Giant enhancement of silicon plasmonic SWIR photodetection using nanoscale self-organised metallic films
Christian Frydendahl, Meir Grajower, Jonathan Bar-David, Noa Mazurski, Joseph Shappir, Uriel Levy
aa r X i v : . [ c ond - m a t . m e s - h a ll ] S e p iant enhancement of silicon plasmonic SWIR photode-tection using nanoscale self-organised metallic films Christian Frydendahl , ∗ , Meir Grajower , Jonathan Bar-David , Noa Mazurski , Joseph Shappir ,and Uriel Levy , ∗ Department of Applied Physics, The Faculty of Science, The Center for Nanoscience and Nan-otechnology, The Hebrew University of Jerusalem, Jerusalem, 91904, Israel*[email protected], [email protected]
Many consumer technologies and scientific methods rely on photodetection of infrared light.We report a Schottky photodetector operating below silicon’s band gap energy, through hotcarrier injection from a nanoscale metallic absorber. Our design relies on simple CMOS-compatible ’bottom up’ fabrication of fractally nanostructured aluminium films. Due tothe fractal nature of the nanostructuring, the aluminium films support plasmonically en-hanced absorption over a wide wavelength range. We demonstrate two orders of magnitudeimprovements of responsivity, noise-equivalent-power, and detectivity as compared to bulkmetal, over a broad spectral and angular range. We attribute this to momentum relaxationprocesses from the nanoscale fractal geometry. Specifically, we demonstrate a direct linkbetween quantum efficiency enhancement and structural parameters such as perimeter tosurface ratio. Finally, our devices also function as bulk refractive index sensors. Our ap-proach is a promising candidate for future cost effective and robust short wave infraredphotodetection and sensing applications.Introduction
Photodetection of near infrared (NIR) and short wave infrared (SWIR) light is of interest for a largevariety of practical applications, such as optical communication , spatial proximity sensing , nightvision , as well as molecular spectroscopy - which itself has applications ranging from topics likemedicine and food science , to astrobiology . As such, there is a large desire for cheap, efficient,and scalable technologies for fabrication of NIR and SWIR photodetectors. However, silicon, thestaple of complementary metal-oxide-semiconductor (CMOS) technologies, becomes transparentin the NIR, as the band gap of silicon is roughly 1.1 eV ( ∼ µ m wavelength). Thus silicon is notthe natural choice for photodetection of wavelengths above ∼ µ m.A Schottky junction between a metal and semiconductor can allow for optical absorption of sub-band gap light in the metal, and correspondingly photodetection of photons below the semiconduc-tor band gap, through a process known as internal photo-emission . After a photon is absorbedin the metal, and a hot carrier is generated, it is possible for this carrier to emit across the Schottkyjunction barrier, Φ B8–11 . This injection of hot carriers into the semiconductor can then be collected1s a photocurrent, which is proportional to the incident intensity of light on the junction. The pro-portionality factor, R , is then known as the photodetector’s responsivity. Schottky junctions thusallow semiconductors, like silicon, to indirectly absorb light with energies that exceed Φ B , whichcan be considerably below the band gap energy of the bulk semiconductor. The magnitude of Φ B depends on the choice of metal and semicondutor, as well as the doping of the semiconductor and its surface quality. Typical values of Φ B for metal/silicon junctions, will range from 0.2 to0.8 eV .However, a common issue with Schottky photodetectors is their low responsivities (typically R ∼ µ A/W). Various schemes have been developed to increase the optical absorption efficiency of themetal, and secondly the efficiency of the transfer of the hot carriers to the semiconductor . Suchschemes consist of optical waveguide structures , plasmonic absorbers - such as antenna and grating arrays , or random nanostructures
25, 26 . Engineering of the metal thickness in thejunction , and optimisation of the optical coupling in the device by cavity modes
27, 28 , has alsoproven successful. Recently, the addition of an intermediate layer of graphene between the metaland semiconductor, has also been shown to drastically increase the transfer efficiency of generatedhot carriers
10, 29 . The role of surface roughness in the interface has also recently been studied ,and combined Schottky-photoconductor devices with gain have also been studied .Metal percolation films (also known as semi-continuous films or metal-dielectric composite films)have previously been noted for their high optical absorption . This is typically attributed to thehigh degree of plasmonic field enhancement found in such films . Optical fields can couple tocollective oscillation modes of the conduction electrons at the surface of metallic nanostructures,known as plasmons . Exciting plasmons results in strongly localised electrical fields that can beseveral orders of magnitude higher than that of the initial incident optical field, resulting in strongerlight-matter coupling, and these plasmons can also decay directly into hot carriers . Percolationfilms consist of a fractal network of random metallic clusters, containing gaps and metallic particlesin a wide range of sizes, ranging from sub-nanometer to hundreds of nanometres
38, 40 . This is whatgives these structures their very broad spectral range of optical enhancement
