Size | PDF

Figure 3 plots the simulated temperature response to varying heating power, and the frequency response of the bolometer in the simulation. It is clear that the temperature in the antennas is much higher than the temperature in the thermometer due to the low thermal conductivity of a-Si. On the other hand, the antennas are not only absorbing but also reflecting and scattering the incident light. In the simulation, the absorption of three antennas is 70% [16]. However, due to fabrication errors in the antennas, the absorption coefficient is usually less than the simulation result. Assuming that the absorbing rate of antennas is 50%, the responsivity of the bolometer is 13.97 K/mW. Under the stationary simulation, the frequency-dependent term, ${\omega ^2}C_{\rm{th}}^2$, can be ignored. Then, according to the calculation, the thermal conductance, ${G_{\rm{th}}}$, is $3.58 \times {{1}}{{{0}}^{- 5}}\;{\rm{W/K}}$. From Fig. 3(b), the 3 dB bandwidth is 129.57 kHz; therefore, the thermal capacitance is $7.62 \times {{1}}{{{0}}^{- 11}}\;{\rm{J/K}}$.

 

Fig. 3. (a) Average temperature in the antennas and thermometers versus the heating power in the antennas in the simulation. (b) Normalized temperature change with respect to the modulating frequency of the heating power in the simulation.

Download Full Size | PDF

The fabrication begins with preparing a bulk silicon sample. To get the designed a-Si on insulator sample, 3 µm ${\rm{Si}}{{\rm{O}}_2}$ was deposited by plasma enhanced chemical vapor deposition (PECVD) firstly. After that, 400 nm intrinsic a-Si and 100 nm $p$-type a-Si were deposited by hot wire chemical vapor deposition (HWCVD) on the sample sequentially. In the $p$-type a-Si deposition, the gas ratio between ${{\rm{B}}_2}{{\rm{H}}_6}$ and ${\rm{Si}}{{\rm{H}}_4}$ is 10:40 sccm. Then, the passive devices, waveguides and gratings, were patterned on the sample. After that, 200 nm thick Au contact pads were deposited on the sample, with a 10 nm thin Cr layer between the Si and Au used as an adhesion layer. Finally, the optical antennas (composed of 1 nm thick Cr as the adhesion layer and 20 nm Au) were deposited. The sketch of the fabrication processes flow is presented in Section 1 of Supplement 1.

The experimental setup is similar to that described in [16], where the DC source was upgraded to Keithley 6487. First, the waveguide propagation loss is measured by the cut-back method, and the loss is 94 dB/cm. In this demonstration, this high loss is avoided by using a short input waveguide length (${\lt}{{1}}\;{\rm{mm}}$). The high loss does not significantly reduce the performance of the bolometer itself because of the very short length of the bolometer. In the future, the a-Si bolometer could be integrated with a c-Si waveguide-based sensing circuit on the SOI platform. Then, the resistance of the bolometer was measured by heating up the whole sample to different temperatures using a Peltier stage. Figure 4(a) presents the IV curve of the bolometer at 30°C. It illustrates that there is a Schottky contact between the contact pad and the $p$-type a-Si. By fitting the linear region of the IV curve, the resistance is found to be $5.69 \times {10^8}\,\Omega$. Figure 4(b) plots a normalized resistance with respect to the temperature. Assuming the sheet resistance of $p$-type a-Si is dominant, the TCR of the $p$-type a-Si is ${-}{1.9}\% {\rm{/K}}$ from the linear fitting.

 

Fig. 4. (a) IV curve of the bolometer at the temperature of 30°C. (b) Normalized resistance of the bolometer versus the temperature of the sample.

Download Full Size | PDF

The normalized transmission spectrum of the antennas is shown in Fig. 5(a). The transmission of the antennas is less than ${-}{{10}}\;{\rm{dB}}$ over the 3.72–3.88 µm wavelength range. At the wavelength 3820 nm, the transmission is even less than ${-}{{30}}\;{\rm{dB}}$. This result illustrates that the antennas have a broadband response, with a 3 dB bandwidth greater than 160 nm. In the response measurement, we measured the current change with a constant voltage instead of measuring the resistance of the bolometer. To investigate the enhancement from the antennas, the bolometer without the antennas is measured as the reference. Figure 5(b) plots the response versus the input power of the bolometers with and without the antennas at the wavelength of 3.8 µm and the chopping frequency of 400 Hz. From the linear fitting, the responsivity of the bolometer is 24.62%/mW, and the reference bolometer has a responsivity of only 0.81%/mW. Therefore, the antennas introduce a 14.83 dB enhancement on the responsivity of the bolometer. In addition, the responsivities of bolometer with and without antennas at different wavelengths are shown in Fig. 5(a).

