Elimination of the bias-stress effect in ligand-free quantum dot field-effect transistors

Colloidal quantum dot (QD) thin-film field-effect transistors (FETs) are important devices for studying charge transport in QD solids and fabricating QD-based electronics.^1,2 Unfortunately, QD FETs commonly exhibit hysteresis in their current–voltage characteristics caused by the bias-stress effect (BSE). The BSE occurs when the application of a gate bias (V_G) triggers the accumulation of immobile charges near the gate/channel interface that progressively screens the gate field and causes a time-dependent (transient) decay of the mobile carrier density and drain current (I_D) in the channel. The immobile charges can be electrons/holes that are trapped in pre-existing or bias-induced traps or ions such as protons, hydroxide ions, and charged ligands that are pre-existing or induced by the bias stress. In addition to causing I_D decay, progressive screening of the gate field also increases the magnitude of the gate voltage needed to induce a conducting channel (the threshold voltage, V_T). V_T shifts and I_D transients under an applied gate bias are hallmarks of the BSE (Fig. 1). Figures 1(b) and 1(c) show generic electronic and ionic processes that can cause the BSE in QD FETs.

The BSE has been a persistent problem for PbX (X = S, Se, and Te) QD FETs.^1–17 In our own work,^{4–6,8,10,17} we have viewed the BSE as a curiosity, an annoyance, and a reason to better understand charge transport in these systems. Hysteresis and an unstable I_D complicate measurements of transport physics and device performance (e.g., carrier mobility) and prevent the use of these transistors in applications that require stable device characteristics. In cases where the BSE transients are sufficiently slow, a workaround for charge transport studies is to scan current–voltage (I–V) curves as fast as possible to minimize the hysteretic distortions caused by the transients,⁸ but this tactic sometimes fails.¹⁷ It would be better to find a way to eliminate the BSE altogether. Proposed causes of the BSE in PbX QD FETs include dynamic charge trapping by either the ligands,⁹ QD surface defects such as Pb dimers,¹⁸ or H₂O/O₂ adsorbates within the QD films.¹³ However, these hypotheses are unproven, and there has been very limited progress in making BSE-free devices. The standard remedy of employing hydroxyl-free gate dielectrics with low interfacial trap densities^19,20 is usually ineffective for PbX QD FETs, which indicates that the immobile charges responsible for the BSE tend to accumulate within the QD films, not on the surface or in the bulk of the gate dielectric.^4,9,13,21 Zhou et al. tried to stop the BSE in p-channel PbS QD FETs by inserting a MoO_3−x electron blocking layer at the source/drain electrodes, but large transients remained after the electrode modification.¹⁵ Our group achieved strong suppression (but not complete elimination) of BSE transients in Na₂S-treated PbSe QD FETs by infilling the QD films with amorphous aluminum oxide (alumina) using atomic layer deposition (ALD).¹⁰ Now we describe an improved matrix engineering approach that utilizes H₂S-induced ligand removal and alumina infilling to completely eliminate the BSE in PbSe QD FETs, yielding hysteresis-free n-channel transistors with high electron mobilities. We show that the BSE is caused by both charge trapping at the QD surface and proton motion within the films and demonstrate that BSE elimination requires both the formation of a ligand-free PbSe/PbS/alumina interface (which removes the traps responsible for the electronic part of the BSE) and full infilling of the QD film with an ALD coating (which stops the proton motion that causes the ionic part of the BSE). The introduction of BSE-free PbX QD FETs should accelerate progress in charge transport studies and optoelectronic applications of QD solids.

Figure 2(a) shows the three-step process that we used to fabricate PbSe QD FETs that are free of the bias-stress effect (see Methods). We began by depositing conductive amorphous QD films onto prepatterned FET substrates [p⁺⁺ (100)-oriented Si coated with a 200 nm thick SiO₂ layer and patterned with metal source and drain electrodes (5 nm Cr/45 nm Au)] using layer-by-layer dip coating^4,8 to quantitatively replace the native oleate ligands with small ligands that have a high vapor pressure when in their neutral protonated form, such as 1,2-ethanedithiolate (EDT²⁻), formate (HCOO⁻), or thiocyanate (SCN⁻). Next, the samples were exposed to a dose of H₂S gas in a glovebox-integrated ALD chamber to protonate the small ligands, which then quantitatively evaporated from the QD films as neutral molecules, leaving behind films that are presumably capped with about one monolayer of HS⁻ ligands. Finally, in the third step, the QD films were infilled and overcoated with amorphous alumina (a-Al₂O₃)^10,22 in the same ALD chamber to produce all-inorganic, ligand-free, n-channel QD FETs that show zero BSE transients at room temperature. As we explain below, the key to eliminating the BSE in these devices is to (i) fully remove the ligands to produce atomically clean PbSe/PbS/Al₂O₃ interfaces with a relatively low concentration of electron traps, and (ii) fully infill the QD films to prevent internal ion (proton) motion. We found that the H₂S-treated, alumina-infilled QD FETs showed very similar performance characteristics—including high linear electron mobility (3–14 cm² V⁻¹ s⁻¹), total long-term air stability,^10,22 and zero BSE transients (no current-voltage hysteresis)—regardless of which small volatile ligand was used in their fabrication.

