1. Introduction

Due to the attractive properties of terahertz waves, such as high transmission, low energy, and fingerprint spectrum, terahertz imaging and spectroscopy as tools are extensively involved in industrial non-destructive testing [1], medical drug inspection [2], biomedical identification [3], food safety [4] and homeland security [5,6]. However, the lack of terahertz detectors that combine broadband, high sensitivity, high speed, cost-effectiveness, and portability is a central obstacle to the mass popularity of the above applications in our daily lives, since the power of existing broadband terahertz sources is still low [7,8]. Over the last two decades, researchers have been pursuing the improvement of the sensitivity and bandwidth of terahertz detectors [9]. Golay cells [10] and pyroelectric detectors [11–14] deliver an extremely broad response spectrum with a bandwidth of up to 43 THz and 300 THz, respectively. However, their sensitivity and response speed are the main drawbacks limiting the implementation of fast imaging and spectroscopy systems. Cooled bolometers have an ultra-high sensitivity with a noise-equivalent power (NEP) of less than $1~\mathrm {pW/\sqrt {Hz}}$ and a bandwidth of up to 20 THz [15–17]. However, operating at 4.2 K severely constrains the bolometers’ prospective applications. As the most sensitive commercial room-temperature detectors, waveguide-coupled and quasi-optical coupled Schottky-barrier diodes (SBDs) can provide NEPs below $10~\mathrm {pW/\sqrt {Hz}}$ from 0.1 to 1 THz [18]. However, the challenge remains in maintaining high sensitivity at frequency above 1 THz due to its parasitic parameters limiting the cut-off frequency. Field-effect transistors (FETs) can achieve a high responsivity in a broad frequency range far above the transistor’s cut-off frequency [19,20]. FET terahertz broadband detectors integrated with broadband spiral antennas and bowtie antennas have achieved detection beyond 1 THz. Boppel et al. reported a 150 nm silicon complementary metal-oxide-semiconductor (CMOS) detector that operates at discrete frequency from 0.2 to 4.3 THz with a NEP of $487~\mathrm {pW/\sqrt {Hz}}$ at 2.9 THz [21]. Ikamas et al. demonstrated a 90 nm CMOS has state-of-the-art performance and exhibits a nearly flat responsivity up to 2 THz with a NEP of $70~\mathrm {pW/\sqrt {Hz}}$ at 1.5 THz [22]. Regensburger et al. achieved broadband terahertz detection up to 11.8 THz based on AlGaAs/GaAs high-electron-mobility transistors (HEMTs), showing a NEP of $250~\mathrm {pW/\sqrt {Hz}}$ at 0.6 THz [23]. Compared to SBDs, even though the bandwidth of FETs has been significantly increased, the sensitivity is not sufficiently high. Therefore, a FET-based detector as a potential solution, high-bandwidth and high-sensitivity device needs to be explored.

In our previous works, we reported a terahertz detection from 0.2 to 2 THz for incoherent blackbody by three separated antenna-coupled AlGaN/GaN HEMT detectors with different center response frequency of 0.34 THz, 0.65 THz and 0.9 THz [24]. Recently, an antenna-coupled AlGaN/GaN HEMT detector with a NEP of $3.7~\mathrm {pW/\sqrt {Hz}}$ at room temperature [25] and a NEP of less than $1~\mathrm {pW/\sqrt {Hz}}$ at 77 K has been achieved [26]. The results imply that combined multiple such narrowband dipole antennas could be further established to expand the frequency bandwidth of received terahertz wave and would maintain high sensitivity. In this article, a broadband terahertz detector built with an array of dipole-antenna-coupled AlGaN/GaN HEMT is designed and fabricated. Eighteen pairs of dipole antennas are arranged in a bow shape pattern, with the different center frequency of each pair from 0.24 to 7.4 THz. Eighteen transistors have common a source and a drain but different gated channels coupled by the corresponding antennas. The photocurrents generated by each gated channel are combined in the drain as the output port. The detector is characterized by the radiation of both a blackbody and coherent source at 298 K and 77 K.

