1. Introduction

Terahertz (THz) imaging is becoming a valuable tool for a broad range of non-invasive inspection applications, e.g., non-destructive material testing [1, 2], imaging of water content during paper production [3], identification of chemical compositions such as drugs [4,5], or imaging of medical tissue [6]. Security-related THz inspection tasks include the screening for illicit objects and materials in packages [7] or for threats such as explosives and weapons carried on persons, the latter making close-to-real-time imaging over distances of many meters desirable [8–10].

In many applications, spectral discrimination is required. If the spectral response of the materials in a sample under test is known a priori, identification is possible via principal component analysis (PCA) [11, 12]. An example is the multi-spectral THz imaging and spatial component analysis of explosive simulants with bow-tie diodes [13,14]. However, the development of imaging systems with high frame-rates combined with spectroscopic analysis remains a challenging task. A multi-detector scheme is advisable, e.g., focal-plane arrays integrated on a chip, as they have become available in recent years [15–19], and multi-color approaches covering various spectral regimes can be pursued [20].

Antenna-coupled field-effect transistors for THz detection (TeraFETs) offer an elegant solution for room-temperature multi-spectral THz imaging. After the first imaging demonstration with GaAs-based devices in 2008 [21], TeraFETs gained large interest and showed rapid development, mainly because they can be realized with mainstream silicon CMOS technology [18, 19, 22]. Silicon TeraFETs have shown their potential to reach high sensitivities with noise-equivalent powers (NEP) down to between ten and a few tens of $\text{pW}/\sqrt{\text{Hz}}$ [22,23]. Here, we explore for the first time antenna-coupled single-FET detectors as potential candidates for multi-spectral THz imaging of explosive simulants in both transmission and reflection geometry in the range of 0.76–4.25 THz. Extracted transmittance values from the THz images are in good agreement with Fourier-transform infrared (FTIR) spectroscopy data. On this basis, the spatial content distribution of the explosive simulants is extracted from transmission data using PCA.

2. Optimized TeraFET detectors

TeraFET detectors were fabricated in a commercial 90-nm silicon CMOS process provided by United Microelectronics Corporation, Taiwan. THz radiation is coupled to the drain of each field-effect transistor by integrated patch antennas with simulated antenna directivities of roughly ∼6.3 dBi. A single-transistor layout was chosen in contrast to our previous double-FET designs [19,23,24]. The rectified signal is read out at the symmetry axis (virtual ground) of the antenna. TeraFET detectors were commercially wire-bonded and packaged into PCB-board-ready detector chips for front-side illumination. The NEPs of TeraFETs with antennas designed for 216 GHz and 608 GHz were experimentally determined using electronic multiplier-chain sources; other TeraFETs at 2.52, 3.18, and 4.25 THz were designed to match the discrete emission lines of a molecular gas laser, and characterized with the same. NEPs of 59, 20, 63, 85 and $110\hspace{0.17em}\text{pW}/\sqrt{\text{Hz}}$, respectively, were determined using beam profile scans (some shown in [19]). Figure 1(a) shows a comparison of the obtained NEP values to those of previous TeraFET devices in 150-nm CMOS technology with double-FET design [23]. An improvement of roughly an order of magnitude is achieved at 216 GHz and between 2.5 and 4.3 THz by a combination of revised antenna design, reduction of parasitic effects and enhanced impedance matching. The detector noise was measured at 333 Hz modulation frequency using a low-noise current amplifier at zero drain bias, verifying that Johnson-Nyquist noise is the dominating noise source (see inset of Fig. 1(a)). At low gate voltages, the noise floor of the current amplifier is reached.

 

Fig. 1 (a) NEPs of 90-nm TeraFETs of this work (red circles) and those of previous works (black stars). Improved NEPs are 59, 20, 63, 85 and $110\hspace{0.17em}\text{pW}/\sqrt{\text{Hz}}$ at 0.216, 0.590, 2.52, 3.11 and 4.25 THz, respectively. Inset: Noise current of a 90-nm TeraFET (circles) and Johnson-Nyquist noise (green line). R is the measured gate-voltage-dependent drain-source DC resistance. (b) Transmittance of explosive simulants measured with an FTIR (solid lines) and the 90-nm TeraFETs (symbols).

Download Full Size | PDF

In the following spectroscopic experiments, the three TeraFET detectors resonant between 2.5 and 4.3 THz were each used at their designed resonance frequency, while a 648 GHz pixel and a 1.63 THz pixel (not characterized) were operated off-resonance at laser frequencies of 0.7619 and 1.8397 THz, respectively. However, even for these frequencies, both TeraFETs were sufficiently sensitive for imaging application.

