Solution-processable and photo-programmable logic gate realized by organic non-volatile floating-gate photomemory

Abstract

Programmable inverters using non-volatile floating-gate photomemories as basic building blocks instead of field-effect transistors enable the manipulation of threshold voltage by photons, providing an additional degree of freedom for applications in integrated circuits. However, the development of organic photo-controllable inverters is challenging due to issues such as solubility constraints for film stacking and the immaturity of photo-recordable devices. Notably, the development of organic non-volatile floating-gate photomemories (ONVFGPs) with n-type charge-transporting layers still lags behind that of the p-type layers due to the limited availability of suitable solution-processable charge-trapping materials and charge-transporting material pairs. Herein, photo-crosslinkable polystyrene-b-poly(methacrylic acid) (PS-b-PMAA)/5,10,15,20-tetraphenyl-21H,23H-porphine zinc (ZnTPP), which follows anti-Kasha's rule, is adopted as the charge-trapping layer for ONVFGPs. Both the second and first excited states of ZnTPP participate in photo-induced charge transfer, achieving the state-of-the-art photo-programming time of 0.1 second for ONVFGPs. The transfer curve of the derived photo-programmable inverter can be fine-tuned across a broad spectrum spanning from 405 nm to 830 nm, leading to at least six output states for the same input signal. This research confirms the possibility of integrated organic optoelectronics, opening avenues for solution-processable system-on-chip, neuromorphic computing and organic photonic integrated circuits.

---

New concepts

This work presents the first solution-processable and photo-programmable inverter, integrating a p-type organic field-effect transistor (OFET) and an organic non-volatile floating-gate photomemory (ONVFGP) with an n-type channel fabricated through photolithography. Unlike previous approaches that relied on thermal evaporation or mechanical exfoliation for n-type channel fabrication, this study introduces 1,11-diazido-3,6,9-trioxaundecane as a photo-crosslinker, enabling the formation of a robust, solution-processed, and photolithographically patterned inverter. This novel configuration significantly advances in-memory computing by overcoming longstanding solubility challenges in organic material systems. Furthermore, it uniquely leverages visible and near-infrared (NIR) light to fine-tune the conductance of the n-type channel, creating photo-programmable logic gates that serve as foundational elements for photo-sensitive in-memory computation. By merging the principles of optoelectronics and in-memory computing, this work offers transformative insights into energy-efficient hardware design for data-intensive applications. It paves the way for new functionalities in adaptive logic systems, expanding the horizon of materials science research and its applications in emerging technologies.

1. Introduction

Emerging data-intensive and adaptive logic applications, such as self-driving vehicles, smart industries, and telesurgery, demand highly energy-efficient hardware. However, traditional von Neumann architecture, with its significant latency and energy consumption because of data movement between processing and storage units, falls short of meeting these demands. Consequently, there has been a surge in the development of next-generation architectures that blur the boundary between processing and memory units, such as near-memory computing and three-dimensional monolithic integration techniques.^1,2 In-memory computing, which emulates the working principles of the biological brain by using neurons as the center of both logic operations and data storage, is emerging as a promising hardware configuration for addressing the challenges posed by data-intensive and adaptive logic applications.^1–11 However, organic memory devices for in-memory computing applications remain exclusive because of solubility issues during the fabrication of crossbar memory arrays.

Organic memory devices with photo-recordable characteristics, namely organic non-volatile floating-gate photomemories (ONVFGPs), are promising candidates for alleviating energy consumption and signal interference in conventional electrically driven memory devices. The mechanism of ONVFGPs resembles that of a floating-gate transistor-type memory, wherein charge carriers are trapped within a floating gate driven by photon energy rather than an electric field. When a photo-induced exciton occurs in the photoactive material, electrons and holes will be separated between the floating gate and the charge-transporting channel due to a discrepancy in the highest occupied molecular orbital (HOMO) and the lowest unoccupied molecular orbital (LUMO) levels of the two materials. The success of this strategy hinges heavily on identifying an ideal material system capable of fully exploiting the potential of this configuration. Nevertheless, ONVFGPs with n-type channels continue to lag behind their p-type counterparts due to solubility challenges encountered during film stacking.^12–17 Consequently, channel fabrication in NVFGPs with n-type channels typically relies on thermal evaporation for small organic molecules^5,12,14 and mechanically exfoliated 2D inorganic materials.^{4,6–8,13,18,19}

Organic materials, known for their facile modification of chemical structure and solution processability, have sparked a revolution in flexible optoelectronics, including organic field-effect transistors (OFETs), organic memory devices, and organic photovoltaics. However, the solubility issue during film stacking has hindered the advancement of organic material-based integrated circuits compared to their inorganic counterparts. To address this challenge, bis-azide or bis-diazirine-based photo-crosslinkers have been introduced into conjugated polymers during the photolithography process.^20–27 Those enable the formation of photo-crosslinked conjugated polymers through nitrene or carbene insertion into the alkyl side chain of the conjugated polymers under UV light illumination, respectively. Despite these advancements, the introduction of photo-crosslinkers in ONVFGPs for photo-programmable inverters remains unexplored.

Herein, a photo-crosslinker, 1,11-diazido-3,6,9-trioxaundecane(azide), is introduced in the fabrication of an organic photo-programmable inverter, comprising a p-type OFET and a ONVFGP with a n-type channel through photolithography to enable the first solution-processable and photo-programmable inverter, where poly{[N,N′-bis(2-octyldodecyl)naphthalene-1,4,5,8-bis(dicarboximide)-2,6-diyl]-alt-5,5′-(2,2′-bithiophene)} (N2200) served as the charge-transporting layer and polystyrene-b-poly(methacrylic acid) (PS-b-PMAA)/5,10,15,20-tetraphenyl-21H,23H-porphine zinc (ZnTPP) served as the charge-trapping layer in ONVFGP. The conductance of the n-type channel in ONVFGP can be fine-tuned by visible and near-infrared (NIR) light with wavelengths ranging from 405 nm to 830 nm. This will enable the future use of photo-programmable logic gates as building blocks for photosensitive in-memory computation and organic photonic integrated circuits.

2. Results

2.1 Fabrication of photopatterned and photo-programmable logic gate

The schematic images and the corresponding optical images of the photopatterned and photo-programmable inverter are shown in Fig. 1(a), where the photo-programmable logic gate consists of a p-type OFET and ONVFGP with an n-type channel, enabling photo-programmable characteristics. The chemical structures of the materials used are shown in Fig. 1(b). The studied logic gate contains a stack composed of Au serving as electrodes, p-type DPP-DTT and n-type N2200 as the charge-transporting layer, PS-b-PMAA/ZnTPP (BCP/ZnTPP) as the charge-trapping layer, and 300 nm-thick SiO₂/highly doped Si wafer as the gate dielectric material and the bottom-gate electrode, respectively. For film stacking and patterning of the above functional thin film, a bis-azide based photo-crosslinker, 1,11-diazido-3,6,9-trioxaundecane, named as azide, was employed and blended with the above organic functional materials, which generates nitrene under UV light illumination (254 nm) and then joints two alkyl chains through nitrene insertion into the C–H bond (Fig. 1(c)), which enable crosslinked networks with an antisolvent feature for the above functional layers (Fig. 1(d)).