2. Device information and experimental implementations

The broadband detector is fabricated on AlGaN/GaN two-dimensional electron gas (2DEG) with a silicon carbide substrate. The 2DEG offers an electron density of ${n}_0 = 7.89 \times 10^{12}~\mathrm {cm}^{-2}$ and an electron mobility of $\mu = 2290~\mathrm {cm^2/Vs}$ at 298 K, ${n}_0 = 8.35 \times 10^{12}~\mathrm {cm}^{-2}$ and $\mu = 1.89 \times 10^{4}~\mathrm {cm^2/Vs}$ at 77 K, respectively. As shown in Fig. 1(a), eighteen pairs of dipole antennas are arrayed into a bow-tie pattern which is symmetrical about the $x$ axis. Nine pairs of dipole antennas form one side and nine pairs form another side. Each pair of dipole antenna is composed of two quarter-wavelength antennas: one of them (g-antenna) is connected to the gate and another (i-antenna) is placed on the side of the gate, as schematically shown in Fig. 1(b). The 2DEG region under the gate is the gated channel corresponding to each pair of dipole antennas, as indicated in the red dashed in Fig. 1(b). The incident terahertz wave creates a confined and asymmetrical electric field. The component in the source-drain direction (along $x$) modulates the electron velocity, while the other component in the direction perpendicular to the 2DEG plane (along $z$) modulates the electron density. Each antenna pair couples the incident terahertz wave into the corresponding gated channel and generates the self-mixing photocurrent. The detector has eighteen transistors with common a source and a drain so that the detector outputs the self-mixing photocurrent in parallel. The center response frequency ${f}_0$ of each antenna pair is designed from 0.24 to 7.4 THz according to ${f}_0 \sim c /(4\sqrt {\bar {\epsilon }} L_{\mathrm {A})}$, where $L_{\mathrm {A}}$ is the length of antenna block, ${c}$ is the speed of light in vacuum, $\bar {\epsilon } \approx (1+\epsilon ) / 2, \epsilon \approx 9.8$ is the relative permittivity of GaN on the substrate, respectively. The length of antenna block $L_{\mathrm {A}}$, the width of antenna block $W_{\mathrm {A}}$ and the distance between antenna pairs $D_{\mathrm {A}}$ decrease as the resonant frequency increases, as listed in Table 1. Now the location of the antenna is defined in rectangular coordinates, using the center of the chip as the origin of the coordinate (0,0), as shown in Fig. 1(a). We have marked the center of the i-antenna $y_{\mathrm {in}}, y_{\mathrm {in}^{\prime }}$ and the center of the g-antenna $y_{\mathrm {gn}}, y_{\mathrm {gn}^{\prime }}$ on the y-axis, as shown in Fig. 1(b), where $n = 1,2,3,4,5,6,7,8,9$. The y-axis coordinates of the antennas are listed in Table 1. All g-antennas are directly connected with a wide gate which is shared by the eighteen transistors with have the same gate length $L$ of 600 nm and different channel width ${W}$. The distance between the i-antenna and gate is $D$ = 600 nm for each dipole antenna. Table 1 lists detailed information on antennas and channels.

 

Fig. 1. (a) Broadband terahertz detector based on an array of antenna-coupled AlGaN/GaN HEMTs. (b) Zoom-in view of the right part of the central active region. (c) Simplified equivalent circuit of the detector. (d) FTS with the broadband detector.

Download Full Size | PDF

Table 1. Parameters of the antennas and channels.

