Site-specific chemical doping reveals electron atmospheres at the surfaces of organic semiconductor crystals

Chemical doping controls the electronic properties of organic semiconductors, but so far, doping protocols and mechanisms are less developed than in conventional semiconductors. Here we describe a unique, site-specific, n-type surface doping mechanism for single crystals of two benchmark organic semiconductors that produces dramatic improvement in electron transport and provides unprecedented evidence for doping-induced space charge. The surface doping chemistry specifically targets crystallographic step edges, which are known electron traps, simultaneously passivating the traps and releasing itinerant electrons. The effect on electron transport is profound: field-effect electron mobility increases by as much as a factor of ten, and its temperature-dependent behaviour switches from thermally activated to band-like. Our findings suggest new site-specific strategies to dope organic semiconductors that differ from the conventional redox chemistry of randomly distributed substitutional impurities. Critically, they also verify the presence of doping-induced electron atmospheres, confirming long-standing expectations for organic systems from conventional solid-state theory. Organic semiconductor crystals can be selectively doped at the crystallographic step edges, deactivating shallow traps and recovering band-like transport. The space charge induced by chemical doping is observed by scanning Kelvin probe microscopy.

C hemical doping plays a central role in the science and applications of organic semiconductor materials [1][2][3][4][5][6] . Decades ago, the discovery that crystals and films of pi-conjugated molecules could be made highly conducting, and even metallic, by chemical doping (for example, conducting polymers) sparked world-wide interest in these materials and marked a critical turning point in the field of organic electronics [7][8][9][10][11][12][13] . Today, p-and n-type doping of organic semiconductor films is used industrially to lower the operating voltages of organic light-emitting diode displays 1,14,15 . Still, many structural and electronic aspects of chemical doping in organic semiconductors remain poorly understood, particularly for crystalline or polycrystalline systems where strong anisotropic intermolecular interactions, crystalline defects and microstructure can be decisive in determining the doping mechanism and ionization efficiency [16][17][18][19][20][21] .
Here we report two discoveries related to chemical doping of organic semiconductors. First, we have found that crystallographic step edges on the surfaces of specific organic semiconductor crystals can be selectively n-doped. This site-specific doping eliminates shallow electron traps at the step edges and results in dramatic recovery of band-like transport properties for crystals that, before doping, had low field-effect electron mobilities and strongly activated (trap-limited) transport. This intriguing result indicates site-specific doping may be a productive new strategy for improved material performance in organic electronics. Second, we have discovered that doping-induced electron atmospheres along the step edges can be visualized by scanning Kelvin probe microscopy (SKPM). The SKPM images represent direct detection of space charge associated with chemical doping in an organic semiconductor and provide critical verification of microscopic doping effects in organic systems that match expectations from classical inorganic semiconductor physics.
Our study focuses on crystals of Cl 2 -NDI (refs. 22,23 ) (space group P1) and PDIF-CN 2 (refs. 24,25 ) (space group P1), respectively, Fig.  1, well-known semiconducting materials that exhibit outstanding n-channel performance in field-effect transistors (FETs). Exposure of the (001) surfaces of both crystals to an N-silane vapour ((3-(2-aminoethylamino)propyl)trimethoxysilane, Fig.  1a) results in n-type doping that is evident in FET measurements, as described below. Our motivation to use this amine as a dopant was based in part on the known redox reactivity of Cl 2 -NDI and PDIF-CN 2 with amines, and in part on trial-and-error experiments with a variety of different amines (Supplementary Figs. 1-3) [26][27][28][29][30] . Spectroelectrochemistry and ultraviolet-visible light (UV-vis) titration studies confirm that the N-silane in Fig. 1a reduces Cl 2 -NDI and PDIF-CN 2 in solution (Supplementary Figs. 4 and 5), and we hypothesized that spontaneous electron transfer from the amines to the crystals could produce n-type surface doping.