37, 38 .Percolation structures can be achieved from simple metal evaporation processes, as their forma-tion relies on the intrinsic growth patterns of metals on dielectric substrates during deposition .Because of the surface energy between metals and dielectrics, the formation of a thin metal filmwill occur based on a 3D island growth pattern, also known as Volmer-Weber growth . As metalis deposited, the metal adatoms will have a preference to ’stick’ to other metal adatoms, ratherthan the substrate. This results in initial seeding into separated metallic islands in the early stagesof deposition. With increased deposition of metal, the isolated islands will slowly grow laterallyacross the substrate, until they eventually start to connect together and form a dominant ’supercluster’. This point is known as the percolation threshold, and marks the point where the thinmetal film’s properties become dominated by the one unified cluster of metal . For example, thepercolation film becomes electrically conductive at the percolation threshold, as it is now possibleto find an electrically connected path from one side of the film to the other (however windy or2ortuous). Additional metal deposition after this point only serves to close any remaining gapsin the film, and will eventually result in full metal coverage of the substrate, forming a smoothmetal film . Metal percolation films have been a popular topic of research in past decades, dueto their high intrinsic optical field enhancement, and ease of fabrication. They have previouslybeen used for such applications as surface enhanced Raman scattering (SERS) substrates , ex-tinction/absorption enhancement , enhancement of non-linear optical processes in gold, such astwo-photon luminescence
36, 45 and white-light generation
36, 48, 49 . Recently, they have also beenincluded in novel plasmonic laser printing schemes
40, 48, 50, 51 .Here we present a simple method for high-efficiency sub-band gap photodetection in silicon, usinga cheap, CMOS-compatible, and highly scalable fabrication technique, eliminating the need forcostly and complicated nanoscale lithographic processes. By forming a Schottky contact betweena p-doped silicon substrate and fractally nano-patterned aluminium metal percolation films, we areable to guarantee high optical responsivities across a wavelength range of 1304 nm to 1550 nm,with peak responsivity of R ∼ . mA/W and peak internal quantum efficiency of η i ∼ . both at a wavelength of 1304 nm. In addition, we also demonstrate how the absorption in ourdevices is sensitive to the bulk refractive index surrounding the metal films, because of changes tothe optical coupling efficiency. This allows the devices to function as refractive index sensors forliquids placed on top of the metal films, with a direct electrical read-out. This establishes a newarea of Schottky photodetector applications in the chemical- and biosensing domains
16, 52 , beyondjust cost-effective short wave photodetectors.
ResultsSample design:
We have fabricated 5 devices, differing only in the amount of aluminium de-posited for the top electrode forming the Schottky junction to the p-doped silicon substrate. Thesamples have nominal deposited thicknesses of 5.5, 6, 7, and 8 nm. All of these thicknesses areabove the percolation threshold , and thus electrically conductive. A device with a 75 nm bulkaluminium electrode was also fabricated to act as a control sample. All studied diodes have squareactive areas of × µ m . Scanning electron microscope (SEM) images of the nanostructuresgenerated by varying the aluminium deposition, can be seen on Fig. 1.a.In the sample fabrication, we have utilised the technique of local oxidation of silicon (LOCOS) todefine the square diode windows, see Fig. 1.b. This ensures a smooth slope of the SiO sidewalls ,allowing the thin aluminium films to only touch the exposed silicon in a small area, and withoutbreaking the electrical connection of the film from a steep change in height across the sample. Forthe full details of the sample fabrication, see methods and Supplementary Fig. 1. As we illuminatethe samples from below, through the silicon, the top surface of the percolation film electrode is leftfreely accessible to introduce liquid droplets for sensing applications.A simple schematic outline of the internal photo-emission process can be seen in Fig. 1.c. The3 .5nm 6.0nm7.0nm 8.0nm a bc d p-SiSiO Al nano-structured Al P in VA I photo p-Si Φ B E F Al E C E V Photo-excitationTransport Emission ħω g Figure 1: Overview of investigated samples. a) SEM micrographs of the nanostructures achievedat the diode areas for different metal deposition thicknesses. All images at same scale, scale baris 125 nm. b) Schematic side view of the diode geometry, showing the sloped side walls of theinsulating oxide. The illumination scheme is shown with P in . The applied bias, V g , across thejunction and the measured photocurrent, I photo , are also defined. The active diode area is d =40 × µ m . c) The internal photo-emission process of hot holes into silicon. After opticalabsorption in the aluminium of a photon with energy (cid:126) ω > Φ B , there is a chance to emit the hothole into the silicon valence band, resulting in a photocurrent.process consists of 3 steps. First, when a photon of an energy (cid:126) ω is absorbed in the aluminium, ahot electron-hole pair is generated. The hole will then propagate inside the metal with a randommomentum. If the energy of this hole exceeds the Schottky barrier of Φ B , and the momentumof the hole matches to the silicon, it is possible to emit the hole into the p-type silicon’s valenceband. Here it will contribute to a measurable photocurrent. The efficiency of this emission processdefines the overall internal quantum efficiency, η i (and in turn, the responsivity R ), of a Schottkyphotodiode operating in the sub-band gap regime. It has generally been reported, that making themetal electrode in the Schottky junction thinner than the thermalisation length of the carriers in themetal, results in a large increase in responsivity
8, 9, 17 . This is because of two reasons: 1 - there ishigher probability for the carriers to arrive at the metal-semiconductor interface before undergoingthermalisation, and 2 - even if the hot carrier is not emitted when encountering the Schottky barrierfor the first time (due to a momentum mismatch when crossing between the two materials ), itcan be reflected back and forth between the thin metal film’s interfaces. This gives the carriermultiple attempts to cross the Schottky barrier, before it is attenuated inside the metal and loses itsenergy