3. Transmission imaging of explosive simulants

Crystalline sucrose (SC) and tartaric acid (TA) are known to exhibit absorption features which are similar to those of PETN and RDX powder, respectively, constituting the energetic materials of commonly used plastic explosives [25]. SC shows two strong absorption peaks at 1.83 and 2.85 THz, which, apart from a small shift, can be identified with absorption maxima of PETN at 2.00 and 2.84 THz, respectively. For RDX, absorption lines at 0.8, 1.6 and 2 THz are well reflected by three peaks of its simulant TA at 1.08, 1.83 and 2.61 THz, where again the spectrum is slightly shifted [25–27].

Samples were prepared mixing polytetrafluoroethylene (PTFE) powder with crystalline SC and/or TA, and pressing the mixtures into pellets of 1.3 mm thickness and 13 mm diameter (details in [14]). Three samples were fabricated with defined amounts (weight/weight) of explosive simulants, i.e., one pellet with 10% SC, one pellet with 10% TA and one pellet containing a mixture of 5% of each SC and TA. A fourth pure-PTFE pellet was fabricated to act as a reference sample. Figure 1(b) shows transmittance spectra of all samples recorded in an evacuated FTIR spectrometer over the range from 0.59 to 5.09 THz with a resolution of 60 GHz.

Transmission images were recorded with a raster-scan imaging system [14] at five discrete frequencies of a CO₂-laser-pumped molecular gas laser operating in continous-wave mode, namely 0.7619, 1.8397, 2.5242, 3.1075, and 4.2475 THz. A chopper was used to modulate the laser beam at 520 Hz, which was split into a reference path for monitoring the laser output power and a detection path, where the beam was focused onto the samples by an off-axis paraboloidal mirror (OAP). The samples were scanned through the focus to obtain two-dimensional THz images. The focal spot diameter was measured to be 0.5 to 1.1 mm depending on wavelength and laser mode [19]; the pixel size of the images was 0.1×0.3 mm². Using a pair of OAPs, the transmitted beam was guided to a detector chip containing the TeraFETs described above. The rectified signal was read out by a lock-in amplifier with a time constant of 10 ms. In addition, we performed normal-incidence reflection imaging. In this case, a high-resistivity silicon wafer was used as a beam-splitter to extract the reflected THz beam from the beam path.

Figure 2(a) shows THz transmission images of the test samples. The samples were mounted on a metallic holder, which served as a noise level reference. Images at various THz frequencies were recorded via switching between discrete THz laser lines by tuning the laser cavity and/or exchanging the gas (HCOOH for 0.7619 and 1.8397 THz and CH₃OH for 2.5242, 3.1075 and 4.2475 THz). After each frequency change, the position of the TeraFET chip was realigned to the corresponding detector and tuned for maximal signal. The images exhibit a dynamic range of >22 dB, only the measurement at 1.84 THz shows a slightly higher noise level and a reduced dynamic of ∼18 dB, which can be attributed to the off-resonant operation of the 1.63 THz TeraFET. A distinction of the pellets with pure SC and TA content is readily possible from the different contrasts at 0.76, 1.84, 3.11 and 4.25 THz. The SC pellet exhibits a higher transmittance at 0.76 and 3.11 THz, while that of the TA pellet is higher at 1.84 and 4.25 THz. At 2.52 THz, a clear distinction is more difficult, although the TA pellet is slightly more transparent than the SC pellet. It can also be observed in the 4.25-THz data, that the absorption by SC is that strong that both the pure-SC pellet as well as the mixed-component pellet show close-to-zero transmission; the pellet areas exhibit a transmission level similar to that of the metallic sample holder. The PTFE pellet shows only weak absorption at all the measured frequencies and has thus been used as the transmission reference for further evaluation.

 

Fig. 2 (a) Multi-spectral THz transmission images of explosive simulants. The images were acquired with various TeraFET detectors at the labeled THz laser lines. (b) Spatial distribution of SC (blue) and TA (red) determined by PCA.

Download Full Size | PDF

By averaging over the pellet areas, transmittance values were extracted from the images at the five distinct frequencies. Due to visible diffraction effects at the pellet edges, the pellet’s rims were excluded from the averaging. The noise level of each measurement was determined at the area of the fully opaque metallic sample holder. The open symbols in Fig. 1(b) display the resultant mean transmittance values. They agree quantitatively with the values obtained in the FTIR spectroscopy measurements as shown in Table 1. Comparing with the full spectra, it becomes clear that the substantial contrast difference at 0.76, 1.84, 3.11 and 4.25 THz results from the significantly different absorption strength of the two chemical compounds at the respective frequencies. The contrast at 2.52 THz is weak because both compounds strongly absorb there. Taking into account the correspondence of the THz spectra of SC and TA with PETN and RDX, respectively, it can be concluded that the presented TeraFET-based multi-spectral transmission imaging system is well suited for realistic explosive detection scenarios.

