Photomultiplication Enabling High‐Performance Narrowband Near‐Infrared Organic Photodetectors

Continuous monitoring of food quality, blood oxygen, or industrial processes require high‐throughput near‐infrared photodetectors. Due to excellent properties like low‐cost fabrication, flexibility and narrowband response, organic photodetectors (OPDs) have a huge market potential for such applications. An organic donor–acceptor blend with a low‐energy and broad charge transfer (CT) feature is utilized, circumventing the difficulties of obtaining organic materials with significant absorption beyond 1000 nm. The increasing recombination of such low‐energy gap materials that is detrimental for the quantum efficiency is overcome by applying two photocurrent multiplication (PM) mechanisms to the donor–acceptor blend. Combined with an optical micro‐cavity, this OPD achieves a spectral response (SR) of 15 A W−1 at 1092 nm. With its spectrally narrow response of only 18 nm, this OPD technology can be used for highly resolved measurements. Contrary to OPDs working in the photovoltaic mode, this detector is optimized for operation under reverse bias. With its high spectral response, low‐cost readout circuitry like CMOS can be used for signal detection.


Introduction
Over the last few years, the performance of organic photodetectors (OPDs) expressed by the specific detectivity has continuously improved, reaching or exceeding the performance of their inorganic counterparts such as silicon, indium gallium arsenide (InGaAs) or germanium [1] at least in the ultraviolet and visible region. However, medical monitoring, material sensing, foodquality control, and artificial vision system require near-infrared (NIR) detection due to unique absorption features of various DOI: 10.1002/aelm.202201350 materials, allowing their identification. Furthermore, fog and human skin are more transparent in the NIR compared to the visible spectrum. Low-cost NIR photodetectors could revolutionize the market with cheaper and more versatile solutions, improving life quality by automatized daily tasks. Inorganic semiconductors like Si, Ge, or InGaAs show high specific detectivities and a broad detection range. Additional features like narrowband detection, having bio-compatibility, flexibility, and being lightweight are of equal importance for modern applications. Here, organic semiconductors find an excellent application perspective due to their intrinsic properties. Devices processed by thermal vapor deposition are of special interest since they can be easily scaled, integrated with readout circuits, and have an established fabrication technology developed for OLED manufacturing. While many applications utilize broad absorption features, the characteristic absorption of molecules in the NIR can be used as a unique fingerprint to identify materials and their composition. There are two key parts to achieve this type of material analysis; absorption of the active layer in the NIR and a tunable narrowband response. Most materials in the organic photovoltaic community show visible to deep red absorption. There are fewer reported materials for wavelengths above 1000 nm, since their lower optical gap leads to worse open-circuit voltages and power conversion efficiency. [2] The reported external quantum efficiency (EQE) for devices without gain can range from a few percent up to 50%. [3][4][5][6] They exhibit specific detectivities up to 10 12 Jones. However, these values are based on thermal noise or shot noise calculations. As shown in this contribution and other works, the real noise can differ by multiple orders of magnitude from the calculated one. [7] For devices with an intrinsic gain and response above 1000 nm, the measured specific detectivity is in the range of 10 9 Jones. [8,9] To the best of our knowledge, so far, there are no publications on OPDs combining properties like very narrowband, easy tunability, and a high spectral response above 1000 nm.