8, 10 . As the films in our structures are all in the percolation regime, we can safely expect thatthe clusters of metal are generally only a few nanometres thick, i.e. smaller than the thermalisation4ength.The second important aspect in our percolation structures is the effect of plasmonic field en-hancement. Due to the self-similar/fractal nature of the nanostructuring in percolation films, ithas been shown previously that they support plasmonic ’hotspots’ in a wide range of opticalfrequencies
37, 38, 40 . The effect of such plasmonic hotspots is generally to confine any incidentoptical fields close to the surface of the metal, increasing the local field intensity by several ordersof magnitude
37, 48 . By confining and concentrating optical fields down to the nanoscale, it is pos-sible to greatly increase light-matter interaction - such as optical absorption into the aluminiumelectrodes. Metal percolation films have previously been reported to show absorption of around20 to 30% in visible and NIR
33, 35, 45, 54 . The expectation then, is a large increase in the device’sexternal quantum efficiency, η e , when compared to a bulk film. The general relationship betweenthe internal and external efficiencies is η e = Aη i , with A being the absorption of the device . Theresponsivity of a device is then likewise given as : R ( ω ) = Aη i e (cid:126) ω , (1)with e as the elementary charge. From this we see that it is essential to increase the absorption ofa device, as well as its internal quantum efficiency, to insure the highest levels of responsivity.The noise equivalent power (NEP), is the lowest amount of incident power that a detector candetect at a signal to noise ratio (SNR) of 1, at an output bandwidth of ∆ f = 1 Hz:NEP ( λ ) = √ eI R ( λ ) , (2)With I as the dark/leakage current. Likewise, the specific detectivity of a detector, D ∗ , is thendefined by the detector’s area, A det , and the output bandwidth (integration time), as: D ∗ ( λ ) = √ ∆ f A det NEP ( λ ) . (3)The NEP of a detector thus helps define its sensitivity more accurately than the responsitivity alone,and the specific detectivity is an even more precise figure of merit for a detector, as it accounts forthe total active area of the device, which is likely to affect the dark current and noise. Optical responsivity and quantum efficiency:
We investigated the spectral dependence of ourdevices’ responsivities for wavelengths of 1304 and 1550 nm (see methods), the results can be seenon Fig. 2. The devices are all illuminated from below through the silicon, as indicated in Fig. 1.b.An example IV -measurement from a diode with a 5.5 nm Al deposition can be seen on Fig. 2.a.By recording the generated photocurrent at different incident optical powers, we can determine theresponsivity from a simple linear regression, see Fig. 2.b.5 µ A ] a bc fit: R Figure 2: Responsivity characterisation. a) Example of IV -characteristics of a 5.5 nm thick filmdiode, with and without various powers of 1304 nm illumination. The dotted vertical line marks the0.3 V reverse bias used for the responsivity calculations in b-c . b) Example of the linear regressionfor responsivity determination of the 5.5 nm device in a , at 0.3 V reverse bias. c) Responsivityvs. Al deposition thickness, for two different wavelengths of excitation. All responsivities arecalculated for V g = 0 . V. A bulk film (75 nm) has also been included for comparison. Error barsare multiplied by 6 for better visualisation.From Fig. 2.c, we find generally high responsivities in the range of R ∼ − . mA/W for all 4investigated percolation film thicknesses. This is 2 orders of magnitude higher than the responsivityof our bulk control sample (a 75 nm thick film), which is of the order R bulk ∼ . mA/W.There is no appreciable increase in dark current or noise from our percolation devices when com-pared to the bulk device. For the 5.5 nm devices, we get a NEP of ∼ . · − W/ √ Hz, and a D ∗ of ∼ . · Jones for 1304 nm light, when operating at a 0.3 V reverse bias. For the full NEPand D ∗ results for all devices and wavelengths, see Supplementary Fig. 2.We find no strong dependence of the responsivity on the polarisation of the incident light. This isalso to be expected, as the nanostructuring of the percolation films is random/isotropic. Althoughthe exact spatial distribution of plasmonic hotspots will be different for different polarisations oflight , the average amount and intensity of the hotspots will be similar for two mutually perpen-dicular polarisations. Due to the quite large refractive index of silicon ( n Si ∼ ◦ and ◦ of free-space angles of incidence, seeSupplementary Fig. 3. 6 ulk refractive index sensing: Plasmonic systems have previously been exploited for sensingpurposes to great success in the past . However, such sensing devices are typically reliant on avery complex optical read-out of the changes in the plasmonic resonance. Here, we can directlymonitor the absorption changes in the aluminium percolation film from the measured photocurrent.The results of our bulk sensing tests can be seen on Fig. 3.a and b. cab P in n Si n eff,2 n liquid n Si n eff,1 n air P in Figure 3: Liquid sensing experiments. a) Photocurrent of a 6 nm film device in air and withdifferent liquid droplets added on top, as a function of the incident optical power. Dotted lines showfits for extracting R . b) Responsivities of a as a function of liquid refractive index, and calculatedthin film absorption (see main text). Both a and b were recorded with 1304 nm excitation. c) Fresnel equation thin film model. The effective index for the metal film is calculated from themetal filling fraction and the surrounding refractive index.We tested a 6 nm device using two different liquids, a 95% isopropyl alcohol (IPA) solution ( n IPA ∼ . ) and Nikon microscope immersion oil Type A ( n oil ∼ . ). From Fig. 3.a and b, we see thatthe responsivity of the device goes down when a higher index liquid is placed on top.We attribute the loss in