Fig. 1 Fabrication of a photopatterned and photo-programmable logic gate. (a) The schematic image and the photographic image of a photopatterned and photo-programmable logic gate consisting of a p-type field-effect transistor and an n-type ONVFGP. (b) The chemical structures of the materials. (c) The mechanism of the photo-crosslinking reaction (d) The schematic image of the crosslinked networks for the above functional layers. (e) The absorption spectra of N2200, BCP/ZnTPP and the corresponding BCP/ZnTPP/N2200 film. (f) UPS spectra of N2200 and BCP/ZnTPP film. (g) Energy-level diagram of the functional materials in ONVFGP.

The schematic fabrication procedures for the p-type semiconductor, n-type semiconductor, and charge-trapping layer by the photopatterning process using azide are summarized in Fig. S1 (ESI†). A typical logic gate was fabricated in a bottom-gate top-contact (BGTC) configuration on a SiO₂/Si wafer substrate. The first step involved patterning the BCP/ZnTPP charge-trapping layer. This was done by spin-coating a solution mixture of BCP/ZnTPP and azide onto the SiO₂/Si wafer substrate, followed by selective exposure to UV light (254 nm, 1.47 mW cm⁻²) through a mask. The second step involved patterning the p-type polymer film, achieved by spin-casting a solution mixture of the p-type polymer and azide, followed by UV exposure through a mask to define the desired area and subsequent washing to remove the un-crosslinked polymer regions. In the third step, the n-type polymer film was photopatterned using the same method as in the second step. The consecutive application of the photopatterning process was possible due to the structural robustness of the underlying crosslinked films. Finally, a bottom-gate, top-contact configuration was defined by 50 nm-thick thermally evaporated gold electrodes. This study first demonstrates the fabrication of a photopatterned, solution-processed, and photo-programmable logic gate achieved through a single processing protocol—photopatterning using the photo-crosslinker for different functional components.

Our previous reports have indicated that well-isolated photoactive materials in microphase-separated block copolymers can inhibit the dissipation of trapped charge carriers in photomemory, thereby ensuring long-term memory behaviors compared to homopolymer systems.^28–34 To achieve this goal, the block copolymer PS-b-PMAA is utilized, where the oxygen atom from the C=O group in the hydrophilic PMAA can chelate with the Zn atom in ZnTPP, forming a PMAA/ZnTPP moiety. To validate this assumption, solution ¹H nuclear magnetic resonance (¹H NMR) spectra were obtained for zinc tetraphenylporphyrin (ZnTPP), poly(methacrylic acid) (PMAA), PS-b-PMAA and PS-b-PMAA/ZnTPP (with MAA : ZnTPP molar ratio of 1 : 1) in N,N-dimethylformamide-d₇ (DMF-d₇), as shown in Fig. S2a (ESI†). In all the ¹H NMR spectra, the proton signals at 8.02, 2.92 and 2.75 ppm belong to DMF-d₇, with a signal at 3.46 ppm assigned to H₂O due to the hydrophilic nature of DMF. For ZnTPP (Fig. S2b, ESI†), the resonance absorptions of pyrrole protons, o-protons, and m- and p-protons are located at 8.89 ppm, 8.29 ppm and 7.84 ppm, respectively, which could be found in PS-b-PMAA/ZnTPP (Fig. S2c, ESI†) as well, confirming the existence of ZnTPP in PS-b-PMAA/ZnTPP. Upon adding ZnTPP to the PS-b-PMAA solution, a slight de-shielded shift in the resonance absorptions of methyl protons is observed (Fig. S2f, ESI†), which confirms the strong electron-withdrawing effect of ZnTPP on the carboxylic group of PMAA. Besides, due to the deshielding phenomenon resulting from the chelating interaction between PS-b-PMAA and ZnTPP, the original merged methyl protons from MAA in PS-b-PMAA are distinguishable and labeled at 1.10 ppm and 1.09 ppm in the ¹H NMR spectrum of the PS-b-PMAA/ZnTPP solution. The above solution ¹H NMR spectra confirm the chelating interaction between PS-b-PMAA and ZnTPP in solution. Fourier transform infrared (FTIR) spectra also demonstrate a shift of the C=O stretching signal from 1705 cm⁻¹ to 1695 cm⁻¹ after the BCP was blended with ZnTPP (Fig. S3a, ESI†). The UV-visible absorption spectrum of BCP/ZnTPP exhibits two characteristic electronic transitions centered at 436 nm and 555 nm, which are assigned as the Soret band (S₀ → S₂) and Q-band (S₀ → S₁) of ZnTPP, respectively.^35–37 It is noted from the UV-visible absorption spectrum that the intense peak centered at 441 nm belonging to the Soret band of ZnTPP in pure ZnTPP film is blue-shifted to 436 nm in the BCP/ZnTPP film, which may indicate that the isolation of ZnTPP by BCP inhibits the delocalization of the conjugated porphyrin network (Fig. S3b, ESI†). Fig. 1(e) demonstrates the absorption spectra of N2200, BCP/ZnTPP and the corresponding N2200/BCP/ZnTPP films. The broad absorption spectrum of the N2200 film from visible light to the NIR region, with an onset wavelength of ∼855 nm, enables broadband photoresponsivity for the derived ONVFGP. The optical absorption analysis reveals approximate bandgap values of 1.45 eV for N2200 and 2.01 eV for ZnTPP. Additionally, the findings from ultraviolet photoelectron spectroscopy (UPS) (Fig. 1(f)) indicate the highest occupied molecular orbitals (HOMOs) of N2200 and ZnTPP to be at −5.26 eV and −4.82 eV, respectively. By integrating these data with the corresponding optical band gaps, the estimated lowest unoccupied molecular orbitals (LUMOs) are determined to be −3.81 eV for N2200 and −2.81 eV for ZnTPP. These energy-level assessments for the N2200 and ZnTPP films are consolidated in Fig. 1(g). Notably, ZnTPP exhibits elevated HOMO and LUMO levels compared to those of the N2200 film, facilitating photo-induced charge transfer between N2200 and ZnTPP. It is worth noting that ZnTPP is known as its triplet–triplet annihilation upconversion (TTA-UC), 2T₁ → S₂ + S₀, which enables the generation of a high-energy S₂ state to trigger optoelectrical reactions such as photo-thermal therapy and photoinitiated polymerization.^35–37 The feasibility of such excited-state electron transfers further elevates the driving force for photoinitiated electron transfer from ZnTPP to N2200 in ONVFGP.