For our intentions, narrowband photoresponse is needed, which for the inorganic technology requires additional optical elements consisting of bulky dispersive components as gratings or prisms, unendearing the photodetector's production and their application. Organic photodetectors offer monolithic approaches www.advancedsciencenews.com www.advelectronicmat.de to accomplish narrowband photoresponse, which are less cumbersome, much cheaper and robust. [10][11][12] By tuning organic molecules, narrow absorption with full width at half maximum (FWHM) in the range of 100 nm has been demonstrated. [13][14][15][16] However, depending on the use case, FWHMs in the range of 25 nm are required for practical applications. [17] Thus, different approaches have been proposed to achieve easily tunable narrowband photoresponse. By introducing very thick active layers, Armin et al. demonstrated an architecture that selectively produces current depending on the location where photons are absorbed. [18] The so-called charge collection narrowing process (CCN) generated responses at the absorption edge with FWHM below 100 nm. [19] By introducing a very low acceptor (or donor) concentration, artificial traps are introduced. The same architecture like in CCN can be used to achieve an electric gain, called charge injection narrowing (CIN), while retaining the narrow response. [20,21] Despite offering a reasonable performance, this design requires 1 μm thick active layers while offering a limited peak position tuning. Another approach uses optical micro-cavities to enhance the optical field within the device, where specific wavelengths constructively interfere, leading to narrowband photoresponse. The main advantage over the CCN approach lies in the reduced thicknesses required for peak tuning, which, besides cheapening the production, also makes it a more functional means of achieving wavelength selectivity. Apart from that, the targeted wavelength can be easily predicted by the device thickness L, the effective refractive index n, and the order of resonance m: This mode has a spectral width depending on the number of roundtrips one photon does before being absorbed. Low absorption within the whole device and high reflection of the mirrors should be present to achieve narrow spectral features. In this work, the absorption feature of the charge-transfer (CT) state is used for the NIR wavelength range. The CT state forms between the donor's highest occupied molecular orbital (HOMO) and the acceptor's lowest unoccupied molecular orbital (LUMO). Due to its intermolecular nature, the absorption is much weaker and, therefore, suitable for cavity enhancement. By carefully selecting the donor and acceptor molecules, we are able to shift the absorption far into the NIR, with up to 1665 nm already observed. [10,22] The silver electrodes, acting as reflective mirrors, introduce parasitic absorption to the cavity that can be significantly higher than the CT state absorption. These photons are lost and do not contribute to the device's response. Depending on the absorption coefficient of the CT state, the parasitic absorption can be 95% or even more. [12] Together with the roll-off in absorption of the CT states, they lead to a low EQE. [11] While distributed Bragg reflectors have been shown to yield higher EQE due to their low parasitic absorption, they add more complexity to the overall device. [23] The low response and specific detectivity of up to 10 10 Jones at zero bias require high sensitivity for readout circuits, increasing the system cost. Furthermore, for standard p-i-n OPDs, the highest specific detectivity is achieved at zero bias voltage, which is incompatible with CMOS readout technology. By implementing photomultiplication (PM) into OPDs, small electrical signals can be increased by many orders of magnitude. Contrary to inorganic photodiodes, where the avalanche effect is used for photomultiplication, the disordered nature of most organic solids and the high exciton binding prevent the avalanche effect from taking place. [24,25] While there are different approaches to achieving photocurrent gain in organic semiconductors, they often introduce localized energy level band bending by an accumulation of one charge carrier type, as first shown by Hiramoto et al. in 1994. [26] Due to the similar phenomenon of observing a photoinduced gain in both organic and inorganic semiconductors, we will refer to this process as photomultiplication. The band bending can be done by trap states at interfaces, [27][28][29] dedicated trapping centers, [30][31][32][33] and ultra-low donor or acceptor concentrations. [20,[34][35][36] It has also been shown that single molecule OPDs can trap charge carriers with designated side groups. [37] These localized holes (electrons) introduce a Coulomb force to the electrons (holes) from the circuit, enhancing their tunneling probability through the injection barrier. Upon light illumination, a mixture of photo-excited electrons (holes) and electrons (holes) from the circuit are dragged through the device by the external bias. This increases the number of charge carriers that go through the device and thereby generates more current than photons being absorbed, which is interpreted as photomultiplication. [38] This process is depicted in Figure 1. If the lifetime of accumulated charge carriers is longer than the transit time, additional charge carriers can pass through the device compared to the photo-excited ones. Consequently, internal quantum efficiency (i.e., ratio of absorbed photons and generated charge carriers) above unity can be observed. [36] However, the low energy of the charge-transfer state between the D:A interface results in higher recombination rates of charge carriers. [39] This will decrease the trapping efficiency and therefore reduce the gain mechanism.