responsivity to the change in effective index of the metal film. As the metalis porous the liquid can fill out the voids in the film, and this changes the coupling efficiency tothe metallic layer . Likewise, any Fabry-P´erot cavity effect will be lessened as the higher indexliquids act as an anti-reflection coating, again lowering the total coupling efficiency to the metallicabsorber. We can calculate the absorption in the metal film layer as a thin coating on the silicon,using the complex Fresnel equations (model shown in Fig 3.c). Because the metal film is porous,we use the Maxwell-Garnett equation to calculate the effective refractive index between the mixingof the metal and the bulk dielectric. Each added liquid thus changes the effective index of themetallic layer, and results in a change in total absorption, seen in Fig. 3.b. See the supplementarymaterials for the full calculation details. It should be noted however, that the true absorption inthe metal films is higher, as neither the Fresnel nor the Maxwell-Garnett equations consider thepossibility of plasmonic resonances. The result presented here is thus only an estimate, but the7elative shifts in absorption predicted by the model, and the overall trend should still apply to ourcase.An alternative explanation could be additional loss mechanisms opening up from the increasedspill-out of the electron density of the metal, with increasing surrounding bulk index . Althoughour devices appear to be in the Drude-loss dominated regime, due to the relatively low refrac-tive indices tested, and from the fact that we are working with infrared resonance frequencies ,the extreme nanoscale dimensions of the films might also make them more sensitive to quantumeffects .Because our percolation film electrode structure hosts a wide continuum of resonances, the sensingis not based on the redshifting of resonances from adding a higher index material on top of the film,as the density of resonances is expected to stay similar within a narrow shift of wavelengths . Fractal properties of the film geometries:
We investigate the features of the different film ge-ometries by using image analysis of SEM images recorded of each of the film geometries. Becauseall of our samples are above the percolation threshold, we cannot easily study the properties of themetal itself, as there is only one connected component of metal. We choose instead to study the in-verse geometry, and investigate the absences of metal in the films, i.e. the voids. The full results ofthis analysis can be seen in the supplementary materials (Supplementary Figs. 4-6), but the overallconclusions are as follows: a b
MetalVoid
Figure 4: Momentum relaxation. a) Total perimeter of voids normalised to the area investigated(dotted line), and internal quantum effeciencies, η i for 1304 nm (points), for the different filmgeometries. The inset shows a binary image of a 5.5 nm film as compared to an 8.0 nm film. b) Schematic of how the localisation of carriers in the nanoscale features in the films results inmomentum relaxation, easing the transmission process of carriers from the metal to the silicon.The thinner metal depositions results in fewer, but significantly bigger and more complex voidshapes. This is shown in the supplementary materials by the void clusters of the thin films having8ignificantly larger correlation lengths, ξ , average areas, S , and higher fractal Hausdorff dimen-sions, D . As shown in Fig. 4, the thinner structures are also significantly more dominated by theperimeter of voids per the total film area (i.e. the images contains more metallic edges). Thismeans that corresponding metallic clusters are more separated into thin filaments, rather than acontinuous film with small isolated voids as in the thicker depositions (see Fig. 4.a inset).We determine the internal quantum efficiency of our devices from absorption measurements per-formed with an integrating sphere, see Supplementary Fig. 7. From these measurements, we seethat absorption goes from roughly 0.45 to 0.55 from the thinnest to the thickest films. This is likelydue to enhanced coupling efficiency to the thicker films, from the Fabry-P´erot effect discussed inthe section above. Despite the increasing absorption, we see that the responsivity drops for thethicker films in Fig. 2. To explain this, we calculate the internal quantum efficiencies of the filmsusing the measured responsivities and absorption values for 1304 nm, and follow eqn. 1. Fig. 4.ashows the found internal quantum efficiencies. The peak value is of roughly 1% for the 5.5 nmfilm devices. However, what is extremely interesting, is how there seems to an almost direct rela-tionship between the change in quantum efficiency and the total void perimeter ratio of the films(i.e. how much of a film image is made up of metallic edges). This is also mirrored for the case of1550 nm, shown in Supplementary Fig. 8Films with a higher total void perimeter ratio result in more isolation of the metal into thin filamentsand nanoclusters. This isolation enables a higher degree of hot carrier momentum relaxation, asthe carriers are more likely to encounter a metal/vacuum interface and elastically scatter off of itto change their momentum, before the carriers can thermalise. See Fig. 4.b for a schematic of theprocess. Schottky barrier measurements:
Finally, we have characterised the Schottky barrier height inour different devices (see methods). The results can be seen on Fig. 5.Fig. 5.a shows an example of the temperature dependent IV -characteristics of a 5.5 nm devicein reverse bias. The dotted lines indicate fits to the linear part of the IV -curves, using the ex-pression: I = R − s V g + I s , where R s is a serial resistance and I s is the saturation current. Fromthermionic emission theory, we get the saturation current’s evolution with temperature, as I s ∝ exp ( − Φ B /k B T ) . Fig. 5.b shows how Φ B can be extracted by fitting such an exponential expres-sion to the I s values found in Fig. 5.a. Finally, Fig. 5.c shows the determined Schottky barrierheights found for our different device deposition thicknesses. We find generally similar values forthe percolation samples, as compared to the bulk sample, with only the 5.5 nm sample having aslightly lower value of Φ B . 9 .02.03.0 a bc Figure 5: Schottky barrier characterisation. a) Temperature dependent IV-curves for a 5.5 nmdevice in reverse bias. The dotted lines indicate linear regressions for extracting the saturationcurrent, I s , which is used for determining the Schottky barrier (see main text/methods). b) Expo-nential fit for determining the Schottky barrier height of the 5.5 nm device highlighted in a , usingthe extracted values of I s . c) Comparison of Φ B determined for devices of different depositionthicknesses. Discussion
As mentioned previously, the high responsivities observed in the percolation samples versus thebulk film control sample, can likely be explained by two aspects of the percolation geometry.First, the overall thinness of the percolation metal layers in the Schottky junctions helps ensurea higher probability of hole emission (increasing the device’s internal quantum efficiency, η i ).This by itself is due to two major mechanisms: 1 - lower probability of thermalisation due to thenanoscale distance between the location of carrier generation and the Schottky interface, and 2 - thelocalisation into small metallic grains of ∼
50 nm or smaller (see Fig. 1.a) provides an additionalavenue of momentum relaxation, which is lost as the feature sizes increase for the thicker depositedfilms. Finally, due to plasmonic field enhancement and resonant coupling, the metallic part of thejunction absorbs a greater deal of incident light.We see otherwise that responsivity decreases for increasing metal deposition for the percolationstructures. This is most likely the result of the metal clusters becoming physically thicker for in-creasing deposition thickness, lowering the probability of hot carrier emission due to increasedtransport time/distance, and the nanostructing that contributes to momentum relaxation is de-creased for the thicker depositions. 10or sensing, we observe a noticeable shift in the overall device responsivity when either an IPAor index matching oil droplet is placed on top of the active diode area, fully embedding the metalstructure in the higher index material. Importantly, we also see the diodes recover back to theirintrinsic responsivity after the removal of the liquid, demonstrating that the responsivity change isindeed not caused by a chemical change to the metal electrodes. We explain the drop in responsiv-ity as being from a general lowering of optical coupling efficiency to the metal/liquid layer becauseof a change in the film’s effective index, and from lowering the efficiency of any Fabry-P´erot cavityeffect in the samples.It is worth briefly highlighting how sensing devices such as these could be affected by quantumeffects
58, 59 . It has recently been demonstrated how a high bulk refractive index surrounding plas-monic structures will increase the electron density spill-out, causing additional damping mecha-nisms of the plasmon resonances to become relevant . We believe that the devices investigatedhere are generally in the classical regime due to the low refractive indices and the low plasmon res-onance energies investigated, and as such the plasmonic loss is dominated by phonon/Drude-loss.However, quantum effects such as these could play an important role in future plasmonic indexsensing devices, in particular if operating within visible wavelengths. Conclusion
We have demonstrated a simple, cheap, scalable, and CMOS-compatible fabrication techniquefor Schottky diodes with sub-band gap photodetection in silicon. We utilise self-organsied fractalmetasurfaces, known as metal percolation films, to achieve plasmonic enhancement of hot car-rier generation across a broad spectral regime. In addition, we have shown how such hot carrierphotodetectors can be utilised for bulk refractive index sensing.Our devices show high responsivities, with a peak value of ∼ ∼ · − W/ √ Hz and a corresponding D ∗ of ∼ · Jones, both for 1304 nm and 0.3 Vreverse bias operation. Such numbers are in the same order of magnitude as some of the commer-cial photodetectors, e.g. those made of PbS. Our nanostructured devices show internal quantumefficiencies of η i ∼ . , which we attribute to momentum relaxation processes made possible bythe random structuring and overall thinness of the percolation film geometries. We see generallythat the responsivity of the devices decreases with the deposition thickness of the percolation layer,despite the increase in optical absorption. In general we see Schottky barrier heights of ∼ ∼ ◦ . This makes them a potential candidatefor robust and reliable NIR- and SWIR photodetection in many practical applications.11n this work, we have not investigated any ways to optimise the absorption in the aluminium,besides varying the deposition thickness. Future investigations could look into utilising an anti-reflective coating on the illuminated silicon interface, to further increase the transmission of lightinto the metal. For purely optical photodetection purposes, a thick metallic mirror could be addedabove the percolation layer, to create a Salisbury screen effect, or even allow plasmonic couplingto the mirror. Such a structure, using gold films, has recently been demonstrated to allow forwideband near-perfect absorption in the visible and for contrast enhancement in plasmonic colourprinting . Again, we emphasise the relative simplicity of the nanofabrication of our devices,requiring only oxidation of the top silicon surface to define the diode areas, and metal evaporationthrough a simple mechanical mask while relying purely on CMOS compatible techniques andmaterials. The fabrication is thus suitable for easily making complex optical photodiode arraystructures, for applications like NIR-cameras, beam positioning sensors , and as demonstrated,could also be used for cheap liquid/humidity sensors. MethodsSample fabrication:
The samples were fabricated from a two-step UV-lithography process. Formore details please see Supplementary Fig. 1. 500 µ m thick double side polished 4” wafers ofp-doped silicon ( ρ ∼ Ω cm) were used. First, a 40 nm thermal oxide was grown. After this,150 nm nitride was grown on the top of the wafer, using plasma-enhanced chemical vapor deposi-tion (PECVD). Next, a pattern for the active areas of the diodes were defined with UV-lithography,and the unmasked nitride was etched with reactive ion etching (RIE). Using local oxidation ofsilicon (LOCOS), 250 nm of oxide was grown next to the diode areas. After the LOCOS, theremaining nitride was etched in 180 ◦ C phosphoric acid.Then, the topside of the wafer is covered with photoresist, and the wafer was baked at 120 ◦ C for2 minutes. Then the wafer was submerged in buffer hydrofluoric acid (HF) to remove the oxidefrom the back of the wafer. Then the topside resist is removed by solvent and piranha cleaning.Next, the wafer is turned over and the any native oxide on the back is removed with HF. Immedi-ately afer, a 100 nm aluminium film is deposited on the back of the wafer. Using UV-lithography,a pattern for the aluminium Ohmic contact is defined, and the uncovered aluminium is removedwith chemical etching. Leftover resist is removed, and the wafer is placed in a 460 ◦ C oven with / N /H atmosphere to make the back Al contact Ohmic by alloying.Before depositing the thin metal films for the Schottky electrode, the wafer was diced into chips.Each sample is then cleaned, and quickly rinsed in 1:10 HF:H O, to strip any native oxide at theactive diode areas. After this, thin metallic films are deposited with electron-beam evaporationthrough a mechanical mask. The films are deposited at a rate of 0.6 ˚A/s, in a vacuum chamberpressure of ∼ − mbar. 12 esponsivity characterisation: The samples were characterised using a 1550 nm Thorlabs fibercoupled laser (model S1FC1550), and a 1304 nm laser diode powered by a Thorlabs ITC510Benchtop Laser Diode and Temperature Controller.A Keysight B2901A Precision Source/Measure Unit was used for recording IV -curves. The sam-ples were electrically contacted to the Ohmic back contact with a piece of carbon tape, and amicro-mechanical probe was used to make contact to individual diodes on the top of the sample,by mechanically touching the surrounding metal films.Light from the corresponding light source was shone through the window of the Ohmic back con-tact, using an 8 mm pigtail-style fiber lens from Oz Optics. Using a 3D stage, the light was focusedon the individual diode areas, with a spotsize of roughly ∼ µ m diameter. IV -characteristics ofthe diodes were then recorded from − . V to . V, for different wavelengths and powers of illu-mination. The output power from the fiber lens was measured, and we measured the reflectivityof the silicon substrate to roughly R = 0 . . The reported incident powers have been corrected as60% of the power measured from the fiber lens, to account for the reflection from the bottom ofthe silicon.From these measurements, the responsivities were found by fitting the expression: I photo ( V g ) = R ( V g ) P in + I ( V g ) , where I photo ( V g ) is the measured photocurrent for a certain reverse bias of the diode, V g . R ( V g ) is the responsivity for the bias, P in in the input optical power, and I ( V g ) is the measured darkcurrent ( P in = 0 W) for the same bias. A reverse bias of V g = 0 . V was used for all responsivitycalculations.
Index sensing:
To measure the effect of different refractive indices on top of the percolation films,the samples were mounted as in the general optical characterisation experiments. The diodes werethen characterised for 5 different optical powers using a 1304 nm laser diode powered by a ThorlabsITC510 Benchtop Laser Diode and Temperature Controller. After characterising the dry diode, aliquid droplet of either 95% IPA or Nikon immersion oil type A was then added, and the diode wasmeasured again for the same 5 optical powers. After each liquid test, the diode was dismountedand washed with acetone and left to dry, and then remounted and remeasured.
Schottky barrier measurements:
The magnitude of the Schottky barriers in the diodes weremeasured by heating the samples to different temperatures with an Ohmic heater, while recording IV -curves from − . V to . V for the different temperatures. From these measurements, thesaturation currents from reverse bias, I s ( T ) , was found at the different temperatures by fitting thelinear parts of the IV -curves ( . V ≤ V g ≤ . V) with: I = R − s V g + I s , R s as a serial resistance. The different saturation currents were fitted to the expression : I s ( T ) ∝ exp ( − Φ B /k B T ) , where Φ B is the Schottky barrier energy, k B is Boltzmann’s constant, and T is the temperature. Data availability:
Acknowledgments:
The authors would like to thank Maurice Saidian and Evgenia Blayvas, as well as therest of the Harvey M. Kruger Family Center of Nanoscience and Technology for assistance in the samplefabrication and SEM characterisation.This work was financially supported by the Israel-USA Binational science foundation (BSF).C.F. is supported by the Carlsberg Foundation as an Internationalisation Fellow.
Author contributions:
The sample geometry was designed by all authors in collaboration. The sampleswere fabricated by N.M. IV -measurements, Schottky barrier measurements, and sensing experiments weredone by C.F. with M.G assisting. Absorption measurements were done by J.B.D. and C.F. Image and fractalanalysis was done by C.F. The manuscript was written and drafted by C.F., with M.G. and U.L. contributing,and approved by all authors. J.S. and U.L. supervised the entire project. Conflict of interests:
The authors declare no competing financial interests, or conflicts of interest.
Corresponding authors:
Requests for research materials (data/supplementary information) or generalcorrespondence should be addressed to Christian Frydendahl ([email protected]) or UrielLevy ([email protected]).