View Table

According to the self-mixing model of antenna-coupled HEMTs without source-drain bias [20], the self-mixing photocurrent generated in each transistor can be expressed as

The gated channel resistance and the source-drain series resistances of each transistor can be estimated approximately based on the electron density, the electron mobility of the 2DEG and the geometry of the active region, but it is challenging to measure accurately the resistances of each transistor in practice. It is relatively easy to obtain the whole channel resistance ($r_\mathrm {m}$) and series resistances ($r_\mathrm {s}$ and $r_\mathrm {d}$) of the detector. The source-drain resistance $r$ of the detector can be expressed as $r \approx r_\mathrm {m}+r_\mathrm {d}+r_\mathrm {s}$. Hence, the extracted photocurrent $i_\mathrm {THz}$ becomes a fraction of the internal photocurrent $i$ from the gated channel: $i=\gamma i_0$ with factor $\gamma =1-\left (r_{\mathrm {d}}+r_{\mathrm {s}}\right ) / r$, which can be increased from 0.1 to near 1 at 298 K and from 0.13 to near 1 at 77 K when the gate voltage is gradually reduced from 0 V to −4 V. The detector chip is carefully assembled in the center of the planar surface of a high-resistivity silicon hyper hemispherical lens with a diameter of 6 mm and a height of 3.87 mm. Placing the detector module in a cryogenic dewar at 77 K permits a significant increase in electron mobility and thus in the sensitivity of the detector [24,26]. To characterize the response spectrum of the detector to a blackbody, a Fourier-transform spectrometer (FTS) is employed, as shown in Fig. 1(d). A chopper is placed in the optical path to modulate the terahertz source. Boosted by a preamplifier (DL1211), the photocurrent $i_{\mathrm {THz}}$ is measured by a lock-in amplifier (Signal Recovery 7265).

3. Results and discussions

Firstly, we characterize the detector’s terahertz photoelectric properties at 298 K and 77 K, as well as the electrical properties of the field-effect channel. As shown in Fig. 2(a), the detector gives a zero-$V_\mathrm {g}$ conductance of about $G=3.55~\mathrm {mS}$ at 77 K, a factor of 4.6 increase over that at 298 K. As shown in Fig. 1(c), the drain-source resistance $r \approx r_\mathrm {m}+r_\mathrm {d}+r_\mathrm {s}$ includes the gated 2DEG channel resistance $r_\mathrm {m}$ and series resistances $r_\mathrm {d} \approx r_\mathrm {s}$. The total series resistances $r_\mathrm {d}+r_\mathrm {s}$ are fitting parameter which is about $1160~\Omega$ at 298 K and $246~\Omega$ at 77 K. Hence, the pure gate-controlled channel conductance $G_\mathrm {m}$ is extracted by $G_\mathrm {m}=G /\left [1-\left (r_\mathrm {d}+r_{\mathrm {s}}\right ) G\right ]$ shown as the solid curves in Fig. 2(a). The photocurrent tuned by the gate voltage $V_\mathrm {g}$ is measured under coherent terahertz irradiation with an illumination power of 2.7 $\mathrm {\mu }$W at 931 GHz, as shown in Fig. 2(b). The maximum photocurrent is increased by a factor of 16.3 from 27.5 nA at 298 K to 450 nA at 77 K. The solid curves in Fig. 2(b) are fitting based on the self-mixing model and agree well with the experiment data. The optical NEP tuned by the gate voltage $V_\mathrm {g}$ is calibrated and shown in Fig. 2(c). The minimum optical NEP at 298 K is as low as $282~\mathrm {pW/\sqrt {Hz}}$ at $V_\mathrm {g}$=−3.64 V and at 77 K is $15~\mathrm {pW/\sqrt {Hz}}$ at $V_\mathrm {g}$=−3.38 V, respectively. The optical NEP is improved by a factor of 18.8 as the temperature decreased from 298 K to 77 K. The value of the factor $\gamma$ is 0.60 at $V_\mathrm {g}$=−3.64 V and 298 K and is improved by 1.2 times to 0.70 at $V_\mathrm {g}$=−3.38 V and 77 K, respectively. As shown in Fig. 2(d), incoherent terahertz radiation from a blackbody of 1173 K is detected by the detector at 77 K and 298 K. The solid curve describing the field-effect factor matches the experimental data quite well. The maximum photocurrent is increased by a factor of 57 from 2.6 nA at 298 K to 0.45 nA at 77 K. Compared to the 931 GHz coherent source, the cooling effectively improves the response of the detector to an incoherent blackbody source. This is mainly due to the significantly wider response bandwidth of the detector as a result of the cooling as shown in Fig. 4(c).