In this work, we present a low energy-gap organic photodetector that combines two mechanisms to achieve photomultiplication in organic blends. The ultra-low donor concentration decreases the number of percolation paths for holes, while an additional blocking layer is needed to increase the accumulation and thereby achieve EQEs above 100%. The fully vacuum-processed OPD stack is embedded into an optical micro-cavity. The broad and low energy absorption of the CT states is used for detection above 1000 nm, which is enhanced by fine-tuning the donor concentration within the active layer to the field intensity. A spectral response of 15 A W −1 at 1092 nm with an FWHM of 18 nm is realized.

Photomultiplication in D6:C 60 Blends
A versatile cavity-enhanced, NIR OPD with a narrow response must have a broad and significant CT absorption in the NIR. Therefore, we use a donor with a high occupied molecular orbital (HOMO) level, 2,2,6,6-tetraphenyl-4,4-dithiophenylidene (D6), [22] and C 60 as electron acceptor. Figure 2a shows the EQE of an OPD consisting of a D6:C 60 blend with a broad CT absorption feature below the absorption of the neat layers. We attribute this sub-gap response to the intermolecular CT state absorption, where the CT state energy (E CT ) is 0.95 eV with a reorganization energy of 0.20 eV. [40,41] First, we use a very small donor concentration to decrease the number of percolation paths for holes since they should accumulate close to the anode when applying a negative bias to the device. Between the active layer (AL) and the anode, we insert an injection barrier (IB) of N,N,N″,N″-tetrakis(4-methoxyphenyl)benzidine (MeO-TPD) to decrease the dark current. Upon illumination, the energy level bending induced by the accumulation of holes should increase the tunneling current of electrons from the anode through the IB into the active layer. However, in Figure 3a, even for a very low donor concentration of 2 wt% and a reverse bias of −10 V, the EQE does not exceed 100%. [22,[42][43][44][45][46] By inserting an additional hole-blocking layer (HBL) between the AL and the IB, the unintended extraction of holes from the anode can be suppressed. We employ three different materials to investigate their blocking character in this device. In Figure 2b, the ionization potential and electron affinity of hexaazatriphenylene-hexacarbonitrile (HAT(CN) 6 ), C 60 , and 1,4,5,8-naphthalenetetacarboxylic dianhydride (NTCDA) are depicted. [44,45,47] The EQE of the device with C 60 HBL is well below 100% and similar to the performance of the devices without the HBL. This suggests that there is significant hole transport present in the C 60 layer and a lower D6 concentration combined with other materials need to be used to generate the required hole accumulation. [48] By introducing an intrinsic NTCDA layer of 10 nm, photomultiplication is achieved with EQE values in the range of 1000% for visible light. At the same time, the dark current is increased drastically, an effect which is less pronounced when choosing HAT(CN) 6 as HBL, cf. Figure S1a, Supporting Information. In Figure S4, Supporting Information, it is demonstrated that the EQE further increased if the donor concentration is decreased; reaching values above 100%. However, at the same time the EQE for wavelengths above 800 nm is decreasing with less D6 content, which we attribute to the reduced CT state density due the decreasing D:A interface. [40] Later, we propose a strategy to maintain high gain while not sacrificing CT absorption. Figure 2. a) Normalized absorption of C 60 and D6 in dashed lines and the normalized EQE of a bulk heterojunction OPD with 10 wt% D6. A very broad and strong CT feature is visible at wavelengths above 800 nm. b) Stack structure of final device architecture, including the glass substrate, silver anode (Ag), injection barrier (IB), hole blocking layer (HBL), active layer (bulk heterojunction of C 60 and D6), electron transport layer (ETL) and silver cathode (Ag). c) Sketch of molecular energy levels of the materials used in this work. [22,[42][43][44][45][46]  However, high EQE or responsivity is not the only figure characterizing the performance of photodetectors. The capability to detect faint light signals is expressed by the specific detectivity D*. It combines the EQE with the power spectral density (PSD) generated by noise currents of the device: where q is the elementary charge, h is Planck's constant, and c is the speed of light. Apart from thermal noise and flicker or 1/f noise, shot noise is present in our OPDs. It arises from the dark current density j dark and is expected to be the main contribution in our detectors because low E CT blends generally have higher dark currents. [2,49] For a shot noise dominated device, the PSD can be estimated by: With that, the specific detectivity can be approximated by: with Δf as the electrical bandwidth. An estimate of the voltage dependent specific detectivity is done by utilizing the current voltage characteristic at 1000 W m −2 illumination condition. By assuming a linear relationship between the EQE and the photocurrent density j(V), a voltage dependent EQE(V) is calculated from the known values at −10 V bias denoted by a subscript. In Figure S5, Supporting Information, it is shown that this assumption is valid for our devices in the relevant voltage range.
