Figure 2a shows the cross section of the digital X-ray detector-on-foil. It consists of (1) an IGZO-based active-matrix TFT backplane fabricated on a polyimide foil, followed by (2) an inverted stack OPD frontplane that is formed by a ~300 nm thick photoactive organic layer sandwiched between bottom pixel electrode and non-patterned optically transparent top electrode, (3) an optically transparent thin-film barrier, and finished by (4) a bendable CsI scintillator, developed by Hamamatsu Photonics (Japan).
The IGZO TFT backplane is processed on a SiNx bottom encapsulation barrier on polyimide foil/glass carrier via a series of lithographic mask steps. The display-compatible TFT fabrication process is discussed in detail both previously11 and in the “Methods”. Transfer characteristics of 25 transistors with a channel length of 20 μm and a channel width of 60 μm equally distributed over the 320 × 352 mm substrate are presented in Fig. 3a. The TFTs show a typical mobility of 15.1 ± 0.8 cm2/Vs, onset voltage (VON) of 0.4 ± 0.3 V, ON/OFF-current ratio ∼107, and a subthreshold swing of 0.4 ± 0.02 V/decade.
In the current digital detector, we employed a single pass transistor configuration per pixel, as used in most current X-ray imaging products. The TFT switch requires a high ON/OFF ratio, a reasonably sharp turn-ON region but particularly a low leakage. The low OFF-current of substantially <1 pA compares favorably to amorphous and polycrystalline silicon transistors. This is inherent to IGZO as the low leakage current is related to the large bandgap of IGZO (∼3 eV). The formation of the OPD on top of the IGZO transistor does not affect the IGZO TFT properties. This is also not expected as the OPD deposition occurs at much lower temperatures than the maximum process temperature during IGZO TFT fabrication.
The photodetector is based on a bulk heterojunction (BHJ) structure, with a donor polymer and an acceptor fullerene derivative. BHJ photodetectors show high photogeneration sensitivities, fast response times, and their absorption can be tuned from the ultraviolet to the near-infrared12,13. Moreover, they can be solution-processed at low temperatures and on flexible substrates by roll-to-roll and sheet-to-sheet large-area processing techniques14,15. In this work, we have used a 280 nm thick blend of poly[N-9′-heptadecanyl-2,7-carbazole-alt-5,5-(4′,7′-di-2-thienyl-2′,1′,3′-benzothiadiazole)], PCDTBT, a p-type (donor) polymer16,17 and [6,6]-phenyl-C61-butyric acid methyl ester, PCBM, (acceptor) fullerene18, followed by a thermally evaporated MoOx hole transport layer and a thin Ag semitransparent anode. Finally, a top transparent thin-film encapsulation barrier19 is processed (Fig. 2b). Figure 2c, d show photographs of curved image sensor and flexible scintillator, respectively. Details of the step by step fabrication can be found in the “Experimental” section.
OPD test devices having the same layer stack as the OPD frontplane and processed in the same way were also fabricated. These devices were used to measure the OPD performance and consisted of a series of 1 mm2 pixels. Figure 3b shows the current density–voltage (J–V) characteristics of the OPDs in dark (Jd) and under illumination of low intensity, white light (Jphoto). The nonzero Jd at 0 V results from transient displacement currents that occur during a J–V sweep even at very low scan rates. We consider the static measurement to be a more reliable way of determining Jd. Figure 3c depicts a static measurement of Jd at −2 V over a time interval of 5 min. Typically, the OPDs have a leakage current density of 0.3–1 × 10−7 mA/cm2 at −2 V (Supplementary Fig. 3). This value is lower than that of low-temperature a-Si PDs20,21 (10−6–10−7 mA/cm2) and on par to the 10−7 mA/cm2 value reported for high-temperature a-Si p–i–n photodiodes20. It compares favorably with other reported dark current densities for BHJ based OPDs with ~300 nm thick active layers22,23. For an OPD pixel of 126 × 126 µm2 with a fill factor of 50% this would translate to a leakage current of 8 × 10−15 A. The leakage current of the TFT used in the curved detector (W = 60 µm) is below 6 × 10−15 A, hence even lower than that of the OPD. The leakage current of the TFT is estimated from TFT measurements on devices with larger channel widths. It can be seen that Jphoto is orders of magnitude larger than Jd. We found Voc to decrease reciprocally with ln (I) with I the light intensity. The so-called ideality factor n was 1.6. Voc was determined to be 0.85 V under simulated solar light (AM1.5G, 100 mW/cm−2).