To test this hypothesis, we grew lath-like crystals of Cl 2 -NDI and PDIF-CN 2 with thicknesses ranging from 1 to 50 μm by physical vapour transport and laminated the crystals onto Au-coated polydimethylsiloxane (PDMS) stamps to make four-terminal 'air-gap' FETs ( Fig. 1a) 25,31 . In these FETs, the empty space, typically air or vacuum, between the recessed gate and the crystal surface forms the gate dielectric. Such devices have the advantage that the gated semiconductor surface is pristine. The transport characteristics of the completed crystal FETs were then measured under vacuum. Subsequently, doping of the same crystals was accomplished by exposing the FETs to the N-silane vapour using the quartz hot wall apparatus shown schematically in Fig. 1a. Contact of the N-silane vapour with the crystal face forming the FET channel was possible because of the air-gap FET design. Figure 2a shows representative FET drain current versus gate voltage (I D -V G ) characteristics for a relatively thick (47 μm) Cl 2 -NDI crystal before and after exposure to 100 μl of N-silane (for all FET results the source is grounded, V S = 0). The undoped, as-grown crystal exhibits n-type field-effect conduction as expected (turning ON as V G becomes more positive), and after exposure to the N-silane there is a clear negative shift in the onset voltage, as well as improvement of the maximum conductance in the ON state. Note for example that at V G = 0 V, I D is five orders of magnitude greater after doping. The drain current-drain voltage (I D -V D ) characteristics also reveal substantial differences, Fig. 2b. After doping, the crystal does not show characteristic current saturation, and the currents at the same V G and V D values are more than an order of magnitude higher than for the same crystal before doping. The transport results in Fig. 2a,b clearly indicate n-type doping for the thick Cl 2 -NDI crystal. However, thin (2.9 μm) Cl 2 -NDI crystal results are much different (Supplementary Fig. 6). Exposure of thin crystals to the N-silane does not produce a large onset voltage shift, nor does it result in any notable change in maximum ON current. There is only a change in the OFF current at negative V G values. The results in Fig. 2a and Supplementary Fig. 6 are representative of many experiments on thick and thin crystals. Evidently, crystal thickness somehow plays a vital role in the doping effect. We carried out the same doping experiments with PDIF-CN 2 crystals and found essentially the same results ( Supplementary Fig. 7).
We have shown previously that the crystallographic step-edge density on the (001) surface of as-grown Cl 2 -NDI and PDIF-CN 2 crystals is highly correlated with crystal thickness 23 ; others have shown similar results for rubrene crystals 32 . Furthermore, we have demonstrated that the step edges (Fig. 1c) serve as electron traps in FET measurements on Cl 2 -NDI and PDIF-CN 2 crystals 23 . It therefore seemed reasonable that the doping effects were actually directly related to step density, rather than crystal thickness. Figure 1d shows atomic force microscopy (AFM) images of one-unit-cell-tall steps on both the thick and thin crystals of Cl 2 -NDI (note the difference in scale bars). Figure 2c displays ON-state sheet conductance σ s at V G = + 60 V for a large number of as-grown (93) and doped (64) Cl 2 -NDI and PDIF-CN 2 crystals as a function of step density. For as-grown Cl 2 -NDI and PDI-CN 2 crystals, σ s decreases strongly with step density. However, the dependence is opposite for doped crystals. For doped crystals, σ s increases with step density. These trends do not reflect changes in contact resistance on doping as the contact resistance in these devices is small relative to the channel resistance ( Supplementary Fig. 8). The FET threshold voltages V T shown in Fig. 2d also trend oppositely for as-grown versus doped samples. For as-grown crystals, V T becomes more positive with increasing step density, consistent with step-edge trapping that we have reported previously 23 . For doped crystals, V T is increasingly negative as step density increases, that is, turning the FET channel OFF requires strongly negative gate voltages, consistent with increased n-doping as step density increases. The data in Fig. 2c,d are indeed consistent with the critical role of step edges in the doping mechanism. Definitive resolution as to whether crystal thickness or surface step-edge density is the key factor in the doping mechanism-as the two structural characteristics are proportional to each other-is presented below, but first we complete the discussion of transport effects.
