Highly sensitive broadband AlGaN/GaN TeraFETs

Technological optimization of an alternate GaN MMIC process recently developed at FBH allowed the design and fabrication of a third improved generation of AlGaN/GaN TeraFETs. In the optimized fabrication process, the transistor dimensions could be even further reduced compared to the devices described above. The gate length was shortened by a factor of 2.5 down to onlyL_g = 100 nm and furthermore, the gate-to-channel dielectric could be thinned down to 12 nm. This should have an immediate effect on plasma wave generation in the FET channel, where in general, the gated plasmon frequency is directly proportional to the square root of the gate-to-channel separation [138]. The carrier density in the channel is controlled more effectively with a thinner gate dielectric as the gate capacitance rises (comp. Eq. 2.34).

A schematic of the improved GaN technology is shown in Fig. 6.15 on the left.

Again, the structures were grown on SiC substrate of roughly 470 µm thickness for backside THz illumination. We incorporated most of the design features of the

6.3 AlGaN/GaN TeraFETs

Fig. 6.16: Fit of measured DC drain-source resistance of a TeraFET with reduced gate length of 100 nm. The inset shows the result of the first fit stage. Extracted device parameters are listed in the box.

detector generation described above choosing again a broadband bow-tie antenna design. A MIM capacitor was implemented between the transistor’s source an gate terminals to realize drain coupling boundary conditions, and the metal sheets were laid out in the shape of one of the bows of the integrated bow-tie antenna. The length of the source and drain side ungated access could be further reduced to L_ug = 300 nm. The gate width was kept at 3 µm. The scanning electron microscopy image on the right of Fig. 6.15 show the bow-tie antenna design embedded in an eletric stabilization circuit as introduced before. In the magnified detail, the transistor structure with source, gate, and drain contacts as well as the implemented MIM capcitor can be recognized.

Figure 6.16 presents the fitted DC resistance of a 100 nm TeraFET with extracted device parameters in the inset box. Improved ohmic contacting was achieved by evaporation of the drain and source metals while the gate metal was deposited by electroplating. In total, the contact resistance was slightly reduced compared to the devices presented before and amounted to r_c = 0.8 Ω×m, while the sheet resistance of the Al0.32Ga0.68N gate dielectric barrier was determined at FBH⁴³ to be r_s = 610 Ω/sq. From the device geometry we calculate a total contact resistance (together for both source and drain side) of R_C = 2r_c/W ≈533 Ω and a sheet

resis-tance of R_S = 2r_s(L_ug/W)≈61 Ω. With these values we calculate an expected total ungated resistance of approximately R_ug =R_S+R_C ≈ 655 Ω which is reasonably

43Measurement of the specific sheet and contact resistances at FBH were performed by the so-called van der Pauw method.

500 520 540 560 580 600 Frequency (GHz)

0 2 4

6 ×10^-11

at 500 GHz at 600 GHz

Fig. 6.17: Absolute optical characterization measurement of the optimized AlGaN/GaN TeraFET between 490 and 610 GHz. The achieved values in optical sensitivity constitute record values for GaN-based TeraFETs in the investigated frequency region, in particular, for broadband devices.

well reproduced by parameter extraction from the DC resistance fit.

A measurement setup similar to the one described in the previous section was used to characterize the TeraFET’s broadband performance. A detailed measurement of the TeraFET’s optical NEP in the region between 490 and 610 GHz performed with the tunable, all-electronic source described in Section 6.2.2 is shown in Fig. 6.17.

The optical power in the THz beam was measured with a photoacoustic calibrated powermeter. To give a reference, we obtained power levels of 75 and 18 µW at 500 and 600 GHz, respectively, measured after the two PTFE lenses at the position of the TeraFET detector module. With these powers TeraFETs optical NEPs of 25.4 and 31.2 pW/√

Hz were determined at the above two frequencies assuming again thermal noise as the dominating noise source in our detectors. To the authors knowledge, these constitute record values for broadband-designed AlGaN/GaN HEMT-based TeraFETs in the investigated frequency range. Roughly comparable performance was reported in Ref. [116] employing a floating antenna design working around 900 GHz. Recently, optimized detectors were presented by the same group with optical NEPs of 30 to 300 pW/√

Hz at 0.7 to 1.1 THz, respectively [70]. It should be critically mentioned though that the exact procedure of power calibration is not further elucidated in the reference.