2.2 Characterization of functional materials photopatterned using azide

To evaluate the initial dose of UV light, the conjugated polymer-to-azide ratio was fixed at 1 : 0.1 by weight. The photochemical reaction of the azide crosslinker inside the N2200 and DPP-DTT films was confirmed by FTIR spectroscopy, with the spectra ranging from 400 cm⁻¹ to 4000 cm⁻¹ (Fig. 2). For the N2200 system (Fig. 2(a)–(c)), compared to the pure N2200 thin film, the characteristic vibration peak of the azide group located at 2105 cm⁻¹ assigned to (the asymmetric stretching ν_as) of N₃ can be specified in the spectrum of N2200 and azide blend film. The intensity of the characteristic vibration peak of the azide groups decreases gradually by increasing the UV light exposure time from 0 min to 30 min, confirming the photoreaction of the azide group. Accordingly, the expected main photo-crosslinking reaction is the insertion of nitrenes into the nonconjugated C–H bonds, as evidenced by the reduction of the CH₂ vibration (symmetric (ν_s) at 2853 cm⁻¹; ν_as at 2925 cm⁻¹) and CH₃ vibration (ν_s at 2871 cm⁻¹ and ν_as at 2954 cm⁻¹). Similar phenomena can be observed in the DPP-DTT system, as shown in Fig. 2(d)–(f). With increasing the UV illumination time from 0 min, 10 min, 20 min to 30 min, the N₃ signal at 2109 cm⁻¹ cannot be eventually distinguished (Fig. 2(e)) and the correspondingly diminished signals from both CH₂ vibration and CH₃ vibration (Fig. 2(f)), demonstrating the universal strategy for the crosslinked network using azide as photo-crosslinker. Based on the absorbance of the N₃ stretching signal in the FTIR spectra of N2200 and azide blend film, as well as the DPP-DTT and azide blend film, the nitrene-forming efficiency can be evaluated. (Table S1, ESI†). It reaches a maximum at an exposure time of 30 min with values of 52.47% and 86.8% for the N2200 and DPP-DTT systems, respectively.

Fig. 2 FTIR characterization of the conjugated polymer and azide blend film. (a) FTIR spectra of N2200 and azide blend films under various UV light exposure times. Enlarged spectra (b) from 2000 to 2200 cm⁻¹ and (c) from 2600 to 3300 cm⁻¹. (d) FTIR spectra of DPP-DTT and azide blend films under various UV light exposure times. Enlarged spectra (e) from 2000 to 2200 cm⁻¹ and (f) 2600 to 3300 cm⁻¹.

To confirm the integrity of the above crosslinking strategy, film retention was examined using UV-visible absorption spectroscopy and grazing-incidence wide-angle X-ray scattering (GIWAXS). Fig. 3(a) displays the absorption spectra of the N2200 and azide blend film after exposure to UV light for different periods. The absorption spectrum of the pure N2200 film is also provided for comparison. Except for the pure N2200, each sample was developed in chloroform before recording the absorption spectrum. The film retention is defined by the ratio of the characteristic absorption peak at 700 nm between the blend film and pure N2200. The characteristic absorption peak at 700 nm of the N2200 and azide blend film shows a tendency to increase with the prolonged UV light irradiation time and eventually saturates at 0.91 after a UV light exposure of 20 min, corresponding to a dose of 2.4 J cm⁻² (Fig. 3(b)). Film retention for DPP-DTT and azide blend film also approaches saturation after a continuous illumination time of 20 min when probed at a wavelength of 820 nm. (Fig. 3(c) and (d)). It is worth noting that, despite a part of the non-crosslinked conjugated polymer being etched away during chloroform development, the electrical performance of the derived organic OFETs is comparable to that of the pure conjugated polymer-derived one, as discussed below. Under the same dose of UV light illumination, high-resolution photopatterned conjugated polymer with a feature line width (W) down to 5 μm (Fig. 3(e)), confirmed by atomic force microscope (AFM) (Fig. 3(f)) and the line-cut profile (Fig. 3(g)), can be realized using the nitrene–insertion strategy in the photolithography process. Semiconducting polymers containing patterns with large-scale dimensions were also successfully obtained (Fig. S4, ESI†).

Fig. 3 (a) Absorption spectra of N2200 and azide blend film under different illumination times and (b) the corresponding film retention. (c) Absorption spectra of DPP-DTT and azide blend film under different illumination times and (d) the corresponding film retention. Optical images of photopatterned DPP-DTT films with various patterned topologies. (e) Optical images of feature line widths of 10 μm and 5 μm. (f) AFM height image of the photopatterned N2200 film with a line gap of 5 μm and (g) the corresponding line-cut profile.

The morphology can be probed by GIWAXS to confirm the crystalline structure of the conjugated polymer before and after the photo-crosslinking reaction with azide, as demonstrated in Fig. 4. With the prolonged illumination time, the scattering signal from the ordered lamellar stacking and π–π stacking labeled as (h00) and (001), respectively, in both the in-plane (Fig. 4(i)) and the out-of-plane (Fig. 4(j)) directions show a tendency to increase for the N2200 and azide blend film and gradually saturates over 20 min. The higher intensity of the (h00) signals in the in-plane direction compared to the out-of-plane direction indicates that the photo-crosslinked N2200 is oriented dominantly in a face-on structure, similar to pure N2200. On the other hand, the GIWAXS image recorded from the DPP-DTT and azide blend films shows increasingly ordered lamellar peaks in the out-of-plane direction and a π–π stacking peak in the in-plane direction over the studied period, demonstrating that the photo-crosslinked DPP-DTT adopts an edge-on orientation. It is worth noting that the consistent scattering vector of lamellar stacking and crystalline orientation, both in the pure conjugated polymer and blended film, demonstrates that the nitrene insertion on that alkyl side chain does not alter the morphology of conjugated polymers.

Fig. 4 GIWAXS characterization of crosslinked N2200 and DPP-DTT. (a) pure N2200. N2200 and azide blend film under UV light illumination of (b) 10 min (c) 20 min (d) 30 min. (e) pure DPP-DTT. DPP-DTT and azide blend film under UV light illumination of (f) 10 min (g) 20 min (h) 30 min. The line-cut diffraction profiles of the diffractograms for the N2200 system films along the (i) in-plane direction and (j) out-of-plane direction. The line-cut diffraction profiles of the diffractograms for DPP-DTT system films along the (k) in-plane direction and (l) out-of-plane direction.