The temperature dependences of the FET I D -V G characteristics for the same Cl 2 -NDI crystal in its as-grown and doped states, respectively, are shown in Fig. 3a,b. The linear regime electron mobility (μ) and threshold voltages extracted from these traces are displayed in Fig. 3c,d (see also Supplementary Fig. 9 for V G dependence of mobility). The doped crystal exhibits a factor of ten higher room temperature electron mobility and the mobility increases as temperature decreases, reaching 9.4 cm 2 V −1 s −1 at 150 K, which is indicative of band-like transport. For the as-grown crystal the mobility is significantly lower at all temperatures and clearly thermally activated, consistent with trapping. The profoundly different μ-T behaviour in Fig. 3c suggests substantial reduction of shallow traps on the surface of the doped crystal [33][34][35] . The V T behaviour is also consistent with this interpretation. For the doped crystal, the slope of the V T -T correlation is two to three times smaller than the corresponding slope for the same crystal in its as-grown state, and V T is only a few volts even at 150 K for the doped version, whereas it is tens of volts for the as-grown case ( Fig. 3d and Supplementary Fig. 10). The smaller V T values and the weaker temperature dependence are consistent with less trapping for the doped crystal 36,37 . Together, the μ and V T behaviours thus reveal a strong qualitative improvement in the transport behaviour for Cl 2 -NDI crystals on doping.   Again, similar observations were made for as-grown and doped PDIF-CN 2 crystals (Supplementary Fig. 11). Furthermore, pronounced band-like charge transport was observed in all doped devices, regardless of thickness or step density, which was in sharp contrast to the as-grown crystals that displayed band-like behaviour only for the very thinnest crystals, that is, those with the lowest step densities (Fig. 2c). We note also that the doped devices were stable. Mobilities for devices stored in air for two months degraded by only about 18% (Supplementary Figs. 12 and 13).
We turn now to the mechanism of n-doping. The size of the N-silane molecule and the tight-packing of the Cl 2 -NDI and PDIF-CN 2 crystals suggested to us that the doping mechanism was a surface, not a bulk, phenomenon, as has been reported before for fluoroalkyl silane surface doping of rubrene crystals 16,38 . To prove this point we carried out depth profiling experiments to measure the concentration of N-silane as a function of depth into the crystal. The data indicate that indeed the N-silane does not penetrate into the bulk of the crystal, but is confined to the surface, as expected ( Supplementary Fig. 14). Additionally, X-ray diffraction revealed no evidence of crystal expansion after doping ( Supplementary Fig. 15) 18,39 .
To understand the surface effects of N-silane exposure, we undertook scanning probe microscopy analysis of the crystal surfaces. not evident in untreated crystals, indicating the N-silane selectively interacts with molecules at the step edges (see another example in Supplementary Fig. 16). As noted already, N-silane readily undergoes electron transfer reactions with Cl 2 -NDI and PDIF-CN 2 in solution to make the corresponding radical anions ( Supplementary  Fig. 4, 5). This is illustrated for the case of Cl 2 -NDI in Fig. 5a, which shows the UV-vis absorption spectra for Cl 2 -NDI on spectroelectrochemical reduction in solution in comparison to the observations for a thin film of Cl 2 -NDI exposed to N-silane. The bands at roughly 480, 520 and 600 nm are clear signatures of the Cl 2 -NDI radical anion (Supplementary Fig. 4) 40 . Similar results are obtained for PDIF-CN 2 in solution ( Supplementary Fig. 5). For the crystals, we speculate that N-silane can reduce Cl 2 -NDI and PDIF-CN 2 molecules located at the step edges because the redox-active pi-cores of the molecules are exposed on the step faces (Fig. 1c). In contrast, the crystalline terraces, which correspond to the (001) planes, are terminated with fluoroalkyl chains that effectively block molecular interaction and electron transfer. Thus, it is the unit cell structure of Cl 2 -NDI and PDIF-CN 2 crystals, combined perhaps with the less aggressive reducing power of the N-silane, which provides the step-edge specificity of the doping chemistry. Silanes are known to undergo oligomerization, and once reaction has occurred at the step edge further reactions may follow to create a several-nanometre thick N-silane multilayer stripe as shown in Fig. 4a,c (ref. 41 ). The SKPM images in Fig. 4b,d, which correspond to the topographic images in Fig. 4a,c, show striking electric potential contrast 38,42 . Coincident with each step edge for both doped crystals, positive and negative potential stripes are clearly evident. In particular, one sees that the positive stripe at each step edge is flanked by negative potential stripes on either side. Insets to the panels show line scans across the step-edge potential stripes. From the profile in Fig. 4b for Cl 2 -NDI, it is evident that the peak positive potential is roughly +50 mV and the peak negative potential exceeds −100 mV, which are both very large values. The total width of the entire stripe is nearly 5 μm. For PDIF-CN 2 in Fig. 4d, the results are similar, with the exception that the peak negative potentials are even lower, that is, less than −300 mV. In our previous work on undoped crystals, only positive SKPM potentials were observed at step edges 23 . Thus, the potential signatures measured at the step edges have completely changed after treatment with the N-silane. To assess the robustness of the Fig. 4 results, we have carried out many SKPM experiments under different scanning conditions and modes (for example, amplitude modulation versus frequency modulation). While the precise potentials observed in Fig. 4 depend on tip lift height and a.c. bias, the overall 'triple stripe' pattern at the step edges is always observed for doped crystals, and it is distinct from the step-edge potential measured for undoped samples (Supplementary Figs. 16-30).