We evaluated the performance of these bottom-contact, global back gate FETs [Fig. 2(b)] at room temperature. Figures 2(c)–2(l) show representative results for EDT-capped devices (6.3 nm diameter QDs). Prior to any ALD treatments (H₂S or a-Al₂O₃), the EDT-capped devices behaved as ambipolar transistors with relatively low electron and hole mobility and significant drain current (I_D) hysteresis due to prominent BSE transients [Figs. 2(c) and 2(d)].⁴ Exposing the FETs to one pulse of H₂S at a substrate temperature of 54 °C caused a dramatic change in their current–voltage characteristics: the FETs became p-channel transistors featuring quasi-linear output plots, weak gate modulation of I_D, and somewhat slower I_D transients [Figs. 2(e) and 2(f)]. These H₂S-dosed devices behaved in a very similar fashion to sulfide-capped FETs previously prepared in our laboratory by solution-phase ligand exchange with Na₂S.¹⁰ The similarity in behavior is likely due to the abundance of electronically active sulfides (S²⁻ or SH⁻) adsorbed on the surface of the QDs in both types of films.¹¹ Remarkably, upon fully infilling and overcoating the H₂S-dosed devices with amorphous alumina using ALD at 54 °C, the FETs became n-channel transistors with high linear electron mobility (3–4 cm² V⁻¹ s⁻¹), insignificant hysteresis in output plots, and negligible BSE transients at room temperature [Figs. 2(g) and 2(h)]. Transfer plots acquired as a function of bias stress soaking time clearly show the absence of BSE in the H₂S-treated, alumina-infilled FETs [Figs. 2(i) and 2(j)]. We also characterized I_D vs time by measuring I_D at a constant source–drain voltage (V_SD) while stepping the gate voltage (V_G) from 0 to ±50 V in 10 V increments. Figures 2(k) and 2(l) are log-linear and normalized plots of |I_D| vs time for these devices. While both the EDT-capped and H₂S-dosed devices exhibited quasi-exponential I_D transients after each step in V_G, the H₂S-dosed and alumina-infilled device showed no transients. A linear plot of these data confirmed that the absence of the transients is real, not an artifact of y-axis scaling and higher device conductivity (Fig. S1 in the supplementary material). The staircase-like shape of the I_D vs time plots is characteristic of a well-behaved FET that responds to sudden changes in V_G with sudden and stable changes in I_D. This H₂S/alumina surface chemistry produces what is to our knowledge the first example of a PbX QD FET without hysteresis at room temperature.

We found that the BSE transients were eliminated by the combination of H₂S dosing and alumina infilling only when volatile ligands were used to produce the initial QD films. For example, both formic acid (HCOOH) and hydrocyanic acid (HSCN) have high vapor pressures^23,24 and, thus, readily evaporate at the substrate temperatures used for H₂S dosing (54–75 °C). Fourier transform infrared (FTIR) spectra show that exposing formate- or thiocyanate-capped QD films to an H₂S pulse of 3 × 10⁶ Langmuir (L) at 54 °C results in the complete loss of all ligand vibrational signatures (Fig. 3). The carboxylate peaks of adsorbed formate at 1554 and 1327 cm⁻¹ and the characteristic –C≡N stretch of thiocyanate at 2021 cm⁻¹ disappeared after H₂S exposure (blue traces). This is consistent with protonation and evaporation of the ligands by H₂S. Subsequent infilling with ALD alumina caused little additional change to the FTIR spectra, but the resulting FETs were free of transients (red traces in Fig. 3) and showed I–V plots that were nearly identical to those of H₂S-treated and alumina-infilled FETs made from EDT-capped QD films (Figs. S2 and S3). The electron mobility of these FETs is independent of the V_G sweep rate used to acquire the transfer plots from which the mobility values are extracted (Fig. S4). As with the EDT-capped films, both H₂S ligand exchange and alumina infilling were required to eliminate the transients of formate- and thiocyanate-capped films. H₂S exposure alone removed the ligands but worsened the transients (blue traces), while infilling alone removed most of the ligands but only part of the transients (gray traces). Optical absorption spectra indicated that a similar amount of redshifting and broadening of the first exciton absorption peak occurred for the EDT-, formate-, and thiocyanate-capped films upon H₂S dosing and again upon infilling (Figs. S1–S3). The excitonic peak remained distinct and narrow for all films, indicating that the H₂S and infilling steps did not cause a significant increase in the polydispersity of the QDs by etching, necking, sintering, or ripening.

As a check on the above-mentioned results, we verified that the BSE transients persisted if nonvolatile ligands were used to prepare the initial QD films. Figure 4 shows FTIR and I_D time trace data for films capped with oxalate, trimesate, and 1,4-benzenedithiolate (BDT) ligands, each of which has a very low vapor pressure as a neutral molecule.²⁵ As expected, H₂S dosing did not remove these ligands from the films because the ligands do not appreciably evaporate when protonated. Indeed, the absence of C=O and S–H peaks in the FTIR spectra of H₂S-dosed films suggests that these three ligands are not protonated by H₂S to any significant extent. Since the neutral ligands are kinetically trapped on the QD surface in the absence of solvent, any ephemerally protonated ligands are likely to quickly back react to expel H₂S. Subsequent alumina infilling also had little impact on the ligand loading of the films. All the FETs made using nonvolatile ligands showed strong BSE transients, with the magnitude and time constants of the transients varying somewhat for the different ligands and steps in the fabrication process (Fig. 4 and Figs. S5–S7). Our data show that ligand removal is a necessary but insufficient condition for eliminating the transients. Complete elimination of the transients requires ligand removal with H₂S and infilling/overcoating of the QDs with alumina. As discussed in more detail below, the strong correlation between the chemical state of the QD surface and the presence of transients suggests that the electron BSE in our n-channel PbSe QD FETs is caused by screening of the applied gate field by species on the surface of the QDs rather than on or in the gate dielectric or near the source/drain contacts. It is likely that QDs with “dirty,” ligand-covered, or poorly passivated surfaces suffer particularly strong BSE transients.