 

Fig. 5. (a) Blue line is the normalized transmission spectrum of the bolometer. The plus sign and square marks are the responsivity of the bolometer with and without antennas, respectively. (b) Response of bolometers with and without the antennas versus the input power at 3.8 µm wavelength and 400 Hz chopping frequency.

Download Full Size | PDF

 

Fig. 6. (a) Responsivity and (b) noise of the bolometer versus the chopping frequency.

Download Full Size | PDF

Figure 6(a) plots the responsivity of the bolometer with the antennas versus the chopping frequency. The responsivity of the bolometer does not show an obvious attenuation up to 1 kHz, which is the maximum frequency limit of the chopper. It indicates that the term of the thermal conductance to the surroundings in Eq. (1), ${{\rm{G}}_{\rm{th}}}$, predominantly limits the temperature change in the bolometer. Assuming the absorption coefficient is 50%, the thermal conductance of the bolometer is $3.86 \times {{1}}{{{0}}^{- 5}}\;{\rm{W/K}}$, which matches the simulation result. On the other hand, a free-space coupled bolometer (forming part of a camera array) with a thermal conductance of $2.38 \times {{1}}{{{0}}^{- 8}}\;{\rm{W/K}}$ has been reported in [23]. Suspension is an effective method for improving the thermal insulation [16]. In the simulation shown in Section 2 of Supplement 1, the thermal conductance after the suspension is reduced to $3.46 \times {{1}}{{{0}}^{- 6}}\;{\rm{W/K}}$. Therefore, there is still a high heat loss through the waveguide. It indicates that still responsivity improvement of 2–3 orders of magnitude may be possible in the future through redesign of the device geometry. Meanwhile, the noise with respect to the chopping frequency is plotted in Fig. 6(b). In the frequency range from 50 Hz to 10 kHz, the noise does not decrease quickly as frequency increases, and the mean value of the noise is only $0.0247\% /\sqrt {{\rm{Hz}}}$. The noise equivalent power (NEP) of the bolometer at 400 Hz is $10.4\;{\unicode{x00B5}{\rm W}}/\sqrt {{\rm{Hz}}}$. (The details of NEP calculation are shown in Section 3 of Supplement 1.)

To further investigate the response of the bolometer with a high input power, a light with a power of 1.2 mW is coupled into the bolometer. At these high input powers, we observe an irreversible reduction in the responsivity from 24.62%/mW down to 7.26%/mW. Figures 7(a) and 7(b) present scanning electron microscopy (SEM) images of the antennas of bolometers that have and have not been exposed to a high input optical power. It is clearly visible that the antennas have been damaged after coupling a high input power up to 1.2 mW. According to the high responsivity, 24.62%/mW, and the TCR, ${-}{1.9}\% {\rm{/K}}$, the temperature change in the thermometer is 15.5°C. According to the simulation [Fig. 3(a)], the average temperature in the antennas is 228°C. The simulation assumes that the heating in the antennas is uniform, but the heating is not actually uniform. The melting point of the Au/Si eutectic alloy is around 363°C [24]. It indicates that the temperature in the localized region of the antennas is higher than 363°C. If the thermometer can go though the antennas region while avoiding high optical scattering from the thermometer arms, the responsivity of the bolometer could be greatly improved.

 

Fig. 7. SEM images of the antennas of bolometers that (a) have and (b) have not been exposed to a high input power.

Download Full Size | PDF

In this work, we demonstrated a waveguide-based bolometer on the a-Si platform. Compared with the bolometer on the SOI platform published in [16], the responsivity of the bolometer is improved from 1.13%/mW to 24.62%/mW due to the low thermal conductivity of the a-Si. From the SEM images and the simulation, the temperature change in the region near the antennas is 10 times higher than the temperature change in the thermometer. If the thermometer can measure the temperature change in the region around the antennas, the performance of the bolometer could be greatly improved. In addition, gold, the material of antennas, could be replaced by Al or TiN, which is available Si foundries. The absorption in TiN antennas could be enhanced by increasing the number of antennas. Furthermore, using multiple antennas with different lengths could further increase the bandwidth of absorption. This bolometer will propel the realization of integrated MIR sensors.

All data supporting this study are openly available from the University of Southampton repository at https://doi.org/10.5258/SOTON/D1587.