 

Fig. 2. (a) Total conductances ${G}$ of the detector at 298 K and 77 K tuned by the gate voltage. Gate-controlled conductances ${G}_\mathrm {m}$ excluding the series resistances are shown as solid curves. (b) Terahertz photocurrent at 298 K and 77 K as a function of the gate voltage under continuous-wave coherent irradiation at 931 GHz. The photocurrent from the detector at 298 K is multiplied by a factor of 16.3. (c) NEP at 298 K and 77 K as a function of the gate voltage at 931 GHz. (d) Terahertz photocurrent tuned by the gate voltage under incoherent broadband radiation of a 1173 K blackbody. The photocurrent from the detector at 298 K is multiplied by a factor of 57. The solid curves in (b) and (d) are identical simulations based on the self-mixing model.

Download Full Size | PDF

The coherent response spectrum of the detector could be examined with coherent continuous-waves terahertz irradiating at varied frequencies. We characterize the sensitivity of the detector by obtaining its optical responsivity $R_\mathrm {i}$ and NEP. The calculation formula for the optical responsivity $R_\mathrm {i}$ can be expressed as $R_{\mathrm {i}}=i_{\mathrm {THz}} / P_0$, where $P_0$ is the terahertz radiated power determined by a calibrated Golay cell (TYDEX GC-1P, 102900 V/W) and is the total power coupled into the detector by the silicon lens. The calculation formula for the NEP can be expressed as $\mathrm {NEP}=N_{\mathrm {i}} / R_{\mathrm {i}}$, where $N_{\mathrm {i}}$ is the noise spectral density at modulation frequency measured by a signal analyzer (SR770). Only the frequency range from 0.2 to 1.1 THz of coherent terahertz source is available in our laboratory. The optical responsivity $R_\mathrm {i}$ and NEP in the range of 0.2-1.1 THz are shown in Fig. 3 at the optimal gate voltage at 298 K and 77 K, respectively. As shown in Fig. 3(a), at 298 K, an average responsivity $R_\mathrm {i}$ of 15.5 mA/W at $V_\mathrm {g}$=−3.64 V is achieved. At 77 K, the average responsivity at $V_\mathrm {g}$=−3.38 V is increased by a factor of 16.6 to about 257 mA/W. The responsivity enhancement factor at a low temperature generally tends to rise with frequency, i.e., from 7 to 30. As shown in Fig. 3(b), the average NEP at 298 K is about $188~\mathrm {pW/\sqrt {Hz}}$ at $V_\mathrm {g}$=−3.64 V. At 77 K, the average NEP at $V_\mathrm {g}$=−3.38 V is reduced by a factor of 10 to about $19~\mathrm {pW/\sqrt {Hz}}$. A minimum NEP is $7.0~\mathrm {pW/\sqrt {Hz}}$ at 0.74 THz, which is comparable with that of SBD detectors and the best narrowband Si-CMOS detector at room temperature [28].

 

Fig. 3. (a) Measured current responsivity and (b) NEP of detecter as a function of the terahertz frequency under continuous-wave coherent irradiation from 0.2 to 1.1 THz at $V_\mathrm {g}$ = −3.64 V at 298 K and $V_\mathrm {g}$ = −3.38 V at 77 K. The ratio of the responsivity and the NEP at 77K to that at 298K is plotted to the corresponding right axis.