2.3 Electrical characteristics of photopatterned devices

Before the fabrication of the photopatterned conjugated polymer film, the stability of pure N2200 and DPP-DTT under UV light was investigated. As shown in Fig. S5 (ESI†), the electrical characteristics of the pure conjugated polymer-derived OFETs do not fluctuate significantly under UV light exposure (254 nm, 2.4 J cm⁻²), which implies that the high energy of the UV light does not destroy the conjugated structure. To optimize the electrical performance of the photo-crosslinked conjugated polymer-derived OFETs, the charge-transporting layer in OFETs with varying conjugated polymer-to-azide ratios was investigated. As depicted in Fig. S6 (ESI†) and summarized in Fig. 5(a), the optimal N2200 to azide ratio is 1 : 0.1, which shows the minimum azide content while providing an average electron mobility of 1.05 × 10⁻³ cm² V⁻¹ s⁻¹. For the DPP-DTT series, the same conjugated polymer-to-azide ratio of 1 : 0.1 shows an average mobility of 9.37 × 10⁻³ cm² V⁻¹ s⁻¹ (Fig. S6, ESI† and Fig. 5(b)). Therefore, the conjugated polymer-to-azide ratio is fixed at 1 : 0.1 for the following optoelectronics fabrication. It should be mentioned that conventional treatments to enhance mobility, such as thermal annealing, are omitted here to prevent the morphological variation of the charge-trapping layer beneath the charge-transporting layer in the top-contact and bottom-gate ONVFGP. The output curves of the optimized photopatterned n-type and p-type derived OFETs exhibit typical OFET characteristics, exhibiting linear behavior at low V_DS and saturation behavior at high V_DS (Fig. 5(c) and (d)).

Fig. 5 Electrical characteristics of photo-patterned OFET. (a) Electron mobility of photo-patterned n-type OFET with different N2200 to azide ratios from 1 : 0.01 to 1 : 0.2. (b) Hole mobility of photopatterned p-type OFET with different DPP-DTT to azide ratios from 1 : 0.01 to 1 : 0.2. Output characteristics of optimized photopatterned (c) n-type and (d) p-type OFET. (e) Circuit diagram and OM image of the inverter. (f) Output curves of the inverter and (g) the inverter voltage gain both as functions of the input voltage.

The fabrication of all-photopatterned n-type and p-type OFETs with reliable operation enabled the integration of these devices into functional logic circuits. The circuit diagram and the corresponding optical microscopy (OM) image of the NOT logic gate are presented in Fig. 5(e). The NOT gate was constructed by connecting the n-type and p-type OFETs in series. The voltage transfer characteristics of the NOT gate at various supply voltages (V_DD) of 30, 40, 60, 80, and 100 V are shown in Fig. 5(f). A signal inversion was clearly observed, as an increase in the input voltage (V_IN) resulted in a decrease in the output voltage (V_OUT) to 0 V (logic state “0”). The signal inverter gain of the NOT gate, defined as |dV_OUT/dV_IN|, was approximately 14.9, with a switching voltage of 53 V at a supply voltage of 100 V (Fig. 5(g)).

2.4 Characterization of the photo-crosslinked BCP/ZnTPP photomemory

ONVFGM is a transistor-type memory that consists of a OFET and the charge-trapping layer underneath a charge-transporting layer. Developing an antisolvent and patternable charge-trapping layer, therefore, is the next crucial step in realizing a photo-programmable inverter. To confirm the photo-crosslinkable features of BCP, the BCP and azide blend film with a BCP-to-azide weight ratio of 1 : 4, before and after UV light illumination, were analyzed by FTIR, as depicted in Fig. 6(a). The decreased asymmetric stretching of N₃ at 2107 cm⁻¹ and the reduction of the CH₂ vibration (ν_s at 2870 cm⁻¹ and ν_as at 2922 cm⁻¹) after a 30-min illumination confirm that the nitrene insertion into the nonconjugated C–H bonds of BCP can also be realized. Based on the absorbance of the N₃ stretching signal in the FTIR spectra, the nitrene-forming efficiency can be evaluated with a value of 49.77%. (Table S1, ESI†).

Fig. 6 (a) FTIR of BCP and azide blend film. (b) AFM height image of photopatterned BCP film with a line gap of 2 μm and the corresponding line-cut profile. (c) AFM height image of photo-crosslinked BCP and (d) photo-crosslinked BCP/ZnTPP. (e) The corresponding GISAXS 1D profile of photo-crosslinked BCP and photo-crosslinked BCP/ZnTPP. (f) TEM image of photo-crosslinked BCP/ZnTPP. (g) PL of photo-crosslinked BCP/ZnTPP and photo-crosslinked BCP/ZnTPP/N2200. (h) Normalized TRPL decay probed at 391 nm for BCP/ZnTPP before and after contact with N2200. (i) Normalized TRPL decay probed at 643 nm for BCP/ZnTPP before and after contact with N2200.

To explore the main block for constructing the crosslinked network, homopolymers, PS and PMAA, were blended with azide and probed by FTIR (Fig. S6, ESI†). It is interesting to note that although both homopolymer blend films exhibit lower absorbance of the asymmetric stretching of N₃ and CH₂ vibration after 30 min of UV light illumination, the drop in the PS system (Fig. S7b and c, ESI†) is more prominent compared to the PMAA system (Fig. S7e and f, ESI†). This is believed to stem from the fact that the hydrophilic PMAA can stabilize the highly polar organic azide, thus reducing the generation of nitrene under UV light illumination. It should be mentioned that the high azide content for a high crosslinking degree may stem from the fact that the self-assembled BCP provides the template to selectively control the spatial location of the hydrophilic azide molecules in the PMAA block and thus retards the interaction between the azide and the PS block.