The stable device characteristics of the BSE-free FETs enabled the first unambiguous determination of the dependence of electron mobility on temperature and QD size. Figure 5(a) shows the dependence of the linear electron mobility on temperature (80–300 K) for H₂S-treated and alumina-infilled FETs made from SCN- or EDT-capped films of 6.1 nm diameter QDs. The mobility is weakly thermally activated (dμ/dT > 0) for both types of FETs and increased by only 35%–40% over this temperature range. The simple Arrhenius nearest-neighbor hopping expression μ(T) = μ_∞ exp(−E_A/kT) with a temperature-independent prefactor μ_∞ yielded reasonable fits of the data using very small E_A values of 3–4 meV (data not shown). These activation energies are similar to values recently reported for hole transport in single-grain PbSe QD superlattice transistors¹⁷ and suggest that carriers follow low-barrier percolative pathways through the films.⁸ Whereas the superlattice FETs suffered from thermally-activated BSE transients that suppressed the high-temperature mobility, created (in whole or part) a negative dμ/dT region above ∼150 K, and rendered μ(T) ambiguous, the BSE-free amorphous FETs studied here show unambiguous μ(T) curves that are consistent with hopping transport.

Figure 5(b) shows the room-temperature electron mobility of FETs made from EDT-capped films as a function of QD diameter. In 2010, we reported that the electron mobility of EDT-capped PbSe QD FETs has a maximum at a QD diameter of ∼6 nm (lower orange trace).⁸ However, the devices used in that study suffered from fast and diameter-dependent BSE transients. Despite employing even faster V_G sweeps to acquire transfer curves, we could not be certain at the time that our measured mobility values and resulting mobility-size plots were unaffected by the transients. Here, we reproduced our previous work and found that devices without H₂S dosing and alumina infilling showed the same size dependence as before, but with slightly higher mobility values (upper orange trace). After H₂S treatment and alumina infilling, the BSE transients were eliminated and the mobility values increased by about an order of magnitude, but the basic size dependence of the mobility was unchanged (red trace). These data confirm that our previously-reported dependence of the electron mobility on QD size is real, not an artifact of the BSE. These results also demonstrate that the H₂S/alumina chemistry eliminates the BSE for a range of QD diameters.

Increasing the ALD temperature from 54 to 75 °C substantially increased the FET mobilities. As with the devices made at 54 °C, the highest mobilities were achieved using thiocyanate-capped QD films (Figs. 2, 3, and 5). Figure 6 shows representative data for thiocyanate-capped films dosed with H₂S and infilled with alumina at 75 °C, which we know from past work is a sufficiently low temperature to avoid QD sintering and coarsening during ALD.¹⁹ We measured a room-temperature electron mobility of 12–14 cm² V⁻¹ s⁻¹ for these FETs [Fig. 6(b)]. These are the highest-mobility PbX QD FETs made by our group to date. The devices combine high electron mobility with negligible BSE/hysteresis (i.e., total short-term stability) and indefinite air stability (i.e., total long-term stability).^10,22 We speculate that the higher ALD temperature improves the electron mobility by producing a higher-quality PbSe/PbS/alumina interface and alumina coating that better passivate the QD surface.

To determine how the H₂S/alumina chemistry eliminates the BSE in our ligand-free QD FETs, we began by verifying that the transients are associated with the surface of the QDs rather than the gate dielectric or FET contacts. We tested many different treatments and modifications of the bare FET substrates, including annealing up to 200 °C in different environments, plasma treatments, self-assembled monolayers, the use of alumina as a gate dielectric, the use of different source/drain contact metals, and various combinations of H₂S dosing and alumina coating prior to depositing the QD films (Fig. S11), but these efforts failed to significantly suppress the BSE transients. The fact that the transients were eliminated only by H₂S/alumina treatment of the QD films further convinced us that the BSE in these n-FETs is caused by the accumulation of immobile charge within the QD films rather than in other parts of these devices.