Download Full Size | PDF

Subsequently, the broadband response of the detector to the incoherent broadband terahertz radiation from a blackbody at the temperature of 1173 K is examined to a much broader frequency range than that of the coherent source. The measured blackbody spectroscopy $i_{\mathrm {m}}(\omega )$ at 77 K and 298 K are plotted as the blue and grey dotted curves shown in Fig. 4(c), respectively. At 77 K, a consecutive response from 0.2 to 4.0 THz is obtained at optimal operation gate voltage $V_\mathrm {g}$ = −3.38 V, which has a center response frequency of 1.5 THz. While at 298 K, a consecutive response becomes from 0.2 to 2.2 THz at optimal $V_\mathrm {g}$ = −3.64 V, with a center response frequency of 0.8 THz. The bandwidth of the spectrum at 298 K is significantly narrower than that at 77 K. The noise floor at 77 K and 298 K are plotted as the black solid lines in Fig. 4(c). The noise floor is the root mean square of the noise power spectrum $P_\mathrm {noise}(\omega )$ obtained from the fast Fourier transform of the current noise $i_{\mathrm {noise}}(t)$. The current noise $i_{\mathrm {noise}}(t)$ is related to the noise spectral density $N_{\mathrm {i}}$ of the detector and the integration time $T_{\mathrm {c}}$ of the lock-in amplifier, which can be expressed as $i_{\mathrm {noise}} \propto N_{\mathrm {i}}/\sqrt {4T_{\mathrm {c}}}$. The integration time $T_{\mathrm {c}}$ is 2 s at 77 K and 5 s at 298 K, respectively. And the measured noise spectral density $N_{\mathrm {i}}$ is 4.1 $\mathrm {pA}/\sqrt {\mathrm {Hz}}$ at 77 K and 2.7 $\mathrm {pA}/\sqrt {\mathrm {Hz}}$ at 298 K. The maximum signal-to-noise ratio (SNR) obtained for measured blackbody spectroscopy is 16.5 dB at 77 K and 6.1 dB at 298 K, respectively. The terahertz transmission coefficient in air is plotted as the grey dashed curve in Fig. 4(c) as a Ref. [29]. The measured spectroscopy at 77 K clearly reveals water vapor absorption peak in air. But the measured spectroscopy at 298 K is limited by the SNR, resulting in that water vapor absorption peaks at some frequencies are not clearly visible. According to the self-mixing model in Eq. (1), the photocurrent is proportional to the received terahertz energy flux ${P}_0 (\omega )$ and two key factors: $\Xi$ and $\Lambda (\omega )$. Field-effect factor $\Xi$ does not change with frequency at a given gate voltage. Antenna factor $\Lambda (\omega )$ determined by the antenna design demonstrates a relatively flat broadband response in 0.2-4.0 THz as shown in Fig. 4(a). The formation of simulated spectroscopy $i_{\mathrm {s}}(\omega )$ from the antenna factor $\Lambda (\omega )$, Planck’s blackbody radiation law $P_{0}^{\prime }(\omega )$ and the field-effect factor $\left (\Xi = 8.6~\mathrm {mS}/\mathrm {V}\right )$ at $V_\mathrm {g}$ = −3.38 V is plotted as the orange solid curve in Fig. 4(c). The simulated spectroscopy $i_{\mathrm {s}}(\omega )$ does not quite match the measured from 1.75 to 4 THz. We have built a model of the high-resistivity silicon hyper hemispherical lens coupled with the chip in EM simulation software. The simulated average power density in the antenna area (400 $\mu$m $\times$ 300 $\mu$m) on the surface of the chip under plane wave irradiation with an incident power density of 1.33 $\mathrm {mW/m^2}$ is obtained. The ratio of the power density after convergence with the hyper hemispherical silicon lens to the incident power density is defined as the coupling factor $\eta (\omega )$ of the lens, as shown in Fig. 4(b). The coupling factor $\eta (\omega )$ is used to describe the convergence effect of the silicon lens on terahertz at different frequency. The coupling factor $\eta (\omega )$ has revealed that the hyper hemispherical silicon lens coupled with the chip provides good convergence of terahertz waves in the frequency range of 0.2-1.75 THz, with a dramatic reduction in convergence at frequency beyond 1.75 THz. The simulated spectroscopy $i_{\mathrm {n}}(\omega )$ is obtained by considering the effect of the coupling factor $\eta (\omega )$ on the basis of $i_{\mathrm {s}}(\omega )$. The center frequency and overall spectral shape of the measured spectroscopy at 77 K match well with the simulated one $i_{\mathrm {n}}(\omega )$. Especially from 1.75 to 3 THz, $i_{\mathrm {n}}(\omega )$ is in better agreement with measured spectroscopy $i_{\mathrm {m}}(\omega )$ at 77 K than $i_{\mathrm {s}}(\omega )$. This could indicate that the response bandwidth of the detector is mainly limited by the silicon lens. The response bandwidth of this detector could be greatly expanded with a higher-bandwidth coupling element. The difference between the simulated spectrum $i_{\mathrm {n}}(\omega )$ and the measured spectrum at 77 K from 0.5 to 1 THz may be due to insufficient simulation accuracy of the antenna factor $\Lambda (\omega )$ and coupling factor $\eta (\omega )$. Meanwhile, the fact that the electron mobility decreases with increasing frequency is ignored [30]. This might be the reason for the poor fit at 3.0 to 4.0 THz between the simulated spectroscopy $i_{\mathrm {n}}(\omega )$ and measured spectroscopy at 77 K. One possible conjecture is that as the frequency increases, the decrease in mobility will be stronger at 298 K than that at 77 K because the degree of ionized impurity scattering increases [31]. This may account for the difference between measured spectroscopy at 298 K and 77 K and the discrepancy between simulated spectroscopy $i_{\mathrm {n}}(\omega )$ and measured spectroscopy at 298 K. More accurate simulations in the future to investigate the impact of the underlying conducting 2DEG sheet on the self-mixing factor will be performed.