To pursue the optimized BCP-to-azide ratio for the best film retention, the BCP-to-azide weight ratio ranging from 1 : 1, 1 : 2 to 1 : 4 was evaluated. The film thickness of the as-spun film and developed film in chloroform was proved by AFM, as summarized in Table S2 (ESI†). The consistent film thickness for both the as-deposited film and developed film in all ratios implies that the azide-triggered photolithography can also be realized in PS-b-PMAA. However, the higher root-mean-square roughness (R_q) on the developed film, in the ratios of 1 : 1 and 1 : 2, with values of 1.43 nm and 2.89 nm, respectively, was observed, which may stem from the films with an insufficient crosslinking agent not achieving full insolubility against the developing solvent. Given that the film roughness plays a pivotal role in controlling the quality of the above solution-processable film, i.e., charge-transporting material, a BCP-to-azide ratio of 1 : 4 with an R_q of 0.91 nm was adopted in the following process. A negative BCP pattern after UV light exposure and subsequent development in chloroform with feature sizes down to 2 μm is demonstrated in Fig. 6(b). The micro-phase separation between two moieties in the block copolymer is the key feature that enables the isolation of a photoactive floating gate in the charge-trapping layer. After being illuminated by UV light and chloroform development, hexagonally packed cylinders can be observed, even at a high BCP-to-azide weight ratio of 1 : 4, as shown in Fig. 6(c). The photo-patternable and microphase-separated film ensures the subsequent ZnTPP incubation as the photoactive floating gate. As shown in Fig. 6(d), the AFM height image of the photo-patternable BCP/ZnTPP with the PMAA-to-ZnTPP molar ratio of 1 : 0.5 demonstrates similar hexagonally packed cylinders while displaying loose cylinder packing, which is confirmed by the diffraction pattern from grazing-incidence small-angle X-ray scattering (GISAXS) with the in-plane scattering vector (q_y) shifted from 0.013 Å⁻¹ to 0.011 Å⁻¹, corresponding to the d spacing shift from 46.9 nm to 56.1 nm (Fig. 6(e)). Such a q shift can be inferred from the fact that the incubation of ZnTPP in the PMAA phase enlarges the volume ratio of the PMAA phase to the PS phase. To evaluate the content of ZnTPP in the composite film, UV-visible absorption spectra of the photo-crosslinked BCP/ZnTPP before and after development were probed. As illustrated in Fig. S8 (ESI†), the ZnTPP retention ratio was determined by dividing the maximum absorbance (at a wavelength of 436 nm) of the composite film after development by that of the film before development, indicating that 79% of ZnTPP remains within the composite film. As shown in Fig. S7 (ESI†), the antisolvent feature of the photo-crosslinked BCP stems from the crosslinked PS moiety in PS-b-PMAA. Therefore, the swelling effect on the PMAA/ZnTPP moiety may result in the loss of ZnTPP during development. Fig. 6(f) demonstrates the corresponding TEM image of photo-patternable BCP/ZnTPP, where the white hexagonally packed cylinders dispersed in a dark matrix can be inferred to indicate that the hydrophobic PS cylinders are repelled from the hydrophilic matrix consisting of PMAA, ZnTPP and azide moieties. The photo-induced charge-carrier transfer between ZnTPP and N2200 can be probed by the photoluminescence quenching of BCP/ZnTPP before and after spin-coating of the N2200 film on top of the BCP/ZnTPP film, as depicted in Fig. 6(g). Under a 372 nm laser excitation, both Soret band emission (S₂ → S₀) centered at 391 nm and Q-band emission (S₁ → S₀) centered at 643 nm of ZnTPP display scarcely detectable intensity after the BCP/ZnTPP is in contact with the N2200 film. It is worth noting that the anti-Kash's rule, the emission of a high-lying excited-state S₂, can further elevate the energy-level offset between ZnTPP and N2200, thereby providing sufficient driving force for more efficient exciton dissociation. The dynamics of a low-lying excited S₁ and high-lying excited-state S₂ before and after the BCP/ZnTPP contact with the N2200 film were probed by time-resolved photoluminescence (TRPL) spectra, as demonstrated in Fig. 6(h) and (g), respectively. The transient PL decay can be described using the biexponential fitting curve shown in eqn (1), where τ₁ and τ₂ represent the short-lived and long-lived exciton lifetimes, respectively. The intensity-averaged lifetime (τ_average) was calculated using eqn (2). The charge transfer efficiency (CTE) was evaluated using eqn (3), where τ_BCP/ZnTPP and τ_{BCP/ZnTPP/N2200} are the average exciton lifetimes of the photopatterned BCP/ZnTPP composite film before and after contacting N2200, respectively.

All results are summarized in Table S3, ESI.† The τ_average of S₂ fluorescence drops from 0.45 ns to 0.15 ns with a CTE of 67%, while the τ_average of S₁ fluorescence decreases from 0.74 ns to 0.17 ns with a CTE of 77%. Both the PL and transient PL spectra confirm the photo-induced charge transfer, indicating ZnTPP and N2200 as a decent donor–acceptor pair in ONVFGP.

The photo-recordable behavior under light illumination (60 mW cm⁻², 240 s) with wavelengths ranging from 405 nm to 830 nm is confirmed by a sweeping gate voltage (V_GS) from −20 V to 100 V at a fixed drain voltage (V_DS) of 100 V. As shown in Fig. 7(a), the threshold voltage (V_TH) is negatively shifted under 120 s-light illumination, leading to the memory window increasing from 3.96 V, 4.78 V, 6.32 V, 12.21 V to 20.21 V, with the wavelength decreasing from 830 nm, 650 nm, 525 nm, 450 nm to 405 nm (Fig. S9a, ESI†). The temporal I_DS curve reading at V_GS = 0 V further reveals the kinetic behavior of charge transfer between ZnTPP and N2200, as shown in Fig. 7(b), where I_DS increases gradually and then saturates under a continuous light illumination of 240 s. The distinct absorbance characteristics of BCP/ZnTPP/N2200 across the visible-to-NIR spectrum enable the wavelength-specific behavior with the I_ON/I_OFF ratios of 2.83 × 10⁴, 6.59 × 10³, 2.41 × 10³, 2.79 × 10² and 6.23 × 10¹ for 405 nm, 450 nm, 525 nm, 650 nm and 830 nm, respectively. Operation at a low lateral voltage is achievable, as demonstrated in Fig. S9b (ESI†), where an I_ON/I_OFF ratio of 2.53 × 10² is obtained even at a V_DS of 0.5 V. It is worth noting that the negligible V_TH shift and transient photocurrent observed after a 405 nm-light illumination for the BCP/N2200-based OFETs confirm that the photo-recordable behavior arises from the charge transfer between ZnTPP and N2200 (Fig. S9c and d, ESI†).

Fig. 7 (a) Transfer characteristics of the BCP/ZnTPP/N2200-based photomemory at V_DS = 100 V under different light illumination for 240 s (60 mW cm⁻²). (b) The photo-responsive current toward time under different illumination (240 s, 60 mW cm⁻²) at V_DS = 100 V (c) Temporal I_DS curves of the BCP/ZnTPP/N2200 photomemory at V_DS = 100 V with varying illumination intensities from 60 μW cm⁻² to 60 mW cm⁻². (d) I_ON/I_OFF and photoresponsivity toward light-intensity diagram. (e) Temporal I_DS curves of the studied photomemory at V_DS = 100 V with different illuminating times ranging from 0.1 s to 240 s (405 nm, 60 mW cm⁻²). (f) I_ON/I_OFF toward light programming time diagram. (g) Consecutive 10 cycles of optical writing–reading–electrical erasing–reading test. (h) Multilevel behavior of the studied photomemory and (i) a magnified view of the region from 1500 s to 1920 s with red lines indicating the occurrence of light pulses. (j) The plausible mechanism of photomemory.