We next measured the elemental composition of the H₂S-treated and alumina-infilled films by secondary ion mass spectrometry (SIMS) depth profiling to gain insight into the nature of the QD/alumina interfaces produced by the H₂S dosing and alumina infilling processes. To set a baseline for SIMS studies of the films, we first determined the amount of sulfur and other elemental impurities in a bulk powder of our oleate-capped PbSe QDs using glow discharge mass spectrometry (GDMS). GDMS provides a full elemental survey (Li–U) with ppb–ppm detection limits.²⁶ The GDMS data, compiled in Table I, showed a total of 15 elements above the limit of detection out of 74 elements quantified. Note that matrix elements (Pb and Se), gas-forming elements (H, C, N, O, and noble gases), and radioactive elements were not quantified in this analysis. The elements above 1 ppm (equivalent to ∼5 × 10¹⁶ atoms cm⁻³ in the QD film, see Methods) were phosphorus (2040 ppm), silicon (31 ppm), chlorine (25 ppm), sulfur (6 ppm), boron (3.1 ppm), calcium (1.3 ppm), aluminum (1.0 ppm), sodium (1.0 ppm), and bismuth (0.94 ppm). For the purposes of this paper, the most important conclusion from the GDMS data is that the sulfur content of our as-made QDs was quite low (<10 ppm), which is a prerequisite for studying the sulfur content of the H₂S treated and infilled films (see below). However, several other aspects of the GDMS data are noteworthy. First, the high level of phosphorus (equivalent to ∼15% of a monolayer on the QD surfaces) is reasonable given the use of trioctylphosphine selenide (TOP-Se) and diphenylphosphine (DPP) in the QD synthesis. We speculate that the phosphorus is present mostly as adsorbed molecular species derived from DPP and TOP-Se rather than as point defects in the PbSe lattice. Second, ignoring phosphorus, a total impurity content of only ∼70 ppm is striking because, on the one hand, the QDs were surprisingly pure despite a lack of special effort on our part to ensure high-purity conditions, but, on the other hand, several of the impurities were in sufficiently high concentration to act as dopants, traps, and recombination centers in the QD films (particularly Si, Cl, B, Al, and Na). A better understanding of the location and electronic behavior of these impurities is needed to determine whether they may cause deep traps that limit the performance of QD optoelectronic devices.

We used SIMS mainly to quantify the amount of sulfur present at the PbSe/Al₂O₃ interface of our BSE-free FETs. Initially, we hypothesized that the SH⁻ adsorbed on the QDs during H₂S treatment would be removed by H₂O during subsequent a-Al₂O₃ ALD, resulting in clean PbSe/Al₂O₃ interfaces. However, SIMS showed that about 0.4 monolayers (MLs) of sulfur (1.5 × 10²¹ atoms cm⁻³) remained on the surface of the QDs when formate-capped films were dosed with H₂S and then infilled with alumina in the standard way. This sulfur content is the same as that of Na₂S-treated, ALD-infilled PbSe QD films previously reported by our group.¹⁰ SIMS control experiments on a formate-capped film infilled with alumina without intentional exposure to H₂S (or any other source of sulfur) showed a sulfur concentration about 19 times lower (8 × 10¹⁹ atoms cm⁻³ = 1740 ppm, or about 2% of a monolayer; Fig. S13), confirming that the sulfur observed in the transient-free films came from H₂S exposure rather than some uncontrolled background source. We cannot, at present, account for why the sulfur concentration of the supposedly sulfur-free infilled films was so much higher than that of the as-made QD powder measured by GDMS (1740 ppm vs 6 ppm), but adventitious adsorption of thiols from the atmosphere of our glovebox (e.g., EDT) during the fabrication of the films may be responsible. Additional control experiments on a benzenedithiolate-capped film infilled with alumina without any exposure to H₂S showed tenfold higher sulfur and carbon levels and fourfold higher hydrogen levels than the transient-free film, consistent with the retention of BDT ligands after ALD, as seen in IR spectra (Fig. S13 and Fig. 4). All three films had similar amounts of phosphorus (6–10 × 10¹⁸ at. cm⁻³). The SIMS results show that the structure of the QD/alumina interface in the BSE-free FETs is PbSe/PbS_{(∼0.4 ML)}/Al₂O₃ rather than PbSe/Al₂O₃.

We have demonstrated that in vacuo dosing with H₂S quantitatively protonates and desorbs volatile anionic ligands from the surface of the QD films. H₂O dosing was found to work in a similar fashion, but larger doses of H₂O were needed to achieve complete ligand exchange because H₂O is a significantly weaker acid than H₂S.²⁷ The IR spectra in Fig. 7(a) show the degree of ligand removal for formate-capped QD films exposed to different doses of H₂S and H₂O gas. As seen from the spectra, an H₂S dose of 3 × 10⁶ Langmuir (L) was sufficient to remove all formate from the film surface. In our ALD system, this dose was obtained by either a single large pulse (P_max ∼ 1 Torr) or seven smaller pulses (P_max ∼ 0.12 Torr) of H₂S. The same dose of H₂O removed only 60%–70% of the formate, but we could achieve quantitative removal of formate with a tenfold larger dose of H₂O (∼3 × 10⁷ L). If the formate was removed by H₂S, subsequent infilling with alumina eliminated the BSE transients (Fig. 3). If a large dose of H₂O was used instead, the transients of the infilled, ligand-free films were suppressed but not eliminated [Figs. 7(b) and 7(c)]. It therefore seems that some amount of sulfur at the PbSe/Al₂O₃ interface is beneficial for the complete elimination of the transients. The sulfur sub-monolayer may also be responsible for the higher conductivity of H₂S-treated films [Fig. 7(b)]. The much larger I_D of the H₂S-dosed devices at V_G = 0 V [Fig. 7(b)] was due to a more negative threshold voltage V_T for the H₂S-treated FETs (V_T = −6 V, compared to V_T ∼ 0 V for H₂O-treated FETs) and is not indicative of a lower on/off ratio.