 

Fig. 4. (a) Simulated antenna factor from 0.2 to 4.0 THz. (b) Simulated coulping factor from 0.2 to 4.0 THz. (c) The blue and grey dotted curves are the measured spectroscopy at $V_\mathrm {g}$ = −3.38 V at 77 K and at $V_\mathrm {g}$ = −3.64 V at 298 K under incoherent broadband radiation from an 1173 K blackbody. The blue and orange solid curves are simulated spectroscopy. The grey dashed curve is the terahertz transmission coefficient in air.

Download Full Size | PDF

In order to acquire responsivity and NEP in the range of 0.2 to 4.0 THz, we propose an evaluation method by using measured blackbody response spectroscopy and measured coherent response performance from 0.2 to 1.1 THz. The measured blackbody response spectroscopy $i_{\mathrm {m}}(\omega )$ in Fig. 4(c) is divided by the blackbody radiation intensity calculated from Planck’s law to obtain the responsivity spectroscopy $I_{\mathrm {i}}(\omega )$, which is calibrated by the measured responsivity $R_{\mathrm {ic}}(\omega )$ from the coherent radiation in Fig. 3(a). The responsivity under incoherent radiation can be expressed as $R_{\mathrm {ii}}(\omega ) = C(\omega = \omega _0)R_{\mathrm {ic}}(\omega )/I_{\mathrm {i}}(\omega )$, where $C$ is the calibration factor at $\omega _0$ = 0.47 THz at 298 K and at $\omega _0$ = 0.6 THz at 77 K. A minimum error $\sigma$ calculated by the equation $\sqrt {\sigma ^2} = \Sigma \sqrt {(R_{\mathrm {ii}}(\omega )-R_{\mathrm {ic}}(\omega ))^2}$ can be obtained at $\omega _0$ = 0.47 THz at 298 K and at $\omega _0$ = 0.6 THz at 77 K. The black and red solid curve shown in Fig. 5(a) and (b) depict the calibrated responsivity $R_{\mathrm {ii}}(\omega )$ and NEP at 298 K and 77 K, respectively. At 298 K the NEP is about 1.7 $\mathrm {nW/\sqrt {Hz}}$ at 2.0 THz. The detector benefits from the self-mixing mechanism and the high electron mobility of AlGaN/GaN heterostructures, with response times of less than 1 ns. Also, thanks to the high breakdown voltage of AlGaN/GaN heterostructures [32], the detector at 298 K is a superb choice for high-power and high-speed modulation quantum cascade lasers (QCLs) [33], despite the sensitivity at 298 K is not optimal. The black dashed lines match the trend of responsivity and NEP in a frequency range from 0.75 to 4.0 THz at 77 K. Responsivity follows approximately an $f^{-1}$ decrease up to about 1.75 THz, as suggested by the model for nonresonant operation [34]. Then responsivity decreases with $f^{-7}$ from 1.75 to 4.0 THz. The decrease in responsivity is mainly limited by the low coupling efficiency of the silicon lens from 1.75 to 3.0 THz as shown in Fig. 4(c). The NEP follows a $f^{7}$ increase from 1.75 to 4.0 THz. At 77 K, a NEP of about $3~\mathrm {nW/\sqrt {Hz}}$ at 4.0 THz could be achieved.