A light-intensity distinguishable behavior with 405 nm-light intensity ranging from 60 μW cm⁻² to 60 mW cm⁻² can be realized (Fig. 7(c)). As summarized in Fig. 7(d), the minimum recordable light intensity of 60 μW cm⁻² with an on/off current ratio of 4.6 and maximum photoresponsivity of 1.25 mA W⁻¹ can be achieved. Various programming times were measured, ranging from 0.1 s to 240 s, under a 405 nm-light source with an intensity of 60 mW cm⁻² (Fig. 7(e)). The device displays power-distinguishable characteristics, with a decent on/off current ratio of 1.55 upon a short illumination duration of 0.1 s, as summarized in Fig. 7(f). Compared to the state-of-the-art organic photomemories based on n-type charge-transporting materials (Table S4, ESI†), the BCP/ZnTPP/N2200-derived photomemory represents the first solution-processable and photo-patternable photomemory, achieving the lowest photo-programming time of 0.1 s. Consecutive 10 cycles of optical writing–reading–electrical erasing–reading (WRER) test was performed to examine the switching behavior of the studied photomemory, as depicted in Fig. 7(g), where a cycle test consists of photo-gating (405 nm, 60 mW cm⁻²) for 30 s, reading at V_DS of 5 V for 40 s, electrical-erasing (V_GS of 60 V for 15 s) and then reading at V_DS of 5 V for 40 s. The reliable and distinguishable on and off currents manifest the potential practical application in in-memory computation. Owing to the short photo-programming time, the studied photomemory shows multilevel behavior, which is manipulated by consecutively applying a 405 nm pulse train with a pulse width of 0.7 s and intensity of 60 mW cm⁻² every 30 s at a fixed V_DS = 100 V (Fig. 7(h)). By zooming the time interval ranging from 1500 s to 2000 s (Fig. 7(i)), the clearly increased and step-like photocurrent ensures that 6 bit (64 levels) can be achieved in a single cell. From the above analyses of photophysical and optoelectrical properties, the plausible mechanism for photo-recordable behavior is provided and portrayed in Fig. 7(j), where both S₂ and S₁ states participate in the photo-induced electron transfer from ZnTPP to N2200 and thus enable a hole trapped in ZnTPP and switch the N2200 to a high-conductive state. The initial low-conductive state of N2200 can be recovered through charge recombination by applying positive V_GS. On the other hand, photo-recordable behavior can also be achieved with a light source having a wavelength spanning from 640 nm to 830 nm due to the low band gap of N2200. A similar photo-recordable mechanism triggered by a longer wavelength is depicted in Fig. S10 (ESI†).

The fabricated photomemory can also be applied to neural network simulations based on their tunable conductance (G) values, which correspond to synaptic weight updates in response to learning and forgetting experiences. A three-layer fully-connected multilayer perceptron (MLP) artificial neural network (ANN) was simulated for supervised recognition tasks, as illustrated in Fig. 8(a). The network was trained via backpropagation on three categories of datasets: the Modified National Institute of Standards and Technology (MNIST) dataset, a small handwritten digit dataset, and the Sandia cyberfile format. The ANN architecture comprised input neurons (corresponding to image pixels), a configurable number (N) of hidden neurons, and 10 output neurons (representing digits 0 to 9). Herein, learning, specifically long-term potentiation (LTP), is emulated using optical pulses (utilizing a 405 nm laser pulse train with an intensity of 60 mW cm⁻² and a pulse duration of 0.7 s every 30 s). Conversely, forgetting, or long-term depression (LTD), is simulated using a positive V_GS pulse train (e.g., 1 V amplitude with a pulse duration of 1 s for every 20 s). Fig. 8(b) demonstrates excellent LTP behavior characterized by a very low nonlinearity (NL of only 0.01) and a substantial dynamic range (DR of 380). These characteristics are advantageous for image recognition tasks and can compensate for the suboptimal NL value of the LTD process. The system exhibits high recognition accuracy across various image sizes: above 90% for both small (8 × 8 pixels) and large (28 × 28 pixels) images (Fig. 8(c) and (e)) and approximately 85% for lower-resolution cyber images (16 × 16 pixels) (Fig. 8(d)). The training accuracy of 94.66% achieved in our study is comparable to that of the state-of-the-art organic transistor-based neuromorphic devices (Table S5, ESI†). Additionally, after three months of storage in a nitrogen-filled glove box, the recognition accuracies for small, cyber, and large images remained at 92.43% ± 1.73%, 90.00% ± 2.28%, and 87.65% ± 3.10%, respectively (Fig. S11, ESI†). These simulation results highlight the significant potential of the fabricated photomemory for applications in neuromorphic computing.

Fig. 8 (a) Schematic diagram of the simulated fully-connected three-layer MLP ANN for image recognition. The layers are interconnected by weight matrices W_ij and W_jk. (b) Photonic/electrical pulse trains for weight updates, showing the relationship between conductance change and the number of applied pulses. Nonlinearity of LTP (NL_LTP) and LTD (NL_LTD), as well as dynamic range (DR) are indicated. (c)–(e) Simulation results showing the evolution of recognition accuracy as a function of training epochs based on the weight update characteristics presented in (b). Numbers in parentheses correspond to the number of neurons in each layer (input-hidden-output). Ideal accuracy is also shown for comparison.

2.5 Characterization of photo-programmable NOT gate

Logic gates with programmable behavior triggered by external optical stimuli are considered a promising strategy for reducing the energy cost of data-centric computing. Several photosensitive inverters have been disclosed to shift the voltage transfer curve under light illumination.^38–40 However, photosensitive inverters composed of photosensitive FETs provide transient trapped charges or photo-generated charge carriers and consequently limit their application to brain-inspired computing, which requires long-term memory behavior to mimic LTP and LTD. Therefore, to enable the photo-programmable inverter, photopatterned DPP-DTT-based OFET and BCP/ZnTPP/N2200-based ONVFGP were fabricated and controlled by various light sources (Fig. 9(a)). Given that the V_TH of photomemory can be manipulated by various optical stimuli, as depicted in Fig. 7(a), at least six distinct states of the V_OUT can be differentiated according to the screening effect on the gate electrode by the charges present in the floating gate. As depicted in Fig. 9(b), the wavelength-programmable ONVFGP enables us to fine-tune the transfer curve of the inverter circuit at a V_DD of 100 V under light programming with wavelengths ranging from 405 nm to 830 nm. The time traces displayed in Fig. 9(c) demonstrates a reliable photo-programmable circuit. At the initial state, as no charge is stored in the floating gate, the photomemory functions as a normal OFET, which shows a high V_OUT of 100 V (State 0) as the input voltage is set at 40 V. As the wavelength of the external optical light source decreases, the higher conductance of N2200 enables the higher voltage drop in V_OUT with a value of 99 V (State 1), 98 V (State 2), 87 V (State 3), 67 V (State 4) and 28 V (State 5) for 830 nm, 650 nm, 525 nm, 450 nm, 405 nm, respectively. To clearly illustrate the minimal difference, the switching characteristics of the photo-programmable inverter under the 830 nm and 650 nm-light stimuli, each subjected to 10 cycles, are presented in Fig. S12 (ESI†). Stable V_OUT of 99 V (State 1) and 98 V (State 2) can be achieved with an input voltage of 40 V, respectively. The output voltage and the corresponding logic state are determined by both the input voltage and the trapped charges in ONVFGP. The relationship between these factors is presented in Fig. 9(d). Notably, under 405 nm-light illumination, the inverter composed of a DPP-DTT-based OFET and a BCP/N2200-based OFET exhibited a negligible voltage drop in V_OUT, maintaining a V_OUT of 100 V with an input voltage set at 40 V (Fig. S13, ESI†). This result affirms the effective charge-trapping capability of ZnTPP embedded in BCP.