References
1. Keiser, G. Optical fiber communications. Wiley Encyclopedia of Telecommunications (2003).2. Blais, F. Review of 20 years of range sensor development. J. Electron. Imaging , 231–244(2004).3. Chrzanowski, K. Review of night vision technology. Opto-Electron. Rev. , 153–181 (2013).4. Parker, F. Applications of infrared spectroscopy in biochemistry, biology, and medicine(Springer Science & Business Media, 2012).5. Osborne, B. G. & Fearn, T. Near-infrared spectroscopy in food analysis. Enc. Anal. Chem. ,4069–4082 (2000). 14. Schwieterman, E. W. et al. Exoplanet biosignatures: A review of remotely detectable signs oflife. Astrobiology (2018).7. Mishra, U. & Singh, J. Semiconductor device physics and design (Springer Science & Busi-ness Media, 2007).8. Scales, C. & Berini, P. Thin-film Schottky barrier photodetector models.IEEE J. Quantum Electron. , 633–643 (2010).9. Berini, P. Surface plasmon enhanced Schottky detectors. In Quantum Plasmonics, 191–209(Springer, 2017).10. Levy, U., Grajower, M., Goncalves, P., Mortensen, N. A. & Khurgin, J. B. Plasmonic sili-con Schottky photodetectors: The physics behind graphene enhanced internal photoemission.APL Photonics , 026103 (2017).11. Grajower, M., Levy, U. & Khurgin, J. B. The role of surface roughness in plasmonic-assistedinternal photoemission Schottky photodetectors. ACS Photonics , 4030–4036 (2018).12. Akbari, A. & Berini, P. Schottky contact surface-plasmon detector integrated with an asym-metric metal stripe waveguide. Appl. Phys. Lett. , 021104 (2009).13. Goykhman, I., Desiatov, B., Khurgin, J., Shappir, J. & Levy, U. Locally oxidized siliconsurface-plasmon Schottky detector for telecom regime. Nano Lett. , 2219–2224 (2011).14. Goykhman, I., Desiatov, B., Khurgin, J., Shappir, J. & Levy, U. Waveguide based com-pact silicon Schottky photodetector with enhanced responsivity in the telecom spectral band.Opt. Express , 28594–28602 (2012).15. Knight, M. W., Sobhani, H., Nordlander, P. & Halas, N. J. Photodetection with active opticalantennas. Science , 702–704 (2011).16. Alavirad, M., Mousavi, S. S., Roy, L. & Berini, P. Schottky-contact plasmonic dipole rectennaconcept for biosensing. Opt. Express , 4328–4347 (2013).17. Li, W. & Valentine, J. Metamaterial perfect absorber based hot electron photodetection.Nano Lett. , 3510–3514 (2014).18. Li, W. et al. Circularly polarized light detection with hot electrons in chiral plasmonic meta-materials. Nat. Commun. , 8379 (2015).19. Desiatov, B. et al. Plasmonic enhanced silicon pyramids for internal photoemission Schottkydetectors in the near-infrared regime. Optica , 335–338 (2015).20. Brueck, S., Diadiuk, V., Jones, T. & Lenth, W. Enhanced quantum efficiency internal photoe-mission detectors by grating coupling to surface plasma waves. Appl. Phys. Lett. , 915–917(1985). 151. Sobhani, A. et al. Narrowband photodetection in the near-infrared with a plasmon-induced hotelectron device. Nat. Commun. , 1643 (2013).22. Giugni, A. et al. Hot-electron nanoscopy using adiabatic compression of surface plasmons.Nat. Nanotechnol. , 845 (2013).23. Alavirad, M., Olivieri, A., Roy, L. & Berini, P. High-responsivity sub-bandgap hot-hole plas-monic Schottky detectors. Opt. Express , 22544–22554 (2016).24. Lin, K.-T., Chen, H.-L., Lai, Y.-S. & Yu, C.-C. Silicon-based broadband antenna for highresponsivity and polarization-insensitive photodetection at telecommunication wavelengths.Nat. Commun. , 3288 (2014).25. Lee, Y. K. et al. Surface plasmon-driven hot electron flow probed with metal-semiconductornanodiodes. Nano Lett. , 4251–4255 (2011).26. Wen, L., Chen, Y., Liang, L. & Chen, Q. Hot electron harvesting via photoelectric ejection andphotothermal heat relaxation in hotspots-enriched plasmonic/photonic disordered nanocom-posites. ACS Photonics , 581–591 (2017).27. Casalino, M., Coppola, G., Iodice, M., Rendina, I. & Sirleto, L. Critically coupled sil-icon fabry-perot photodetectors based on the internal photoemission effect at 1550 nm.Opt. Express , 12599–12609 (2012).28. Kosonocky, W. F., Shallcross, F. V., Villani, T. S. & Groppe, J. V. 160 ×