Table 1. Transmittance of test samples at frequencies f recorded with TeraFETs and FTIR.

View Table

Multi-spectral imaging principally allows for the determination of the lateral distribution of the chemical compounds across pellets provided the focal spot size is larger than the ”grain” size of the compounds. In the data of Fig. 2(a), a spatial variation of the signals is visible which is high above the noise level of the measurements. The variation is caused by locally differing concentrations of the compounds in the PTFE matrix which leads to variations in absorption and to scattering. Assuming that the first effect is dominant because density variations are rather small, the lateral distribution of SC and TA within each pellet is found using PCA [11, 13]. Figure 2(b) shows the spatial distribution of SC and TA for all four samples, where shades of blue and red represent the amount of SC and TA, respectively. Mixed colors within the SC-TA-mixed pellet are obtained by additive RGB color mixing. For the pure samples, the spatial variations found in the analysis indicate that the mixing procedure did not result in a homogeneous material distribution. The fact that the pure-PTFE pellet exhibits a color signal at its edge, is another indication of diffraction and scattering effects at the pellet’s rims.

4. Reflection imaging

Reflection measurements are of great value when aiming for stand-off detection of hidden objects in combination with spectroscopic identification of materials, which is desired in most realistic security application scenarios. However, reflection data are notoriously difficult to interpret [27]. The reflectance spectra of the explosive simulants and the reference PTFE pellet were measured with the FTIR spectrometer (data not shown). It was found that a 10% amount of SC or TA component in PTFE matrix is not an optimal amount for a deeper analysis of spectroscopic THz images in reflection mode. For comparison, experimental data from Ref. [28] were taken for 2.52 THz frequency for which the TeraFET detectors were designed. It is found that the largest reflectivity modulation for RDX and HMX in case of 100% of material purity amounts to only 1% and 3%, respectively. In our case, the content is approximately 10 times smaller. Hence, the reflection signal modulation due to presence of the explosive simulants is difficult to measure by both FTIR spectroscopy and multispectral THz imaging with TeraFETs. In addition, in realistic application scenarios, misalignment between samples as well as bowing and scratching of the surfaces will further deteriorate the reflection signals. Not least for this reason, a full analysis of reflection images is beyond the scope of this paper but is, e.g., discussed in Ref. [29] exploring the diffusive scattering of powdered samples by using THz reflection imaging with a quantum cascade laser operating at 2.8 THz. There, reflection images of samples of mixtures of polystyrene and polymethyl methacrylate powder are discussed on the basis of a number of back-scattering theories.

It has to be noted again, however, that our work is different in the sense that we use samples with relatively low amounts of additives below 10% (weight/weight) and an analysis with a similar approach is not easily possible. Here, we only want to address the following points. First, the dynamic range achieved in multi-spectral reflection-mode imaging with the TeraFETs is sufficient for subsequent analysis, as proven by the data displayed in Fig. 3, which shows images of the four pellets at 0.76 and 2.52 THz. A metallic mirror was placed close to the objects and served as a perfect-reflector reference. The dynamic range was found to be more than 15 dB. Second, the image of the pure PTFE pellet at 2.52 THz exhibits interference patterns. In reflection, these are more pronounced than in transmission because of the relatively small reflectance values. Clearly, such artifacts need proper attention and data processing prior to further analysis. However, with regard to the TeraFETs, they show their potential for spectroscopic THz imaging in stand-off detection geometries.

 

Fig. 3 THz reflection images of the explosive simulants at labeled THz laser lines.

Download Full Size | PDF

5. Conclusion

To summarize, TeraFETs fabricated in a commercial 90-nm CMOS process technology, are found to be well-suited for room-temperature multi-spectral THz imaging in the frequency range of 0.76–4.25 THz. To our knowledge, this is the first demonstration of multi-spectral imaging up to 4.25 THz with antenna-coupled FETs designed for multiple frequencies. THz images of PTFE pellets with explosive simulants exhibit a dynamic range of more than 22 dB in transmission geometry and of more than 15 dB in reflection mode using a molecular gas laser as radiation source. The multi-spectral THz transmission images allow to discriminate the content and to evaluate the spatial distribution of the materials under test. All results are found to be in a good agreement with data obtained by Fourier-transform infrared spectroscopy and the suitability of the TeraFETs for further studies on multi-spectral THz reflection imaging was shown. The presented work opens a route for the development of multi-frequency TeraFET arrays to reduce the measurement time for multi-spectral THz imaging and to enable real-time component analysis in the THz region up to 4.25 THz.