Fig. 9 (a) Photo-programmable inverter consisting of DPP-DTT-based FET and BCP/ZnTPP/N2200 ONVFGP. (b) Voltage transfer curves after photo-programming (c) Time traces of the input and output voltage at initial, under different light sources and after erasing. (d) Schematic circuit of photo-programmable inverter and the corresponding table of logic states manipulated by V_IN and light source.

3. Conclusion

A solution-processable and photo-programmable logic gate consisting of a DPP-DTT-based OFET and BCP/ZnTPP/N2200 ONVFGP is successfully realized using the photo-crosslinking agent 1,11-diazido-3,6,9-trioxaundecane. Decent film retention and high resolution of photopatterned functional materials ensure the film-stacking process for the photo-programmable logic gate while maintaining the optoelectrical and morphological features of the conjugated polymer and microphase-separated block copolymer. Owing to the wavelength-distinguishable functionality of BCP/ZnTPP/N2200 ONVFGP, the six states of the photo-programmable inverter can be fine-tuned using various light sources with wavelengths ranging from 405 nm to 830 nm.

4. Experimental section

4.1 Materials

Toluene, acetone, isopropyl alcohol, polystyrene, chloroform anhydrous, dimethylformamide (DMF, anhydrous,), zinc tetraphenylporphyrin (ZnTPP), poly[N,N′-bis(2-octyldodecyl)-1,4,5,8-naphtha-lenedicarboximide-2,6-diyl]-alt-5,5′-(2,2′-bithio-phene) (N2200), poly[2,5-(2-octyldodecyl)-3,6-diketopyrrolopyrrole-alt-5,5-(2,5-di(thien-2-yl)thieno [3,2-b]thiophene)] (DPP-DTT), and 1,11-diazido-3,6,9-trioxaundecane (azide) were purchased from Sigma-Aldrich. Polystyrene-block-poly(methacrylic acid) (PS-b-PMAA, M_n (PS) = 33 100 g mol⁻¹, M_n (PMAA) = 6700 g mol⁻¹) was purchased from polymer source.

4.2 Fabrication of photomemory and logic gate

The highly doped n-type silicon wafer with 300 nm-thick SiO₂ was cleaned with an ultrasonic cleaner in toluene, acetone, and isopropyl alcohol before being used as a gate and dielectric. First, 1 mg PS_33 100-b-PMAA₆₇₀₀, 0.66 mg ZnTPP and 4 mg azide were dissolved in 1 mL DMF. Then, the mixture was stirred at room temperature for one week to achieve a solution in which the mass ratio of BCP-to-azide was 1 : 4, and the molar ratio of PMAA repeating units to ZnTPP was 1 : 0.5. After stirring, 100 μL of the solution was spin-coated onto a wafer at a spin rate of 1000 rpm for 60 s, followed by 2000 rpm for 5 s using a spin-coater inside the N₂-filled glovebox. The BCP/azide/ZnTPP composite film was then exposed to a UV lamp (254 nm, 1.47 mW cm⁻²) through a shadow mask for 30 min. After photo-crosslinking, the film was developed using chloroform at a spin rate of 4000 rpm for 10 s using a spin-coater to remove the non-crosslinked regions. The photopatterned films from all the subsequent components were subjected to the same exposure intensity and development procedure. In the case of the conjugated polymer, high film retention could be successfully achieved with a polymer-to-azide ratio of 1 : 0.1. A blend solution consisting of 1 mg mL⁻¹ n-type N2200 conjugated polymer and 0.1 mg mL⁻¹ azide was spin-coast at a spin rate of 1000 rpm for 60 s and then 2000 rpm for 5 s onto the above photopatterned BCP/ZnTPP300. After the photopatterning of the n-type polymer film, the p-type DPP-DTT conjugated polymer film was formed by the spin-coating of 1 mg mL⁻¹ DPP-DTT and 0.1 mg mL⁻¹ azide blend solution at a spin rate of 1500 rpm for 60 s. After the photopatterning of the p-type polymer film, finally, the top-contact source and drain electrodes with a width and channel length of 1000 μm and 50 μm, respectively, were defined by 50 nm-thick thermal evaporated gold.

4.3 Characterization

The photo-crosslinking reaction between the functional polymer and azide and the interaction between BCP and ZnTPP were analyzed using Fourier transform infrared spectroscopy (FTIR, Miracle ATR, PerkinElmer). The UV-Vis spectra were characterized using a Shimadzu UV-2600 spectrometer. The ultraviolet photoelectron spectroscopy (UPS, ULVAC-PHI PHI 5000 Versaprobe II) with a photon energy of 21.22 eV was used to measure the valence band maxima of ZnTPP and N2200. Grazing-incident wide-angle X-ray scattering (GIWAXS) and Grazing-incident small-angle X-ray scattering (GISAXS) at a TPS beamline 25A in the National Synchrotron Radiation Research Center (NSRRC), Taiwan, were used to probe the morphology of the conjugated polymer and the micro-phase separation of BCP and BCP/ZnTPP, respectively. ¹H-nuclear magnetic resonance (NMR) spectra were recorded at 500 MHz using a BRUKER spectrometer in DMF-d₇. Photoluminescence (PL) and time-resolved photoluminescence (TRPL) spectra were measured to probe the exciton lifetimes using a 375 nm excitation wavelength. The emitted signals were collected via optical fibers and analyzed with a Hamamatsu C10910 streak camera equipped with an M10913 slow single-sweep unit at the National Synchrotron Radiation Research Center (NSRRC) in Taiwan. An atomic force microscope (AFM, Bruker) was used to measure the thickness and roughness of the photo-crosslinked films, as well as to calculate the film retention. Transmission electron microscopy (TEM, JEM-2100F Electron Microscope) was used to investigate the micro-phase separation of BCP/ZnTPP. An optical microscope (OLYMPUS BX53M) was used to measure the linewidth of the photopatterned film. The photo-responsive properties of the photomemory and photo-programmable inverter were investigated using a Keithley 4200-SCS semiconductor parameter analyzer in a nitrogen-filled glove box at room temperature. The laser pulse width and intensity were controlled via a Keithley 2636 source meter. Photoresponsivity was calculated using the equation (I_ON–I_OFF)/light intensity/illuminated area.

4.4 Nonlinearity analysis and neural network simulation

The analog G change is fitted by the following equations to mimic the weight update behaviors in neural networks:⁴¹

NL_LTP or NL_LTD = 1.726/(NL + 0.162) (6)

where G_LTP and G_LTD are the conductance for LTP and LTD, respectively, and P is the pulse number. G_max, G_min, and P_max are the maximum conductance, minimum conductance and the maximum number of applied pulses required to tune the conductance from G_min to G_max, respectively. NL_LTP and NL_LTD are the fitting parameters for the nonlinear synaptic weight update of potentiation and depression. The simulation of a fully-connected ANN for image recognition was conducted on the Cross-Sim platform.⁴² Three categories of datasets were used: large handwritten digits (28 × 28 pixels) from the MNIST dataset, small handwritten digits (8 × 8 pixels) from image versions, and the Sandia cyberfile format with 256 byte-pair statistical features. For each training epoch, the network was fed 8000 randomly selected patterns from a set of 60 000 training images. The recognition accuracy was then evaluated using a separate set of 10 000 test images. The simulation parameters were derived from the updated device conductance.