Let us review what is known about the BSE transients so far. They are (i) thermally activated;^7,9,10,17 (ii) associated with the surface of the QDs rather than the gate dielectric or contacts; (iii) eliminated by ligand exchange with H₂S followed by coating the QDs with ALD alumina to form PbSe/PbS_{(∼0.4 ML)}/Al₂O₃ interfaces; (iv) suppressed but not eliminated by ligand exchange with large doses of H₂O followed by alumina coating. Since the transients are very sensitive to the chemistry of the QD surface, we next tested how they responded to changing (1) the pulse sequence used for alumina deposition, (2) the thickness of the alumina coating, and (3) the ALD material used to infill the QD FETs after H₂S exchange.

We found that the transients were eliminated only if alumina deposition was initiated with a pulse of H₂O. Starting instead with a pulse of trimethylaluminum (TMA) after H₂S dosing resulted in FETs with sizable BSE transients and lower conductivity, regardless of which ligand (EDT, formate, or thiocyanate) was used to make the QD films (Fig. 8). This suggests that the transients are caused by electronic states or mobile species on the PbSe surface that are better passivated by starting the alumina ALD sequence with H₂O instead of TMA. Starting with H₂O may passivate surface Pb ions that are not already coordinated with SH⁻, thereby eliminating surface states while also providing a more hydroxylated QD surface for better alumina nucleation and growth. In contrast, starting with TMA likely creates Pb–S–Al(Me)₂ species on the surface that sterically block the coordination of unpassivated Pb ions by H₂O and lead to a patchier and lower-quality alumina coating as a result.

Not only is the ALD pulse sequence important for eliminating the transients, but so also is the thickness of the alumina coating. We measured a series of FETs made by infilling H₂S-treated QD films with different thicknesses of alumina (0, 10, 20, 30, 40, and 120 ALD cycles) deposited by beginning with an H₂O pulse. Figure 9(a) shows that the transients became gradually weaker with an increasing number of ALD cycles and completely disappeared for devices made with at least 40 cycles (about the number needed to fully infill these QD films¹⁰). The I–V characteristics of these devices also showed a continuous evolution with the number of ALD cycles: the threshold voltage shifted from >60 V (at 0 cycles) to −4 ± 2 V (at ≥40 cycles), an n-channel appeared and gradually dominated the FET behavior, and the electron mobility increased from 0.71 to 1.9, 4.9, and 5.4 cm² V⁻¹ s⁻¹ for 20, 30, 40, and 120 ALD cycles, respectively [Fig. 9(b)]. A similar progression was observed during alumina infilling of Na₂S-treated QD FETs.¹⁰ TEM imaging showed that the alumina formed a continuous conformal coating at all thicknesses, with no evidence for island formation (Fig. S16). Thus, the fact that the transients stopped only after the QD films were fully infilled with alumina (at ∼40 cycles) rather than once the PbSe/PbS surface was passivated by the first few ALD cycles suggests that the BSE was at least partly caused by the drift of ions along the internal free surfaces of the films. We reasoned that protons on the QD and alumina surfaces could migrate in the transverse direction along the internal surface of the interconnected pore network of unfilled and partially-infilled films in response to V_G. In this picture, when V_G > 0 V, protons are gradually repelled from the channel, leaving a sheet of immobile −O⁻ (oxide) anions that progressively screens V_G and causes the BSE [Figs. 9(c) and 1(c)]. Fully infilling the films should fill much of the pore space and isolate the residual pores from each other, thereby preventing surface protons from drifting to screen V_G [Fig. 9(c)]. We note that Song et al. recently attributed I–V hysteresis in PbS QD solar cells and FETs to mobile protons within the QD films.²⁸ 

To test the hypothesis that proton drift is a cause of the BSE, we carried out alkali metal cation exchange studies of partially-infilled QD FETs. Thiocyanate-capped FETs dosed with H₂S and partially infilled with alumina (25 cycles at 54 °C) were soaked for three hours in 0.1 M solutions of alkali metal nitrates (MNO₃) in anhydrous dimethyl sulfoxide (DMSO) at room temperature to replace surface protons with Li⁺, Na⁺, K⁺, or Cs⁺ ions (Scheme ¹). We reasoned that these larger, heavier ions should be less mobile than protons, and the transients of the cation-exchange devices would weaken if proton drift was indeed a significant cause of the BSE. Figure 10(a) shows that the transients were dramatically suppressed after the MNO₃ soaks. LiNO₃ soaks suppressed the transients the least, while KNO₃ and CsNO₃ soaks suppressed them the most [Fig. 10(b)]. Stronger transient suppression by the larger cations can be explained by their smaller surface diffusivities and/or more complete exchange of surface protons. We found that the transient suppression was reversible by soaking the devices in pure water to reprotonate their surfaces [Fig. 10(c)]. Control experiments confirmed that the transients were affected predominantly by the alkali metal cations, not the DMSO or anion (Figs. S20–S22). From these studies, we conclude that proton drift is a second cause of the BSE in our FETs and that full infilling stops proton drift and eliminates the associated BSE transients by closing off pathways for proton motion.

