Perovskite films were spin-coated on hafnium (IV) oxide (HfO2) layers fabricated by atomic layer deposition (ALD) from precursors consisting of methylammonium iodide (MAI) and tin(II) halide (SnX2, X=Cl, Br, I), and then thermally annealed (see more details in Methods). Subsequently, gold source/drain electrodes were deposited, constructing the bottom-gate, top-contact TFTs (Fig. 1a). The TFTs obtained without halide engineering, denoted ‘I-pristine’, exhibited typical p-channel transfer characteristics under continuous mode at room temperature. The representative I-pristine device exhibited a low gate leakage current of ~10−10 A, large Ion/Ioff ratio exceeding 106, and maximum µFE of 1.3 cm2 V−1 s−1 (Fig. 1b). This is the first demonstration of p-channel TFTs based on 3D MASnI3 perovskite films, demonstrating electrical parameters comparable to those of previously reported perovskite TFTs (Supplementary Table 1).
We achieved TFTs with markedly improved performance by carefully engineering the halide compositions of the precursors for perovskite film deposition (Fig. 1b and Supplementary Fig. 1), inspired by the fact that recent breakthroughs in perovskite photovoltaics have been made mostly based on multiple compositions3, 26, 27. Specifically, partially substitution of the iodide source with bromide salt (2 mol%, the devices are denoted ‘I/Br’) resulted in an over three-fold improvement in the µFE of the TFTs (4.3 cm2 V−1 s−1). With the channel films deposited from a precursor with 6 mol% chloride substitution, we obtained devices (denoted as ‘I/Cl’) exhibiting an even higher µFE (9.5 cm2 V−1 s−1). Surprisingly, a rational combination of the two halide engineering strategies, i.e. employing channels deposited from perovskite precursors with simultaneous Br and Cl substitution (2 mol% Br and 6 mol% Cl), led to greatly enhanced TFT (denoted as ‘I/Br/Cl’) performance. As shown in Fig. 1b, the optimised I/Br/Cl perovskite TFTs exhibited a µFE of 19.6 cm2 V−1 s−1 with an Ion/Ioff of 3×107, which is superior to reported Pb- and 2D Sn-based perovskite TFTs (Supplementary Table 1). Textbook-like output curves (IDS versus VDS) with clear linear and saturation currents were observed for all the devices (Supplementary Fig. 2), indicating Ohmic contact between the channel films and electrodes and validating the reliability of mobility extraction28. Furthermore, the I/Br/Cl device operated in an ideal enhancement mode with a VTH of 0 V (Supplementary Fig. 3), suggesting that no applied bias voltage is needed to turn off the transistor, which is highly desirable for simplifying circuit design and minimising power consumption in practical applications29.
In addition to higher mobilities than those of I-pristine, I/Br, and I/Cl devices, the TFTs based on I/Br/Cl perovskite channels also exhibited significantly reduced, even negligible, current–voltage hysteresis. To quantitatively analyse the hysteresis for the I-pristine, I/Br, I/Cl, and I/Br/Cl TFTs, we calculated the difference in VGS (ΔVGS) at |IDS|=10−7 A, halfway between the on and off states30, and presented the data in Fig. 1c. Notably, the I/Br/Cl devices exhibited an average ΔVGS of 0.1 V, which is less than 1/10 of that of the other three types of TFTs, in which ion migration and/or carrier trapping probably occurred (discussed later). The negligible hysteresis for the I/Br/Cl devices is comparable with commercialised amorphous metal oxide TFTs31. Similar to previous perovskite solar cells32, the hysteresis in the dual-sweep transfer curves of the TFTs also causes variations of the extracted performance parameters. We observed notable differences in the maximum mobility values extracted from the reverse (on-to-off, −12 to 7 V) and forward (off-to-on, 7 to −12 V) scans of the I-pristine, I/Br, and I/Cl TFTs (Supplementary Fig. 4), and presented the µFE statistics in Fig. 1d. The I/Cl TFTs demonstrated the largest mobility variations (>70%) because of their largest hysteresis, while the variations for the I-pristine and I/Br devices were slightly lower (50%). Notably, the I/Br/Cl devices exhibited the smallest mobility variations (12%) owing to their negligible hysteresis. Because a strong mobility–hysteresis correlation exists but has been usually neglected in previous research on perovskite TFTs17, we recommend more information on the measurement methods, device hysteresis, and mobilities extracted from both the forward and reverse scans of devices be provided. The hysteresis-free character is highly desired for a wide range of electronic applications, such as logic circuits and backplanes in OLED displays33.