244 element ptsiSchottky-barrier ir-ccd image sensor. IEEE Trans. Electron. Devices , 1564–1573 (1985).29. Goykhman, I. et al. On-chip integrated, silicon–graphene plasmonic Schottky photodetectorwith high responsivity and avalanche photogain. Nano Lett. , 3005–3013 (2016).30. Tanzid, M. et al. Combining plasmonic hot carrier generation with free carrier absorption forhigh-performance near-infrared silicon-based photodetection. ACS Photonics , 3472–3477(2018).31. Dobierzewska-Mozrzymas, E., Radosz, A. & Biega ´nski, P. Optical properties of granularaluminum films. Appl. Opt. , 727–731 (1985).32. Yagil, Y. & Deutscher, G. Transmittance of thin metal films near the percolation threshold.Thin Solid Films , 465–471 (1987).33. Kunz, M., Niklasson, G. & Granqvist, C. Optical and electrical properties of sputter-depositedal films close to the percolation threshold. J. Appl. Phys. , 3740–3742 (1988).34. Yagil, Y., Gadenne, P., Julien, C. & Deutscher, G. Optical properties of thin semicontinuousgold films over a wavelength range of 2.5 to 500 µ m. Phys. Rev. B , 2503 (1992).35. Berthier, S., Peiro, J., Fagnent, S. & Gadenne, P. Infrared absorption of granular metal filmsin the percolation range. Physica A , 1–5 (1997).166. Ducourtieux, S. et al. Percolation and fractal composites: Optical studies.J. Nonlinear Opt. Phys. Mater. , 105–116 (2000).37. Genov, D. A., Sarychev, A. K. & Shalaev, V. M. Plasmon localization and local field distribu-tion in metal-dielectric films. Phys. Rev. E , 056611 (2003).38. Shalaev, V. M. Nonlinear optics of random media: fractal composites and metal-dielectric films,vol. 158 (Springer, 2007).39. Schuller, J. A. et al. Plasmonics for extreme light concentration and manipulation. Nat. Mater. , 193 (2010).40. Frydendahl, C. et al. Optical reconfiguration and polarization control in semi-continuous goldfilms close to the percolation threshold. Nanoscale , 12014–12024 (2017).41. Malureanu, R. & Lavrinenko, A. Ultra-thin films for plasmonics: a technology overview.Nanotechnol. Rev. , 259–275 (2015).42. Gadenne, P., Gagnot, D. & Masson, M. Surface enhanced resonant Raman scattering inducedby silver thin films close to the percolation threshold. Physica A , 161–165 (1997).43. Drachev, V. P. et al. Adaptive silver films for surface-enhanced Raman spectroscopy ofbiomolecules. J. Raman Spectrosc. , 648–656 (2005).44. Perumal, J., Kong, K. V., Dinish, U., Bakker, R. M. & Olivo, M. Design and fabrication ofrandom silver films as substrate for SERS based nano-stress sensing of proteins. RSC Adv. ,12995–13000 (2014).45. Novikov, S. M. et al. Enhancement of two-photon photoluminescence and SERS for low-coverage gold films. Opt. Express , 16743–16751 (2016).46. Wallace, G. Q. & Lagugn´e-Labarthet, F. Advancements in fractal plasmonics: Structures,optical properties, and applications. Analyst , 13–30 (2019).47. Galanty, M. et al. Extinction enhancement from a self-assembled quantum dots monolayerusing a simple thin films process. J. Phys. Chem. C , 24991–24995 (2015).48. Novikov, S. M. et al. White light generation and anisotropic damage in gold films near perco-lation threshold. ACS Photonics , 1207–1215 (2017).49. Chen, J. et al. Evidence of high-order nonlinearities in supercontinuum white-light generationfrom a gold nanofilm. ACS Photonics , 1927–1932 (2018).50. Zheng, Y. et al. Hot luminescence from gold nanoflowers and its application in high-densityoptical data storage. Opt. Express , 9262–9275 (2017).51. Roberts, A. S. et al. Laser writing of bright colors on near-percolation plasmonic reflectorarrays. ACS Nano (2018). 172. Tsukagoshi, T. et al. Compact surface plasmon resonance system with Au/Si Schottky barrier.Sensors , 399 (2018).53. Novotny, M., Bulir, J., Lancok, J., Pokorny, P. & Bodnar, M. In-situ monitoring of the growthof nanostructured aluminum thin film. J. Nanophotonics , 051503 (2011).54. Pribil, G., Johs, B. & Ianno, N. Dielectric function of thin metal films by combined insitu transmission ellipsometry and intensity measurements. Thin Solid Films , 443–449(2004).55. Stewart, M. E. et al. Nanostructured plasmonic sensors. Chem. Rev. , 494–521 (2008).56. Krayer, L. J., Tennyson, E. M., Leite, M. S. & Munday, J. N. Near-IR imaging based on hotcarrier generation in nanometer-scale optical coatings. ACS Photonics , 306–311 (2017).57. Liu, Z. et al. Automatically acquired broadband plasmonic-metamaterial black absorber dur-ing the metallic film-formation. ACS Appl. Mater. Interfaces , 4962–4968 (2015).58. Christensen, T., Yan, W., Jauho, A.-P., Soljaˇci´c, M. & Mortensen, N. A. Quantum correctionsin nanoplasmonics: shape, scale, and material. Phys. Rev. Lett. , 157402 (2017).59. Jin, D. et al. Quantum-spillover-enhanced surface-plasmonic absorption at the interface ofsilver and high-index dielectrics. Phys. Rev. Lett. , 193901 (2015).60. Grajower, M., Desiatov, B., Mazurski, N. & Levy, U. Integrated on-chip silicon plasmonicfour quadrant detector for near infrared light. Appl. Phys. Lett.113