To give an estimate for the electrical NEP of our TeraFETs we assume THz optical coupling as discussed above in Section 6.2.3 including conservative estimates for the Gaussicity with respect to the employed Si substrate lens as well as reflection losses on the Si surface. The detectors were operated in aplanatic optical configuration an no losses due to internal reflection are considered. With these assumptions we calculate a frequency-independent total beam coupling efficiency of 63%. The electrical NEPs of the fabricated AlGaN/GaN TeraFETs is evaluated to yield 16 and 19.7 pW/√

Hz

6.3 AlGaN/GaN TeraFETs

0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2

Frequency (THz) 0

20 40 60 80

SNR (dB)

Golay cell

GaN TeraFET Photomixer

Parasitic antenna

resonance Water vapour

Fig. 6.18: Comparison of the SNRs of THz detection with three different detectors, namely, the AlGaN/GaN TeraFET (blue), a Golay cell (green), and the coherent photomixer receiver (red). The gray dashed lines represent linear fits reflecting the exponential power decay of

the Photomixer THz source.

at 0.5 and 0.6 THz, respectively. These values compare well to state-of-the-art Si CMOS-based TeraFETs [37] and even are almost comparable to resonant TeraFETs in the same technology [24], [139].

We further characterized the TeraFET’s over a broader frequency range from 0.2 to 1.2 THz. For this purpose, the all-electronic source was replaced with the photomixer-based tunable THz source. In the specific experimental setup, calibrated measurements of the THz power at the detector position could no be reliably performed because the power levels over the largest part of the frequency region lay at the noise level of the calibrated, large-area, photo-acoustic powermeter. We therefore determined the detector’s performance in terms of SNR and then compared the obtained results to THz detection measurements with two other detectors, namely, a commercial Golay cell and a photomixer detector module identical to the transmitting photoconductive antenna of the photomixer THz source. Fig. 6.18 shows the obtained SNRs for our TeraFET (blue), the photomixer (red) and the Golay cell (green). The measurement with the TeraFET and photomixer were performed at a modulation frequency of the THz source of f_mod = 12 kHz and a lock-in integration time constant of 100 ms. Measurements with the Golay cell at lower modulation frequency of f_mod = 20 Hz had to be performed at a much longer integration time of 2 ms due to a significantly higher noise level. The equivalent SNR at 100 ms was calculated based on the equivalent noise bandwidth (ENBW) for the applied filter slope setting of the lock-in amplifier from a measured noise voltage of 17.5 µV/√ The TeraFET’s gate bias voltage of best sensitivity was determined asV_G =−1.09 VHz.

from measurements of THz response versus gate voltage at a number of single frequencies. No significant shift of the optimal gate bias point was observed, which allowed sweeping of the THz frequency while keeping the gate bias fixed.

For all three detector’s, an overall linearly (in dB) decreasing SNR reflects the

ex-ponential decay of output power of the THz source, where the approximate slopes were extracted from linear regression and amounted to -0.034, -0.025, and -0.042 dB/GHz for TeraFET, Golay cell, and photomixer, respectively. Note that again, as discussed in the previous section, a drop in the TeraFET’s SNR around 300 GHz is observed, caused by a parasitic resonance of the ground ring surrounding the actual broadband bow-tie antenna, because the detector’s circuit design was mostly adopted from the preceding TeraFET generation. In terms of device performance, THz detection with the TeraFET yielded an SNR from 50 dB down to 17 dB over the investigated frequency range. The only noise source in the detector was assumed to be thermal noise based on the discussion in Section 6.1.1. The measured TeraFET SNRs exceeded the Golay cell’s SNR by roughly 20 dB. On the other hand, the photomixer detector showed significantly higher sensitivity at SNRs from 82 dB down to 41 dB. The reason for the higher SNR is that the photomixer was operated in coherent detection mode while the TeraFET worked as a direct power detector (comp. Eq. (2.15)). It was shown that TeraFET can as well be operated as coherent detectors in heterodyne configurations, and an enhancement of detection sensitivity of approximately 25 dB was reported [140].

Some remarks should be made about the presented data of broadband detector characterization. First, note that the fluctuations visible in all presented measure-ments - with the TeraFET and also the two reference detectors - are not to be confused with noise but rather reflect resonant standing waves in the optical setup.

The noise level of all three detectors lies well below the respective measurement signals in the investigated frequency range. Second, spectral absorption lines can be well recognized with all three detectors around 1 THz. However, the lines seem to be significantly more pronounced in the Golay cell’s SNR curve despite the lowest total sensitivity of this detector. The reason for this behavior could not be determined conclusively up to this point. We assume that spurious signals were emitted by the laser-driven photomixer source due to possible sidebands of the fundamental laser oscillation frequency. The signals are expected to be located at lower GHz frequencies and are still effectively picked up by both, the TeraFET and the photomixer detector.

It should be investigated further if the unexpected observation can decisively be attributed to a faulty operation of the THz source. A possible approach would be the implementation of an adjustable aperture in the THz path to implement a spatial frequency filter to suppress low frequency parasitic signals in the experimental setup.