We also undertook UV photoelectron spectroscopy (UPS) measurements of doped crystals with high step densities to confirm our interpretation of the effects of N-silane exposure. UPS is a classical technique for examining band alignment in doped semiconductors [43][44][45][46] and in this case, it also has the advantage that it averages over a geometrical area much larger than the distances between steps and so provides complementary information to SKPM. Figure  5b compares the UPS results for doped and as-grown Cl 2 -NDI crystals in the valence (highest occupied molecular orbital, HOMO) band and photoelectron cut-off regions (Supplementary Fig. 31). It is clear that the surface-doped Cl 2 -NDI crystals display both a 0.3 eV deeper HOMO and a 0.5 eV smaller work function (higher binding energy at cut-off). These results are depicted in the energy level diagram on the right side of Fig. 5b. Positive ionization of the N-silane by electron donation to subsurface layers of the crystal results in an effective dipole at the crystal surface that lowers the work function (that is, it creates a downward vacuum level shift), consistent with the UPS observation. Likewise, a higher HOMO binding energy relative to the Fermi level for the doped crystal is indicative of a downward shift in the LUMO (lowest unoccupied molecular orbital) and HOMO energies and thus indeed reflects n-doping (the LUMO-to-Fermi level offset is reduced). Thus, the UPS data in Fig. 5b are consistent with the n-type step-edge doping mechanism.
Step edge From our collective experiments, an explanation for the potential images in Fig. 4b,d thus emerges; we conclude that ionized donors, localized at the step edges, create a narrow line of positive charge that is then screened by mobile electrons and the crystal dielectric response. The positive and negative charge distributions associated with a single step edge are depicted schematically in Fig. 5c. Narrow bright stripes in Fig. 4b,d correspond to the positive donor line charge; the broader, flanking dark stripes correspond to the negatively charged electron atmospheres confined to the vicinity of the step edges by the Coulomb potential of the ionized donors. It is this triple stripe space charge distribution at each step edge that is captured by the SKPM potential images, providing a striking visualization of microscopic charge transfer effects in organic semiconductors. Indeed, SKPM has well-known sensitivity to charge [47][48][49][50][51] and, in future work, computational simulation of both the charge distribution and full tip-sample interaction will be used to quantitatively relate the unusual shapes of the measured potentials to the charge density profiles 47 . For the present, we note simply that the shapes and spatial extents of the potentials definitively reflect doping-induced electron distributions in the vicinity of positive surface line charges.
In conclusion, we have discovered a site-specific n-type doping mechanism on the surfaces of single crystals of two benchmark organic semiconductors. The site-specific doping simultaneously eliminates electron traps and increases the background electron concentration, which in turn leads to superior ON-state conductances for FETs based on the doped crystals. The doping mechanism targets specific, well-defined structural features on the surfaces of the crystals-namely, step edges-that have been shown to be detrimental to transport. This site-specific trap elimination and doping in crystalline organic semiconductors offers the intriguing possibility of a general strategy in which targeted doping chemistry selectively 'erases' the effects of well-defined trap states. A general strategy of course hinges on continued identification of other well-defined structural features, such as grain boundaries, which serve as charge traps in crystalline organic materials, and specific doping chemistry that can address them.
Our study has also provided salient images of doping-induced space charge in organic semiconductor crystals. In the context of organic semiconductor science, this is gratifying confirmation that the classical picture of doping in conventional semiconductors applies well to organic systems. Specifically, the SKPM images demonstrate that electrons released into the host crystal by ionized donors are mobile and delocalize subject to the Coulomb potential associated with the spatially localized ionized donors along the step edges. We anticipate that site-specific chemical doping demonstrated here, in combination with SKPM imaging, will be a powerful approach for further understanding of microscopic doping effects and defects in crystalline organic semiconductor systems.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/ s41563-021-01079-z.
Pt coated, resonant frequency 60-90 kHz, k = 2-5.5 N m −1 , R C = 25 nm). SKPM topography and potential scans were conducted with a tip d.c. voltage V AC = 6 V and lift height d = 10 nm inside an Ar-filled glovebox with oxygen levels (roughly 1 ppm) to mitigate surface contamination effects. Other scanning conditions and corresponding images are given in the Supplementary Information.