In Figure 3d, the EQE of a device using an additional electron transporting layer (ETL) of 2,3,8,9,14,15-hexachloro-5,6,11,12,17,18-hexaazatrinaphthylene (Hatna-Cl 6 ) placed between the AL and the cathode is shown. Even though the insertion of this layer results in a small reduction in the EQE, the dark current is significantly decreased, especially at high reverse biases. To see the effect on the specific detectivity, an estimation according to Equations (4) and (5) is presented in Figure 3f. This additional electron transporting layer (ETL) suppresses the drop of specific detectivity at higher reverse bias voltage, enabling the device operation over a broader range of voltages.
The final optimized device architecture consists of 6 layers: cathode (90 nm ITO or 30 nm Ag), injection barrier (10 nm MeO-TPD), hole blocking layer (10 nm HAT(CN) 6 ), active layer (D6:C 60 440 nm, 0.5 wt%), and electron transport layer (15 nm n-doped Hatna-Cl 6 ). The molecular structures can be found in Figure S2, Supporting Information. Figure 1 depicts the working mechanism of our PM-OPD device comprising all layers of the final architecture. The injection barrier suppresses the tunneling of electrons from the anode into the device. Similarly, the electron transport layer suppresses holes entering the active layer from the cathode. Only charge carriers generated within the device are present in the active layer.

PM-OPDs Optimized for Optical Micro-Cavities
By exchanging the ITO semitransparent bottom anode with a semitransparent and highly reflective 30 nm thick silver layer, an optical micro-cavity is formed. Photons are reflected by both electrodes, effectively creating a standing wave with high field amplitude for specific wavelengths with constructive interference. The absorption of these photons is thus much higher compared to the device structure with ITO as an anode. Thereby, this design yields two benefits: the absorption of specific wavelengths is boosted, and the device shows narrowband response. Here, we focus on cavities with resonances at 1100 nm. However, Kaiser et al. showed that for this and similar material systems, resonance wavelengths could range from 800 to 1600 nm. [22] By running optical simulations based on the transfer matrix method (TMM), we infer the active layer thickness for resonance to occur at 1100 nm. In Figure S3a, Supporting Information, the simulated absorption of thin cavities with first-order and thick cavities with second-order peaks are shown. At 1100 nm for both architectures, the simulated absorption is almost 100%. In addition to this, in Figure 4a, there is the relative EQE (normalized to highest EQE value) for both stack designs measured. For the 1st order device, the EQE is measured at zero bias since it is not stable when operated under reverse bias, which might be a consequence of the thinner active layer.
This device has a 20 times lower relative response than the thick device at the resonance of 1100 nm. As the composition of each layer is identical for both designs, we attribute part of this strong deviation in relative EQE to absorption occurring in the silver mirrors since these absorbed photons thermally relax and  6 as HBL and 5 wt% donor concentration at the cavity field maxima. a) Spectral response at various light switching frequencies with an intensity of 225 W m −2 at 660 nm wavelength. b) Slow measurement of sensitive EQE (4 Hz) and the corresponding specific detectivity in the shot noise limit. The EQE resonance peak is fitted to a Lorentzian function, from which a peak EQE of 3000% at 1092 nm with a FWHM of 18 nm is obtained. c) Power spectral density under dark conditions. All measurements are conducted at −9 V reverse bias. therefore cannot contribute to the photocurrent. This parasitic absorption is more prominent for the thinner (1st order) compared to the thicker (2nd order) device since the fraction of photons being absorbed per roundtrip by CT states is lower, while the absorption of the silver mirrors is unaffected. For the same number of round trips in the resonator, the fraction of absorbed photons in the active layer to the photons in the metal contacts is higher in the 2nd order devices.