We also investigated whether the BSE transients could be eliminated by ALD infills other than alumina. Full ZnO, ZnS, CdO, and TiO₂ infills were tested on H₂S-dosed devices processed at 54 °C. Infilling with ZnO produced n-channel FETs with very weak transients, while ZnS and CdO infilling gave n-FETs with strong transients and TiO₂ infilling yielded p-FETs with strong transients (Fig. S23). The failure of these four alternative infills to eliminate the transients shows that there is something special about the H₂S/alumina chemistry. Indeed, we found that ZnO and TiO₂ infills eliminated the transients if they were deposited on at least one cycle of ALD alumina. For example, while weak transients remained in thiocyanate-capped FETs dosed with H₂S and infilled with 250 cycles of ZnO, the transients were absent in FETs infilled with 1–10 cycles of Al₂O₃ followed by 250 cycles of ZnO (Fig. S24). Similarly, we made BSE-free n-FETs from EDT-capped films infilled with 20 cycles of alumina followed by 20 cycles of ZnO or TiO₂ (Fig. S25). These experiments show that the critical requirements for eliminating the transients are to produce a clean PbSe/PbS/alumina interface and then fully infill the films by ALD. The formation of the ligand-free PbSe/PbS/alumina interface evidently reduces charge trapping on the QD surfaces (the electronic mechanism of the BSE), while full infilling stops proton drift within the films (the ionic mechanism of the BSE).

Finally, we note that while H₂S dosing and alumina infilling eliminated the electron transients, residual hole transients persisted after the H₂S/alumina treatment. Close inspection of transfer curves reveals that the n-FETs had very weak p-channels at negative V_G [e.g., Fig. 2(h) and Figs. 2S3(h), and S26]. I_D time traces in the hole conduction regime at negative V_G showed significant BSE transients, despite the elimination of such transients in the electron conduction regime at positive V_G (Fig. S26). We did not study the hole transients in detail, so we can only speculate that they are caused by hole traps on the surface of the QDs or SiO₂ gate dielectric that are not passivated by the H₂S/alumina chemistry. Additional or alternative treatments may be required to eliminate hole transients and hole-related I–V hysteresis in PbSe QD FETs.

Transistors made from colloidal QD solids frequently suffer hysteresis from the BSE, which can complicate fundamental studies of charge transport and limit the development of practical QD-based electronics. We showed here that the BSE can be eliminated in PbSe QD n-FETs by H₂S ligand exchange and alumina infilling with ALD. The resulting H₂S-dosed, alumina-infilled FETs have stable, hysteresis-free device characteristics (total short-term stability), indefinite air stability (total long-term stability), and a high electron mobility up to 14 cm² V⁻¹ s⁻¹, making them attractive n-FETs for constructing high-performance QD circuits and optoelectronic devices. These BSE-free devices were used to unambiguously demonstrate how the electron mobility depends on temperature and QD diameter, illustrating the importance and utility of hysteresis-free devices in basic research. The high mobility of the BSE-free FETs is likely a product of reduced charge trapping, enhanced electronic coupling, and the absence of I_D transients (the latter of which cause systematic underestimation of the carrier mobility).

Our work has revealed evidence for both an electronic and ionic mechanism of the BSE in PbSe QD FETs. The H₂S/alumina chemistry produces ligand-free PbSe/PbS_{(∼0.4 ML)}/Al₂O₃ interfaces that lack the charge traps that cause the electronic part of the BSE. Meanwhile, full alumina infilling stops proton drift along the internal surfaces of the QD films, which is responsible for the ionic part of the BSE. The formation of a clean PbSe/PbS/Al₂O₃ interface was found to be essential for eliminating the BSE transients. Modifications to our standard H₂S/alumina process, including the use of nonvolatile ligands, H₂O rather than H₂S dosing, TMA-first ALD pulse sequences, partial infilling, and infilling with ALD materials other than alumina, failed to eliminate the BSE, suggesting that recipes for BSE elimination may be quite specific and different for different types of QD solids. We nevertheless believe that (i) ligand removal and surface passivation to stop charge trapping and (ii) ALD infilling to stop ion motion are good initial design goals in ongoing efforts to eliminate the BSE in other QD transistors.¹⁷ 

Chemicals. All chemicals were used as received unless otherwise noted. Lead oxide (PbO, 99.999%), selenium shot (99.999%), LiNO₃ (99%), and KNO₃ (99.0%) were purchased from Alfa Aesar. Oleic acid (OA, technical grade, 90%), diphenylphosphine (DPP, 98%), 1-octadecene (ODE, 90%), anhydrous hexanes (99%), anhydrous ethanol (≥99.5%), anhydrous methanol (99.8%), anhydrous acetonitrile (99.99%), anhydrous dimethyl sulfoxide (DMSO, ≥99.9%), anhydrous tetrachloroethylene (TCE, 99%), extra dry acetone (99.8%), acetone (>99.5%) and isopropanol (>99.5%) for substrate cleaning outside of the glovebox, 1,2-ethanedithiol (>98%), ammonium thiocyanate (99.99%), formic acid (98%), oxalic acid (≥99%), trimesic acid (95%), 1,4-benzenedithiol (BDT, 99%), 3-mercaptopropyltrimethoxysilane (3-MPTMS, 95%), sodium hydroxide (97%), NaNO₃ (≥99.0%), CsNO₃ (99.99%), LiClO₄ (≥98.0%), NaClO₄ (≥98.0%), tetrabutylammonium nitrate (97%), tetrabutylammonium perchlorate (≥99.0%), trimethylaluminum (97%), titanium tetrachloride (≥99.995%), and diethylzinc were purchased from Sigma-Aldrich. Trioctylphosphine (technical grade, >90%) was acquired from Fluka and mixed with Se shot for 24 h to form a 1 M trioctylphosphine-Se stock solution. Gold shot (99.99%), titanium, aluminum, and silver pieces, chromium plated tungsten rods, and molybdenum evaporation boats were purchased from Kurt Lesker. H₂S gas (99.3%) was purchased from Airgas. 18.2 MΩ water (Milli-Q Gradient) was used for substrate cleaning, atomic layer deposition (ALD), and cation exchange experiments. For ALD and other experiments inside the glovebox, the water was freeze-pump-thawed three times prior to use.