The external quantum efficiency (EQE) and spectral responsivity (SR) vs. wavelength at a reverse bias of −2 V are depicted in Fig. 3d. Both metrics peak near the emission wavelength of the CsI:Tl scintillator of 550 nm used in this work (https://www.hamamatsu.com/resources/pdf/ssd/e09_handbook_xray_detectors.pdf) at values of 48% and 0.21 A/W, respectively. Simulations show that a photodiode quantum efficiency of 48% with a dark current density lower than 10−6 mA/cm2 image quality is determined by the random variations in the number of X-ray photons that are generated and absorbed by the scintillator24. The noise current density, IN, was measured at reverse bias between 1 and 50 Hz (Supplementary Fig. 4). Below ca. 10 Hz a 1/f behavior is observed. Above 10 Hz, the frequency response was flat, at ≈8 fA Hz½. Taking this value together with SR of 0.21 A/W, the specific detectivities D* = SR √A/IN (in Jones, where Jones = cmHz½/W and A is device area). At −2 V bias, the calculated maximum specific detectivity D* = 3 × 1012 Jones (at ~550 nm). With a dark current density as low as 3 × 10−8 mA/cm2, sensitivity of 0.21 A/W, and specific detectivity of 3 × 1012 Jones in the green wavelength range, the performance of our solution-processed photodetector rank amongst the best so far for OPDs11,22,23.
A sensor array of 480 × 640 pixels, 126 × 126 μm2 in size, is built. The photodiodes are operated with a reverse bias voltage, of typically a few Volt. The positive electrodes of the photodiodes are formed by the optically transparent common electrode (Fig. 2a). The negative, patterned, and bottom electrodes of the photodiodes of a column are connected to the data line via a TFT. The gates of the TFTs of a row are connected to a common gate line. The flat panel sensor is scanned one-row-at-a-time by a dedicated row driver, in a similar way as active-matrix displays. During one frame time all the rows are sequentially selected and TFTs switch from the non-conducting to the conducting (“ON”) state. With the TFTs in the ON state charge sense amplifiers (CSAs) coupled to the data line detect the pixel current as the TFT transfers the charge from the photodiode capacitance to the data line until the voltage across the photodiode is back to its original value. Current levels are digitized using analog-to-digital converters (ADCs). More details can be found here in ref. 11.
In order to characterize the optical response, we illuminate the OPD arrays from the top, using a large-area homogenous LED light source. The mean wavelength of 540 nm is similar to the light produced by a typical CsI-Tl based X-ray scintillator.
Figure 4a shows the spatially averaged image sensor signal of all pixels in the array as a function of light intensity for different reverse bias voltage over the photodiode and two image scan rates, 2 and 4 frames per second (fps). The average light intensity is changed by keeping the driving current of the LED light source constant, and varying the pulse length from 1 ms to the maximum possible length determined by the integration time of the sensor, which is the inverse of the repetition rate (frame rate) of the sensor readout. Light intensity is measured at the position of the sensor array with a light power meter (Advantest TQ9210). With known wavelength and pulse duration the power is converted into the number of impinging optical photons per integration time. The output of the ADC in the unit least significant bit (LSB) can be converted to electrical charge with the known sensitivity of the amplifier of 900 e−/LSB.
The output signal does not depend on the frame rate of the image sensor. The signal increases linearly with light intensity before a plateau is reached indicating saturation of the photodiode capacity. From the readout charge at the plateau values the capacity of the photodiode (C = Q/V) is calculated to be 1.4 pF. This is well in line with the geometrical capacitance of 1.5 pF calculated for a capacitor with an area of 126 × 126 μm2, OPD layer thickness of 285 nm and estimated dielectric constant of 3. From the slope of the curves in Fig. 4a (log–log plot is shown in Supplementary Fig. 5) the sensitivity of the sensor and thus the EQE can be deduced. In Fig. 4b the EQE is plotted vs. light intensity using the data of Fig. 4a (4 fps frame rate only). For a bias voltage of 4.6 V the EQE is about 53%. For a wavelength of 550 nm the spectral sensitivity is R = 0.24 A/W. With decreasing bias the EQE decreases slightly, to 42% for a bias voltage of 1.3 V. These values are in good agreement with the results of single photodiodes (Fig. 3). The sudden drop in EQE at high light intensities for voltages of 1.3 and 2.2 V in the sensor array originates from saturation of the pixel capacitance.
The temporal response of the sensor array is measured by a sequence of illuminated images followed by images where the light source is switched off. As shown in Fig. 4c there is still some charge detected in the non-illuminated images, a phenomenon known as residual signal or lag also from other noncrystalline photodiodes like a-Si25. The residual signal is attributed to charges in the photodiode, which are released from deep traps after a certain time26. The residual signal is in the order of a few tenths of a percent of the signal height during illumination 1 s after end of the illumination. It is decreasing with increasing bias voltage due to faster removal of charge with higher bias field. After about 2 s the residual signal falls below our measurement limit of about 0.2%. The temporal response is comparable to standard a-Si X-ray digital detectors and thus good enough for the initial medical application. The impact of lag or “residual signal” depends not only on the frame rate, but also on the medical application. For long-term low dose X-ray video imaging (fluoroscopy) applications, a-Si-based X-ray detectors are frequently used without problems with 30–60 fps frame rate. For 3D imaging (cone beam CT) with C-arm systems they are used normally up to 30 fps, too. Here the residual signal can produce some artefacts (rings) after 3D image reconstruction, which have to be corrected by software.