 

Fig. 5. The red and black solid curves are (a) calibrated responsivity and (b) NEP of the detector at 77 K and 298 K as a function of the terahertz frequency from 0.2 to 4.0 THz. The blue and grey symbols are measured response characteristics, consistent with those in Fig. 3. The grey dashed curve is the terahertz transmission coefficient in air.

Download Full Size | PDF

Compared to typical thermal high-bandwidth terahertz detectors Golay cells and pyroelectric detectors, this detector has the advantage of higher response time ( <1 ns) and higher sensitivity below 3 THz at 77 K. However, the disadvantage is that the response bandwidth is one to two orders of magnitude worse than theirs. The NEP of this broadband detector at 77 K is lower than the typical NEP value of $140~\mathrm {pW/\sqrt {Hz}}$ for Golay cells with a response time of 30 ms [10] from 0.2 to 2.85 THz. The NEP is also lower than the normal NEP value of $400~\mathrm {pW/\sqrt {Hz}}$ for pyroelectric detectors with a response time of approximately 1 ms [35] from 0.2 to 3.2 THz. However, the NEP of this broadband detector at 298 K is higher than the NEP of the Golay cell from 0.5 THz upwards and higher than the NEP of the pyroelectric detector from 1.0 THz upwards. Compared to the state-of-the-art FET broadband detector [22], this detector has a significantly better response bandwidth and sensitivity at 77 K. At 298 K, the response bandwidth of our detector is the same, but the sensitivity is two to eight times worse. Sensitivity improvement is a key follow-up. For example, by optimizing the width $L$ and gap $D$ of the antenna, as reported in our previous work [25], the responsivity could be enhanced by roughly a factor of 5. Also, a process will be used to make the ohmic contacts on the antenna, which will effectively reduce the source-drain series resistance, resulting in the factor $\gamma$ being raised to close to 1. The responsivity could be enhanced by roughly a factor of 2. Overall, the NEP would be reduced by an order of magnitude i.e., the minimum NEP of the broadband detector could be reduced to below $1~\mathrm {pW/\sqrt {Hz}}$ at 77 K and below $10~\mathrm {pW/\sqrt {Hz}}$ at 298 K. The sensitivity-optimized detector would have greater application value at room temperature.

4. Conclusion

In conclusion, a high-sensitivity broadband terahertz detector based on an array of antenna-coupled AlGaN/GaN HEMTs arranged in a bow-tie shape is designed and demonstrated. Compared to a continuous response to blackbody radiation in the frequency range of 0.2 to 2.0 THz at 298 K, the detector at 77 K exhibits a frequency response characteristic from 0.2 THz up to 4.0 THz with a maximum optical responsivity of 0.56 A/W and a minimum optical NEP as low as $7.0~\mathrm {pW/\sqrt {Hz}}$ at 0.74 THz. The average responsivity and NEP at 77 K is 0.26 A/W and $19~\mathrm {pW/\sqrt {Hz}}$ from 0.2 to 1.1 THz. The minimum NEP of the broadband detector could be reduced to $1~\mathrm {pW/\sqrt {Hz}}$ at 77 K and $10~\mathrm {pW/\sqrt {Hz}}$ at 298 K by further reducing the series resistance, reducing the gate length $L$ and distance between the antenna and the gate $D$, etc. Meanwhile, more suitable optical coupling components and higher mobility materials, can be used to broaden the bandwidth of the detector. With optimization, this type of high-sensitivity broadband terahertz detector is expected to become a core component for terahertz applications.