Leave out the acceptor
Light absorbed by an organic semiconductor makes an exciton: an electron and a hole still bound to each other. The usual way to split it is a bulk heterojunction (BHJ), a blend of a donor and an acceptor in which the energy offset at the interface pulls the pair apart. Blends work, but their nanoscale mixing must be controlled carefully, can change over time, and creates charge-transfer (CT) states below the optical gap that cost voltage and can raise the dark current.
A single-component device uses one material only. Here it is DCV2-5T, an acceptor–donor–acceptor molecule: a chain of five thiophene rings with a dicyanovinyl group at each end. It absorbs strongly from 450 to 700 nm and its molecules stack face to face. Until now it has mostly served as the donor in blends with C60.
Schematic, after Fig. 1a and the introduction. The molecule drawing is a block diagram, not the chemical structure.
Three stacks, all evaporated in vacuum
Every layer is thermally evaporated, without solvents and without heating the substrate. On ITO (90 nm) sits a p-doped hole-transport layer, BPAPF:NDP9 (30 nm), a thin undoped BPAPF layer (5 nm), the 40 nm active layer, BPhen (8 nm) as the electron-side layer, and a silver top contact (100 nm).
The active layer is either the blend DCV2-5T:C60 (2:1, the BHJ reference) or neat DCV2-5T (the single-component device, SC). The optimised SC device adds a caesium-doped BPhen electron-transport layer (ETL) of 50 nm. Light enters through the ITO; the area is 6.44 mm².
After Fig. 1c,d and the Experimental Section; layer thicknesses are not to scale. The position of the doped ETL in the drawing is schematic.
A quarter of the blend's efficiency, without any bias
The external quantum efficiency (EQE) is the fraction of incident photons that come out as electrons in the circuit. At zero bias the blend peaks at 68% (610 nm), the neat film at 20% (675 nm): four times less. Optical simulations show that both active layers absorb about the same, so the internal quantum efficiency of the neat film is also four times lower, about 20%.
A reverse bias helps the neat film a lot. At −3 V its EQE grows by a factor of three, almost to the level of the blend, while the blend itself gains only a little. The neat film therefore makes charges reasonably well; at zero bias many of them are simply not extracted.
Peak values from Table 1; the factor of three from the text. Curve shapes are schematic, after Fig. 2a,c,d.
No CT states, more voltage
Under 100 mW cm−2 of white light the open-circuit voltage VOC rises from 0.98 V for the blend to 1.20 V for the neat film, although both have nearly the same optical gap (1.73 and 1.72 eV). In the blend, sensitive EQE measurements find CT states at ECT = 1.53 eV and the electroluminescence shows CT emission. The neat film shows no CT emission.
Splitting the loss: the radiative limit Vrad is 1.27 V for the blend and 1.38 V for the neat film, and the non-radiative loss ΔVOC,nr drops from 0.29 to 0.18 V, very small for an organic device. For the neat film a weak subgap feature is left out and only the bulk absorption edge is used.
All values from Table 1 and the text.
Two orders less dark current, one order more detectivity
A detector must tell a weak signal from its own noise. At −0.1 V the dark current is 9.6 × 10−9 A cm−2 for the blend and 2.4 × 10−9 A cm−2 for the neat film. Both are set by ohmic leakage through a low shunt resistance: the neat film is rough, with spikes up to 10 nm. A thick doped ETL covers them. With 50 nm of it the dark current falls to 3.4 × 10−11 A cm−2, among the lowest reported and below a typical silicon photodiode, while the photocurrent only halves.
The specific detectivity D* compares the responsivity with the measured noise. It rises from 7 × 1011 to 1 × 1013 Jones. Thermal noise dominates, and it scales with the square root of the leakage, so two orders less dark current buy one order in D*. Reverse bias raises the EQE but also the dark current, so zero bias gives the best D*.
Values from Table 1 and the text; D* from the measured noise spectral density (Fig. 3c).
Linear over 150 dB, fast to 200 kHz
The photocurrent at zero bias stays proportional to the light intensity (slope 1.01) over a linear dynamic range of 150 dB with the doped ETL and 190 dB without it, limited by the measurement setup. A slope this close to one means almost no bimolecular recombination. At the highest intensities the thick-ETL device falls slightly below the line.
Under a 660 nm LED (33 mW cm−2) the signal drops by 3 dB at 330 kHz without and 200 kHz with the doped ETL, both faster than the blend at 120 kHz. The thick ETL slows charge transport a little, which costs some speed and some linearity.
Values from the text (Fig. 3e,f; blend in Fig. S6d). Curves are schematic: straight lines and a first-order roll-off at the quoted cut-off frequencies.
Telling excitons from charges by colour
In transient absorption a short pump pulse (650 nm) excites the film and a broadband probe records how the absorption changes. Below 700 nm the ground-state bleach shows that molecules are excited. In the near infrared, the exciton of DCV2-5T absorbs strongly at 1000 nm. Charges show up as a shoulder at 870 nm and as an absorption just above 700 nm. Late on, a band at 1020 nm appears from triplets.
A multivariate curve-resolution analysis splits the measured spectra into these three components (excitons, charges, triplets) and returns how much of each is present at every delay. Four samples were measured with the same absorbed photon density of 1.3 × 1018 cm−3: a neat film, a film with transport layers, and devices with and without them.
Schematic, after Fig. 4a–c: band positions from the text, shapes illustrative. The probe skips 720–820 nm (two white-light continua).
Charges within a picosecond, in the bulk
Some charges exist already within the 100 fs time resolution: about 20% of the excitations in the films and about 60% in the devices. In the films the charge population keeps growing until about 50 ps, as excitons decay with time constants of a few hundred femtoseconds and tens of picoseconds. In the devices it peaks after about 1 ps.
After 1 ns all samples hold a similar charge population. So the electrodes speed charge generation up but do not raise the final yield, while transport layers hardly matter. Photoluminescence is not quenched by the transport layers either. The charges are therefore made in the bulk of DCV2-5T, not at its interfaces.
Schematic, after Fig. 4d: initial fractions, peak times and the convergence after 1 ns from the text; curve shapes illustrative.
Disorder helps excitons spread out
Calculations by the TU Munich co-authors look at how neighbouring molecules couple. Sliding one molecule along the backbone of its neighbour by a shift Δ changes the electronic and excitonic coupling between them anywhere from −250 to +200 meV. An exciton model of about 36 864 molecules then adds random shifts: none (L1), on 50% of the neighbour pairs (L2) or on 95% (L3). The intermediate case L2 reproduces the measured EQE at −3 V up to 2.3 eV.
With disorder, molecular excitons mix strongly with intermolecular CT states. The proposed path is a superexchange-like step: the hole hops to the neighbour (tHOMO), the electron follows (tLUMO), and the intermediate CT state is only virtual. The exciton becomes more delocalised and splits more easily. A perfectly ordered crystal couples less, so the material does not need to be highly crystalline.
Schematic, after Fig. 5a,b,d: the coupling curve shows only the quoted range, not the calculated values.