An optical image is taken with a b/w photograph printed on a transparent slide on top of the sensor array (Fig. 4d). For this image a dark image is subtracted (offset correction) and a homogeneously illuminated image is used to correct for sensitivity differences between pixels (gain correction). Finally, single pixels and lines with too low or too high signal (“defects”) are corrected by simple interpolation with the neighboring pixels.
The sensor array is produced while the polyimide foil is still attached to a glass carrier. To show that delamination of the foil does not influence the performance of the sensor array, the optical characterization has been done before and after the laser delamination process. The graphs of signal vs. light intensity at two different bias voltages (Fig. 5a) show no difference between the two measurements. We conclude that the delamination process does not affect the sensitivity of the sensor array. In addition, no new defect single pixels, lines or areas with increased dark current were found comparing images before and after releasing the foil from the glass. To prove the long shelf-life, we repeated the same measurement 44 days after the delamination. In between, the sensor array has been kept at ambient air and room temperature. Again, no signs of degradation were found as shown in Fig. 5b. Hence, the bottom and top encapsulation layers, with a water vapor transmission rate WVTR of 10−6 g/m2/day19,27, serve as an excellent barrier for potential degradation by the ambient.
Subsequent to the optical characterization we tested the digital detector also with X-rays. Flexible cesium iodide (CsI) scintillator sheets of 400 and 700 µm, typical thicknesses in current state-of-the-art flat panel X-ray detectors exceeding the attenuation length of 99 µm at the X-ray energy used, were coupled to the sensor array by simply pressing them with the weight of 1 mm Al to achieve a good optical contact. X-ray images with 70 kV tube voltage and 350 mA tube current were acquired with 50 ms exposure time. No additional filtration other than the internal tube filtration (about 2.5 mm Al equivalent) has been used, i.e., X-ray beam quality is defined as RQR5 with about 40 keV average X-ray energy. With a source-image distance of about 1 m a dose of about 450 µGy air/frame was applied. Typical X-ray resolution test phantoms with thin lead lamella (Huettner, type 10) were put as contrast objects on top of the digital detector (Fig. 6).
Similar as with the optical image (Fig. 4d) an offset, gain, and defect correction were performed. The images show a good and homogeneous contrast over the whole area of the digital detector. It is clearly visible that the spatial resolution is higher with the thinner scintillator (Fig. 6a, 400 µm) layer because the lateral spread of optical photons is lower than in the thicker scintillator layer (Fig. 6b, 700 µm). The modulation transfer function (MTF), has been measured with a tungsten edge according to IEC standard IEC 62220-1 CDV. It is 31 and 37% at 1 lp/mm for the 700 and 400 µm thick scintillator. These values are in agreement with the expected MTF for an X-ray detector with a 40 µm thin-film barrier in between the image sensor and scintillator (Fig. 2a).
In case of a flat detector, oblique incident X-rays lead to a degradation of the MTF at the edges and corners, because the point of interaction in the layer determines the lateral displacement and leads to a blur. Following the analysis method of Hajdok & Cunningham28 a MTF degradation of 17% at 2 line pairs per mm is expected. More details can be found in (Fig. S1). By curving the detector with the right radius for a given source-detector distance, all X-rays are impinging perpendicular to the scintillator surface and no degradation due to oblique incident X-ray occurs. The test bench for 3-dimensional X-ray imaging (Fig. 7a) comprises a standard medical X-ray tube (Philips MRC200 0407 ROT-GS 1004), a rotational stage to place objects, and the foil-based digital detector in a curved holder with 32 cm curvature radius. The X-ray beam from the tube passes the object horizontally before hitting the detector. The tube and detector are fixed in their position. The object can be rotated in a controlled manner. This geometry is equivalent to medical CBCT, just in that case the object (patient) is in a fixed position while tube and detector are rotating.
X-ray images of a piece of bone are taken with a rate of 2 images per second while the object is rotating with 4° per second, thus 180 projection images are taken in a full 360° rotation. The steps of offset, gain, and defect correction mentioned earlier were also applied to the 2D projection images. With a standard 3D reconstruction algorithm (FDK filtered back-projection), which has been adapted for the curved geometry, a 3D volume image (Fig. 7b) of 256 × 256 × 256 voxels with a voxel size of (0.234 mm3) is created. Since 3D reconstruction is very sensitive to any nonideal behavior of the digital X-ray detector, our result with the prototype detector is quite remarkable.