To comprehensively understand the benefits of halide engineering for perovskite TFTs, we performed a series of film characterisations. The scanning electron microscope (SEM) images in Fig. 2a reveal a few pinholes in the I-pristine perovskite film, and slight Br substitution supressed the pinholes. The incorporated Br anions could compete with I anions and coordinate more strongly with the metal ions (Sn2+), modulating the nucleation and crystallisation kinetics of the perovskite films34, 35. In comparison, both the I/Cl and I/Br/Cl films exhibited a considerably smoother surface morphology, suggesting the incorporation of Cl in the precursor significantly promoted perovskite formation, similar to observations about Cl in lead halide perovskites for solar cells34, 36, 37. The results are consistent with X-ray diffraction (XRD) analyses, where the I/Cl and I/Br/Cl samples exhibited substantially increased intensities of the main diffraction peaks compared with those of the I-pristine and I/Br samples, suggesting notably enhanced crystallinity (Fig. 2b).
Interestingly, both the I/Br and I/Br/Cl films exhibited identifiable XRD peak shifts to higher angles compared with that of the I-pristine film. The peak shifts indicate a reduced d-spacing due to the incorporation of smaller Br and/or Cl ions into the I-based perovskite lattices. However, the I/Cl sample showed negligible peak shifts (Supplementary Fig. 5), which suggests that the Cl anions in the I/Cl perovskite film did not enter the perovskite lattice but only functioned to improve the film morphology and crystallinity38. The X-ray photoelectron spectroscopy (XPS) Cl 2p core level spectra further confirmed that Cl was undetectable in the I/Cl films but was successfully incorporated into the triple-halide I/Br/Cl films (Fig. 2c). These results are unsurprising considering the large discrepancy in ionic size between I and Cl anions and the potential volatilisation of Cl additives during film annealing39. However, in the I/Br/Cl perovskite film, the Br-substituted I-based lattice was capable of hosting Cl anions owing to the bridging effect of Br40, 41, enabling the formation of the triple-halide MASn(I/Br/Cl)3 perovskite.
We then conducted Hall-effect measurements to investigate the carrier concentrations and Hall mobilities of the perovskite films, which are important for understanding the performance of the resulting TFTs. As shown in Fig. 2d, the I-pristine films exhibited an average hole concentration of 2.8×1017 cm−3, which gradually decreased to 6.9×1016, 4.5×1016, and 2.2×1015 cm−3 for the I/Br, I/Cl, and I/Br/Cl films, respectively. This trend resulted from decreasing hole sources (tin vacancies). Generally, in 3D Sn-based perovskites, Sn2+ can easily oxidise into Sn4+ even under trace oxygen, and intrinsic tin vacancies (VSn) have a low formation energy, which both cause notoriously high hole concentrations. As revealed by the Sn 3d5/2 XPS spectra in Fig. 2e, the I-pristine sample had a substantial Sn4+ signal at a high binding energy (487.5 eV), which is ~42% of the Sn 3d5/2 peak area. Notably, the peak area of the Sn4+ signal gradually decreased in the I/Br (29%), I/Cl (29%), and I/Br/Cl (20%) samples, indicating suppressed Sn2+ oxidation. Furthermore, previous studies have suggested that the incorporation of anions with electronegativity stronger than that of I− raises the VSn formation energy during perovskite crystallisation, further reducing the hole concentration24, 42. Additionally, for both the I-pristine and I/Br samples, a small shoulder peak appeared at a low binding energy (~485.4 eV), which is ascribed to under-coordinated Sn with an oxidation state of δ < 2+ (Snδ< 2+)43. However, this shoulder peak was undetectable in both the I/Cl and I/Br/Cl samples, indicating well-coordinated Sn sublattices and reduced structural imperfections. The results are in good agreement with the improved quality of the perovskite films deposited from Cl-containing precursors, as demonstrated above, which also rationalises the slightly lower hole concentration of the I/Cl film compared with that of the I/Br film despite their similar degree of Sn2+ oxidation.
The Hall mobilities (µHall) of the perovskite films showed a consistent trend with the µFE extracted from the corresponding TFTs, with the average µHall increasing from 25 cm2 V−1 s−1 (I-pristine) to 57 (I/Br) and 92 cm2 V−1 s−1 (I/Cl), respectively. The µHall for I/Br/Cl films was up to 301 cm2 V−1 s−1. According to the simple Drude model, the hole mobility of a p-type semiconductor is determined by the hole effective mass (mh*) and the average scattering time (τ): 𝜇=q𝜏/\({{m}_{\text{h}}}^{\text{*}}\), where q is the elementary charge12. Considering the small halide substitution in the perovskite lattice, negligible changes to mh* were expected. Therefore, µHall should be mainly determined by the scattering time interval during carrier transport, which is dominated by scattering centres, e.g. ionised (negatively or positively charged) defects and crystal disorders, in the perovskite films. As revealed by the characterisations above, halide engineering effectively enhanced the crystallinity and reduced ionised defects, particularly in the I/Br/Cl film, significantly suppressing charge carrier scattering and providing a rationale for the enhanced µHall of the halide-engineered perovskite films.