In Figure S3b, Supporting Information, the parasitic absorption in the simulation is shown. The second-order device (simulated in Figure S3c, Supporting Information) "standard device" has a 100% active layer absorption than to the simulated first order device. To maximize the EQE, even higher absorption in the NIR is required. While thicker cavities with 3rd or 4th resonance orders could achieve this, such thick OPDs have a large material consumption with potentially higher recombination rates. Furthermore, light coherence might play an important role for devices with thicknesses above 1 μm. Therefore, we stick to cavities that use the second-order resonance peak for the desired wavelength. In Figure 4d, TMM simulations of the entire PM-OPD stack reveal that for these cavities, two field maxima exist within the active layer. Since absorption depends on the field intensity, most photons are absorbed in the region of high fields, which can be seen in Figure S3c, Supporting Information. By increasing the donor concentration in these high field locations, cf. Figure 4b, we can increase the density of CT states, which is a prerequisite to producing photocurrent. [50] In Figure 4c, devices are presented that have a low donor concentration of 0.5 wt% but a higher donor concentration at locations where the electric field for the resonant wavelength of 1100 nm is at its maximum, as demonstrated in Figure 4b. As expected, the devices with higher donor concentration and, therefore, with more CT states show higher EQEs. Here, we would like to stress we also tried to keep the regions with low donor concentration as large as possible since that has been shown to increase the photomultiplication effect in Figure 2a, besides keeping the dark current low, as shown in Figure S1, Supporting Information. Even though the EQEs of the PM-OPDs do not exceed 100%, this is a significant improvement compared to OPDs based on this donor-acceptor blend working in the photovoltaic mode, for which EQEs of about 0.4% have been reported. [22]

Detailed Characterization
In-depth characterization of the device with 5 wt% donor concentration in the field maximal reveals a strong spectral response (SR) dependency on the frequency of the signal with a −3 dB cutoff frequency of 10 Hz, cf. Figure 5a. The slow response of the PM-OPD is a consequence of the hole trapping. Just after recombination of photogenerated holes, no tunneling of the electrons into the device is possible. Unfortunately, this process is rather slow and decreases the speed compared to photovoltaic OPDs.
Due to the low cut-off frequency, a more sensitive EQE measurement with a slow light chopping frequency of 4 Hz is conducted. The measured EQE at the peak positions is much higher as compared to Figure 4c, however, this is in agreement with the f-3 dB behavior of the device. At this condition, the peak is fitted by a Lorentzian function, from which an EQE of 3000% at 1092 nm is obtained. In Figure 5b, the outstanding FWHM of 18 nm for this detector is shown. By assuming the shot noise limit, the specific detectivity is calculated to be almost 10 13 Jones at 1092 nm, competing with Si, InGaAs and Ge photodetectors commonly used for NIR photodetection. Direct measurements of the PSD under dark conditions in Figure 5c reveal a strong 1/f noise contribution. In real-world applications, this will reduce the specific detectivity by a factor of 28. However, a specific detectivity of 3 × 10 11 Jones is still competitive. Besides that, above 10 kHz, the noise is dominated by the shot component, as indicated by the red line in Figure 5c, further mitigating the effect of the 1/f noise in real applications. Considering the SR measured at 660 nm and a 4 Hz signal, we estimate a SR of 15 A W −1 at 1092 nm. These results obtained at the extraction mode with high photocurrents make this device suitable for CMOS integration.