PbSe Quantum Dot Synthesis. PbSe QDs were synthesized and purified using standard air-free techniques. In a typical synthesis, 1 g of PbO, 5 g of oleic acid, and 15 g of ODE were mixed in a three-neck round-bottom flask and degassed at 80 °C under vacuum until a clear solution was formed. Once the solution became transparent, it was heated at 180 °C for at least 1 hour to dry the solution. While heating, 15 ml of 1 M TOP-Se solution containing 0.2 ml of DPP was rapidly injected into the hot solution. The QDs were grown for preselected times (30 s–5 min) depending on the desired QD size. The reaction was then quenched with a water bath and 15 ml of anhydrous hexane. The QDs were purified in an N₂-filled glovebox (<0.1 ppm O₂ and H₂O) by three rounds of redispersion and precipitation using hexanes and ethanol/methanol and stored as a powder in the glovebox.

QD film deposition. PbSe QD films were prepared by using a layer-by-layer procedure described elsewhere using a mechanical dip coater mounted inside a glovebox (DC Multi-8, Nima Technology). Briefly, substrates (microscope glass slide, silicon, quartz, or prepatterned FET substrates, all typically cleaned by sonication in Triton X-100 detergent and isopropanol followed by drying under N₂ flow) were alternately dipped into a 5 mg ml⁻¹ solution of QDs in anhydrous hexanes and then a solution of the desired short-chain ligand: 1 mM EDT, BDT, formic acid, trimesic acid, or oxalic acid in anhydrous acetonitrile, or 0.5 mM NH₄SCN in anhydrous acetone. In the case of BDT, NH₄SCN, oxalic acid, and trimesic acid, a third beaker containing neat acetonitrile or acetone was used to rinse the films after each dip in the ligand treatment to remove any ligand residue. We fabricated films with thicknesses in a range of 30–200 nm (thin for FETs, thicker for XRD and optical absorption studies).

Atomic layer deposition. H₂S dosing was performed in a homemade cold-wall traveling wave ALD system within a glovebox at a substrate temperature of 54 or 75 °C. H₂S gas was introduced using computer-controlled diaphragm valves in-line with a 100 SCCM stream of N₂ carrier gas. H₂S doses up to 3 × 10⁶ Langmuir (L) were obtained using either single long pulses (P ∼ 1 Torr) or a series of shorter pulses (P ∼ 0.12 Torr), with identical effects. H₂O dosing was performed in the same fashion but with exposures up to 3 × 10⁷ L.

Amorphous alumina (a-Al₂O₃) was deposited in the same ALD system from trimethylaluminum and water at a substrate temperature of 54 or 75 °C and an operating pressure of ∼0.1 Torr. Pulse and purge times were 20 ms and 120 s, respectively. The alumina growth rate was ∼1.1 Å per cycle at these temperatures.

Polycrystalline ZnO, ZnS, and CdO infills were grown by ALD using diethylzinc and water, diethylzinc and H₂S, and dimethyl cadmium and water, respectively. Amorphous TiO₂ infills were grown by ALD using TiCl₄ and water. All precursors were used at room temperature, and the substrate temperature was 54 °C.

MNO₃ treatments. Metal nitrate treatments of partially-infilled QD FETs (∼2.5 nm alumina) were performed in the glovebox. Samples were soaked in 0.1 M MNO₃ in anhydrous DMSO at room temperature for 3 h and then rinsed with anhydrous DMSO and blown dry. All other salt treatments (e.g., LiClO₄, TBANO₃) were performed in the same way. Soaks in degassed water (also in the glovebox) were used to reverse the cation exchange.

Characterization. Transmission electron microscopy (TEM) imaging was performed using a Philips CM20 operating at 200 kV. SEM images were acquired using an FEI Magellan 400 instrument operating at 10 kV and 100 pA. Optical extinction spectra of films on glass or quartz substrates were acquired using a PerkinElmer Lambda 950 spectrophotometer operating in the transmission mode. Films not protected from air with an ALD overcoat were sealed in optical cells consisting of two mated 1.33″ ConFlat sapphire viewports. FTIR transmission spectra of QD films on double-side polished intrinsic Si substrates were acquired in dry air using a Nicolet 6700 spectrometer at a resolution of 4 cm⁻¹ with a blank Si substrate as the background. X-ray diffraction patterns of QD films were collected using a Rigaku SmartLab with Cu Kα irradiation.