In addition to the improved perovskite film quality, the considerably improved performance of the I/Br/Cl devices is related to the properties of the dielectric–perovskite interfaces. Generally, the density of states (Ns) at the interface, which negatively affects the device performance of TFTs, can be estimated from the average subthreshold swing (SS, Supplementary Fig. 6): \(SS=\frac{\kappa Tln10}{e}\left[1+\frac{{e}^{2}}{{C}_{i}}{N}_{S}^{max}\right]\), where \(\kappa\) is the Boltzmann constant, e is the electron charge, and Ci is the areal capacitance of the dielectric layer44. Accordingly, \({N}_{S}^{max}\) of the I-pristine TFTs was calculated to be 1.5×1013 cm−2 eV−1, which was notably reduced to ~3.3×1012 cm−2 eV−1 in the I/Br/Cl TFTs. This suggests that Br and Cl co-substitution greatly enhanced not only the film quality but also the dielectric–perovskite interfaces of the TFTs.
Having elucidated the benefits of rational halide engineering of MASnI3 precursors on the TFT performance, we attempted to gain an in-depth understanding of the hysteresis behaviour of the devices. The commonly observed hysteresis of transistors utilising 3D lead-halide perovskite films is typically attributed to ion migration in the perovskite channel45. However, ion-migration-induced hysteresis is strongly dependent on the sweep rate during device measurement. For example, transistors based on single crystalline MAPbX3 channels exhibited gradually expanded hysteresis when the sweep rate increased from 0.05 to 0.25 V s−1(ref 46). However, we observed negligible changes to the hysteresis in the transfer curves of both the I-pristine and I/Br/Cl devices when the sweep rate increased from 0.4 to 4 V s−1 (Fig. 3a), suggesting that ion migration did not contribute significantly to TFT hysteresis in the MASnI3-based perovskite films. This can be partially explained by the different defect properties of Pb- and Sn-based perovskites27. In p-type MASnI3, iodine defects, e.g. iodine vacancies (VI) and interstitials (Ii), are less dominant (if even present) than VSn47. Consequently, the migration of iodide ions, which have the lowest activation energy and move the most easily, is less significant in MASnI3 than in Pb-based perovskites, greatly reducing the associated electric-field screening effects during TFT operation.
We then considered charge carrier trapping as the primary reason for the hysteresis in our perovskite TFTs, which exhibited a higher current in the transfer curves during the off-to-on sweep than that during the on-to-off sweep. The established models indicate deep electron and hole traps in the semiconductor channels48 are possible causes for this type of hysteresis in TFTs. Theoretical calculations have predicted that in MASnI3, hole traps induced by Ii and VSn defects are shallow, with thermodynamic ionisation levels close to or inside the valance band maximum47. Thus, they are expected to mainly affect the µFE and SS of the TFTs rather than induce hysteresis.48 We postulated that VI defects, which possess the lowest formation energy among possible deep electron traps in MASnI3, were the root cause of the hysteresis in the p-channel TFTs. As shown in the inset of Fig. 3a, when VGS << VTH, negative charge accumulated in the channel, and VI-related long-lifetime electron traps were filled. The trapped electrons shifted the flat-band voltage; that is, the threshold voltage was reduced. When VGS swept towards negative potentials, more holes were induced, leading to a higher drain current. In comparison, the on-to-off sweep started directly from hole accumulation at VGS=−12 V without the influence of stored negative charge48, demonstrating a lower drain current. Therefore, the on-to-off measurement of the perovskite TFTs should more closely resemble the ideal field-effect transistor model, indicating the mobility extracted from the on-to-off transfer curve may be closer to the actual µFE.