Conclusion
We present a modern organic photodetector approach using photomultiplication combined with cavity-enhanced charge-transfer to achieve highly efficient narrowband detection capability up to 1600 nm. As one mechanism, very low donor concentrations are used to achieve gain. However, the low CT state energy increases the recombination rate of accumulated holes and thus makes it difficult to achieve EQE values above 100%. Therefore, we insert www.advancedsciencenews.com www.advelectronicmat.de an additional hole-blocking layer to achieve more efficient trapping, triggering high photomultiplication in these devices. The broad CT absorption feature of the absorber system, a D6:C 60 blend, is ideal for detection in the NIR region. Within an optical micro-cavity, the absorption of the CT feature can be enhanced to levels of bulk absorption. Thermal vapor deposition makes it possible to precisely control and change the donor content within the active layer. Locating high-field regions with the aid of TMM simulations allows us to place higher D:A mixing ratios where light is mostly absorbed by CT states, thereby doubling the EQE for optimized devices. For slow light signals, a spectral response of 15 A W −1 with an FWHM of only 18 nm at −9 V is reached at 1092 nm. Such a high spectral response and the need for a reverse bias make the OPDs demonstrated in this work ideal for CMOS readout circuits.

Experimental Section
Device Preparation: Glass substrates with or without prestructured ITO (32 Ω □ −1 , Thin Film Devices) were deposited with organic layers-which had been purified by sublimationusing ultrahigh vacuum (pressure < 10 −7 mbar). Beforehand, they were cleaned with NMP solvent, deionized water and ethanol, followed by O 2 plasma etching for 10 min. All organic materials were purified by at least two times sublimation cleaning. The overlap of the top and bottom contact defines the active area of 6.44 mm 2 . After the evaporation process (Kurt J. Lesker), devices were encapsulated by a glass cover and UV-hardened epoxy. Furthermore, a moisture getter (Dynic Ltd.) was inserted between the device and the glass cover to decrease degradation.
Current-Voltage Characteristics: Illuminated current-voltage characteristics were performed at an intensity of 100 mW cm −2 with a Sun simulator (Solarlight Company Inc., USA) controlled by a Hamamatsu S1337 silicon photodiode. The voltagedependent current under illumination and in dark conditions was measured by a source measurement unit (Keithley SMU 2400).
External Quantum Efficiency (EQE): The monochromatic light of broadband a Xe lamp [Oriel Xe Arc-Lamp Apex Illuminator combined with a Cornerstone 260 1/4m monochromator (Newport, USA)] was chopped at a frequency of 172 Hz and illuminates the device under normal incidence. The signal of a calibrated silicon diode (Hamamatsu 1337 calibrated by Fraunhofer ISE) or a device was measured by a lock-in amplifier (Signal Recovery SR 7265). To avoid edge effects, the device was masked (2.78 mm 2 ). An external bias was applied using a source unit (2NEP-6303, Manson Engineering Industrial Ltd., Hong Kong).
Sensitive External Quantum Efficiency (sEQE): A chopped monochromatic light [140 Hz if not otherwise stated, quartz halogen lamp (50 W) used with a double monochromator (Bentham SHD-300A, QD-Europe)] was shined onto a device or a calibrated silicon (Si) or indium-gallium-arsenide (InGaAs) photodiode under normal incidence. The current was preamplified (DHPCA-100, FEMTO Messtechnik GmbH) and measured with a lock-in amplifier (Stanford Research SR830, USA). The preamplifier was also used to apply a reverse bias to the OPDs.
Spectral Response (SR): A LED (M660L4, Thorlabs) used with an LED driver (BLS-1000-2, Mightex Systems Canada) was driven at various frequencies controlled by a lock-in amplifier (Stanford Research SR830). A reference Si diode (SM05PD3A, Thorlabs) and the device were measured after preamplification (DHPCA-100, FEMTO Messtechnik GmbH) with a lock-in amplifier of the same model.
Power Spectral Density: The noise current was evaluated by connecting a very low noise current-voltage preamplifier (DHPCA-200, FEMTO Messtechnik GmbH) to the device. The internal voltage supply enables bias dependent noise meanwhile avoiding additional noise from another voltage source. The signal was recorded by a high speed, low noise oscilloscope (DPO7354C, Tektronix USA) with a high sampling rate of 10 MS s −1 and high bandwidth to increase the precision of the measurement.
Optical Simulation: The optical simulations were performed by TMM. Optical constants of organic materials were obtained by ellipsometry measurements.

Supporting Information
Supporting Information is available from the Wiley Online Library or from the author.