Glow discharge mass spectrometry (GDMS) measurements of oleate-capped QD powders were performed by Evans Analytical Group using a VG 9000 GDMS instrument (Thermo Scientific). The powder was pressed into high-purity In foil (99.99999%) previously cleaned with acid to remove surface impurities. Impurities in the In foil were analyzed prior to the elemental analysis of each sample. Glow discharge conditions of 1.0 kV, 2.0 mA, and 100 Pa of 99.9999% Ar were used for all measurements. Samples were pre-sputtered for 5 min prior to data acquisition. The intensities of the ion beams were measured using a Faraday cup for lead, selenium, and indium isotopes and a Daly conversion detector for all analytes in the samples. The efficiency of the detectors was calibrated using ¹⁸⁰Ta (relative isotopic abundance of 0.012%) measured using the Daly detector and ¹⁸¹Ta (relative isotopic abundance of 99.99%) measured using the Faraday cup during analysis of pure Ta metal. Scan points per peak were 70 channels, DAC steps of 7, and integration times of 100 and 160 ms for the Daly detector and Faraday cup, respectively.

Secondary ion mass spectrometry (SIMS) of alumina-infilled films on silicon substrates was performed by Evans Analytical Group using a Physical Electronics ADEPT-1010 dynamic SIMS instrument using a 1 keV Cs ion beam for anions (S, O, H, C, P, Si, Al, and Se). Estimated lower detection limits for elements in the alumina overlayer were 1 × 10¹⁹ atoms cm⁻³ for sulfur and hydrogen, 1 × 10¹⁸ atoms cm⁻³ for carbon, and 1 × 10¹⁷ atoms cm⁻³ for phosphorus and silicon. The detection limits for elements in the silicon substrate were 1 × 10¹⁹ atoms cm⁻³ for hydrogen, 5 × 10¹⁷ atoms cm⁻³ for carbon and oxygen, and 1 × 10¹⁷ atoms cm⁻³ for sulfur and phosphorus. Atomic concentrations are accurate to within 20% in the alumina layer (using an Al₂O₃ implant standard) and a factor of two in the PbSe/alumina layer (using a ZnSe implant standard). The depth scale for each layer was quantified by measuring the analysis craters for each of the alumina, PbSe/alumina, and Si layers using a stylus profilometer.

Electrical measurements. All field-effect transistors (FETs) were fabricated on p⁺⁺ (100)-oriented Si substrates coated with a 200 nm thick dry thermal oxide layer (Addison Engineering). Metal source and drain electrodes were patterned onto the substrates by standard photolithography and metal deposition. All data presented in this paper utilized Cr/Au electrodes (5 nm Cr/45 nm Au), but Ti, Ag, and Al electrodes were also tested in unsuccessful attempts to suppress the BSE by modifying the FET substrates. Immediately prior to QD film deposition, substrates were sonicated in detergent and isopropanol, rinsed in isopropanol and water, blown dry, and then O₂ plasma cleaned for 5 minutes (Diener Zepto).

See the supplementary material for Figures S1–S26, including characterization of EDT-, formate-, SCN-, oxalate-, trimesate-, and BDT-capped QD FETs, VG sweep rate dependence for BSE-free devices, electrical data vs temperature and QD size, TEM and other supporting data for the high-mobility devices, substrate modification results, GDMS and SIMS data, results of process modifications, TEM images of the alumina coatings vs thickness, FET data as a function of alumina thickness, results of MNO3 and other cation and anion treatments of partially infilled QD films, results for alternative ALD infills, and examples of persistent hole transients in H2S-dosed and alumina-infilled devices.

J.T. acknowledges support from an NSF Graduate Research Fellowship. J. T. and M.G. were supported by the Center for Advanced Solar Photophysics (CASP), an Energy Frontier Research Center funded by the U.S. Department of Energy (DOE), Office of Science, Office of Basic Energy Sciences (BES). A.A. was supported by the UC Office of the President under the UC Laboratory Fees Research Program Collaborative Research and Training Award No. LFR-17-477148. M.L. was supported by the U.S. Department of Energy, Office of Science, Basic Energy Sciences, under Award No. DE-SC0003904. We thank the UCI School of Physical Sciences Center for Solar Energy. SEM and XRD work were performed at the Laboratory for Electron and X-ray Instrumentation (LEXI) at UC Irvine.

The authors have no conflicts to disclose.

J.T. synthesized the QDs, fabricated the devices, and performed the electrical measurements. M.G. performed FTIR and XRD measurements. A.A. assisted with device fabrication and testing. M.L. conceived and directed the study and assisted with data interpretation. M.L. wrote the manuscript with input from all authors.

Jason Tolentino: Data curation (equal); Formal analysis (equal); Investigation (equal); Writing – review & editing (supporting). Markelle Gibbs: Data curation (equal); Formal analysis (equal); Investigation (equal); Writing – review & editing (supporting). Alex Abelson: Formal analysis (equal); Investigation (equal); Writing – review & editing (supporting). Matt Law: Conceptualization (lead); Data curation (equal); Formal analysis (equal); Funding acquisition (lead); Investigation (equal); Methodology (lead); Project administration (lead); Resources (lead); Supervision (lead); Validation (lead); Visualization (equal); Writing – original draft (lead); Writing – review & editing (lead).

The data that support the findings of this study are available from the corresponding author upon reasonable request.