With the hypothesis that deep electron traps dominate the hysteresis in our p-channel perovskite TFTs, we investigated the different VI properties of the I-pristine and I/Br/Cl perovskite films to unveil the mechanism that eliminated hysteresis in the optimised I/Br/Cl devices. In the I 3d3/2 core level XPS spectra (Fig. 3b), the peak shifted by 0.5 eV towards higher binding energies for the I-pristine sample compared with that of the I/Br/Cl perovskite. This peak shift was previously ascribed to iodine loss from the perovskite lattice, indicating a higher probability of VI formation in the I-pristine perovskite film. In addition, a noticeable shoulder peak appeared at ~631.7 eV in the I 3d3/2 core level spectra of the I-pristine film, corresponding to the I3− species. These species were assigned to iodide interstitials/VI+ iodine Frenkel defects, which form preferentially under VSn-rich conditions49. The almost invisible shoulder peak for the I/Br/Cl perovskite can be attributed to the reduced iodine loss from the perovskite lattice, along with the significantly reduced hole concentration (VSn defects) revealed by the Hall measurements. Density functional theory (DFT) calculations further confirmed the benefits of passivating VI sites in the I/Br/Cl perovskite film. As shown in Fig. 3c, Br or Cl anions, if successfully incorporated into the MASnI3 perovskite lattice, possess higher binding affinities towards VI sites than that of I anions (slab models in Supplementary Fig. 7), in agreement with recent Pb perovskites with double anions3. Based on the MASn(I/Br)3 perovskite, the calculated binding affinity of a third anion, Cl−, to VI was further enhanced (Figs. 3c and 3d), and hence the VI sites in the I/Br/Cl perovskite were expected to be notably suppressed, rationalising the elimination of hysteresis in the resulting TFTs.
We then characterised the operational stability of our perovskite TFTs, which is another critical figure of merit for practical applications. We first monitored the on/off switching stability of the devices (Fig. 4a). The I-pristine device exhibited a noticeable current decay during the consecutive on/off switching test, while the currents of both the on and off states of the I/Br/Cl device remained consistent. We also examined the device stability under dynamic VGS scans. The transfer characteristics of the I-pristine TFTs gradually shifted, while those of the I/Br/Cl TFTs overlapped completely over 100 cyclic sweeps (Supplementary Fig. 8), suggesting considerably enhanced reliability of the I/Br/Cl devices. To evaluate the stability of the TFTs more rigorously, we performed a bias stress test, during which −12 V was applied constantly, and the shift in VTH was monitored. As shown in Fig. 4b, the VTH of the I-pristine device shifted significantly by −2 V (~17% of the operating voltage) after only 1000 s during the bias test (Supplementary Fig. 9), reflecting the serious carrier trapping in the devices50. Encouragingly, the optimised I/Br/Cl TFT exhibited greatly improved stability with a small threshold voltage shift (ΔVTH) of 0.52 V even after biasing for 12 h, approaching the stability of previously demonstrated stable transistors based on organic and amorphous silicon channels45, 51.
With the reproducible and stable p-channel I/Br/Cl perovskite TFTs (Supplementary Fig. 10), we moved a step further to explore their compatibility with existing TFTs based on n-channel metal oxides for monolithic complementary circuit integration. We fabricated complementary inverters by integrating perovskite TFTs with IGZO TFTs on a single chip. Figure 4c shows an optical image and diagram of the complementary inverter. The standard rail-to-rail voltage transfer characteristics of the inverter at different VDD (Fig. 4d) show VOUT being either 0 V or the supplied VDD, suggesting an ideal logic ‘1’ to ‘0’ transfer. The inverter exhibited a high gain of over 100 at all measured VDD and a peak gain of 140 at VDD=11 V (Fig. 4e), which is significantly higher than that of the wire-connected complementary inverter or CMOS-like inverters involving perovskite TFTs in previous studies24, 52. To the best of our knowledge, this is also the first demonstration of the monolithic integration of a complementary circuit involving perovskite TFTs. Additionally, the noise margin of the inverter was 3.93 V, reaching 72% of the ideal value (VDD/2) (Supplementary Fig. 11), which is sufficient for most static logic applications53.
In conclusion, we have achieved high-performance and hysteresis-free MASnI3-based perovskite TFTs through rational halide engineering. We revealed the benefits of Br and Cl co-substitution in the precursor: enhanced crystallinity and reduced vacancy defects of the perovskite films, which enabled exceptional performance of the resulting TFTs. Moreover, we found ion migration had a negligible contribution to the hysteresis in our p-channel perovskite TFTs based on MASnI3 films. Alternatively, we correlated the hysteresis of our MASnI3 TFT with deep electron traps induced by VI defects, which are notably reduced by rational Br and Cl co-substitution in the precursor. By combining our operationally stable p-channel perovskite TFTs with n-channel IGZO TFTs, we demonstrated monolithically integrated high-gain complementary inverters, suggesting high compatibility and processability for electronic applications.