Author contributions

Y.-D. Lu, C.-R. Hsu carried out most of the experimental work and data analyses. S.-H. Ke and K.-L. Lai assisted with the data analysis. H.-L. Cheng and Y.-W. Wang were responsible for simulating the fully-connected ANN for image recognition. J.-Y. Chen conceived the concept, designed the experiments, and supervised the work. All authors discussed the research progress and contributed to editing the paper.

Data availability

The data supporting this article are included in the ESI.†

Conflicts of interest

There are no conflicts to declare.

Acknowledgements

The authors appreciate financial support from the National Science and Technology Council (NSTC) in Taiwan (MOST 110-2636-E-006-025-, MOST 111-2636-E-006-020-, MOST 111-2628-E-006-008-, NSTC 112-2628-E-006-004- and NSTC 113-2628-E-006-018-). This research was also supported by the Higher Education Sprout Project, Ministry of Education to the Headquarters of University Advancement at National Cheng Kung University. The authors acknowledge the use of JEOL JEM-2100F Cs STEM[EM000800] and NSTC 112-2740-M-006-001, belonging to the Core Facility Center of National Cheng Kung University. The authors also thank Professor Ho-Hsuan Chou for assistance with NMR analysis and Dr. Bi-Hsuan Lin at the National Synchrotron Radiation Research Center, Taiwan for the help of time-solved photoluminescence analysis.

References

1. A. Sebastian, M. Le Gallo, R. Khaddam-Aljameh and E. Eleftheriou, Nat. Nanotechnol., 2020, 15, 529–544.
2. W. Zhang, B. Gao, J. Tang, P. Yao, S. Yu, M.-F. Chang, H.-J. Yoo, H. Qian and H. Wu, Nat. Electron., 2020, 3, 371–382.
3. D. Nishioka, T. Tsuchiya, W. Namiki, M. Takayanagi, M. Imura, Y. Koide, T. Higuchi and K. Terabe, Sci. Adv., 2022, 8, eade1156.
4. M.-Y. Tsai, C.-T. Huang, C.-Y. Lin, M.-P. Lee, F.-S. Yang, M. Li, Y.-M. Chang, K. Watanabe, T. Taniguchi, C.-H. Ho, W.-W. Wu, M. Yamamoto, J.-L. Wu, P.-W. Chiu and Y.-F. Lin, Nat. Electron., 2023, 6, 755–764.
5. C. Lee, C. Lee, S. Lee, J. Choi, H. Yoo and S. G. Im, Nat. Commun., 2023, 14, 3757.
6. G. Migliato Marega, Y. Zhao, A. Avsar, Z. Wang, M. Tripathi, A. Radenovic and A. Kis, Nature, 2020, 587, 72–77.
7. C. Liu, H. Chen, X. Hou, H. Zhang, J. Han, Y. G. Jiang, X. Zeng, D. W. Zhang and P. Zhou, Nat. Nanotechnol., 2019, 14, 662–667.
8. H. Lai, Y. Zhou, H. Zhou, N. Zhang, X. Ding, P. Liu, X. Wang and W. Xie, Adv. Mater., 2022, 34, 2110278.
9. J. Pei, X. Wu, W. J. Liu, D. W. Zhang and S. J. Ding, ACS Nano, 2022, 16, 2442–2451.
10. Y. Shingaya, T. Iwasaki, R. Hayakawa, S. Nakaharai, K. Watanabe, T. Taniguchi, J. Aimi and Y. Wakayama, ACS Appl. Mater. Interfaces, 2024, 16, 33796–33805.
11. H. Hong, Z. Yang, Y. Ye, S. Ye, Z. Cheng, Z. Zeng, Z. Meng, Q. Lan, H. Zhang, Y. Chen, J. Wang, Y. Bai, X. Jiang, B. Liu, J. Hong, Z. Shen, T. Guo, Y. Zhang, Z. Weng and Y. Chen, Chem. Eng. J., 2024, 492, 152244.
12. W. C. Yang, Y. C. Lin, M. Y. Liao, L. C. Hsu, J. Y. Lam, T. H. Chuang, G. S. Li, Y. F. Yang, C. C. Chueh and W. C. Chen, ACS Appl. Mater. Interfaces, 2021, 13, 20417–20426.
13. S. Hong, S. H. Choi, J. Park, H. Yoo, J. Y. Oh, E. Hwang, D. H. Yoon and S. Kim, ACS Nano, 2020, 14, 9796–9806.
14. Y. C. Chiang, C. C. Hung, Y. C. Lin, Y. C. Chiu, T. Isono, T. Satoh and W. C. Chen, Adv. Mater., 2020, 32, e2002638.
15. D. Lee, E. Hwang, Y. Lee, Y. Choi, J. S. Kim, S. Lee and J. H. Cho, Adv. Mater., 2016, 28, 9196–9202.
16. M. Kim, H. Seong, S. Lee, H. Kwon, S. G. Im, H. Moon and S. Yoo, Sci. Rep., 2016, 6, 30536.
17. W.-C. Chen, Y.-C. Lin, Z.-S. Syu, Y.-S. Wu, K.-W. Lin, C.-L. Liu, C.-C. Kuo and W.-C. Chen, Chem. Eng. J., 2024, 494, 152897.
18. D. Lee, E. Hwang, Y. Lee, Y. Choi, J. S. Kim, S. Lee and J. H. Cho, Adv. Mater., 2016, 28, 9196–9202.
19. X. Huang, C. Liu, Z. Tang, S. Zeng, S. Wang and P. Zhou, Nat. Nanotechnol., 2023, 18, 486–492.
20. R. Q. Png, P. J. Chia, J. C. Tang, B. Liu, S. Sivaramakrishnan, M. Zhou, S. H. Khong, H. S. Chan, J. H. Burroughes, L. L. Chua, R. H. Friend and P. K. Ho, Nat. Mater., 2010, 9, 152–158.
21. H. J. Kim, A. R. Han, C.-H. Cho, H. Kang, H.-H. Cho, M. Y. Lee, J. M. J. Fréchet, J. H. Oh and B. J. Kim, Chem. Mater., 2011, 24, 215–221.
22. M. Y. Lee, H. J. Kim, G. Y. Jung, A. R. Han, S. K. Kwak, B. J. Kim and J. H. Oh, Adv. Mater., 2015, 27, 1540–1546.
23. S. Wang, J. Xu, W. Wang, G. N. Wang, R. Rastak, F. Molina-Lopez, J. W. Chung, S. Niu, V. R. Feig, J. Lopez, T. Lei, S. K. Kwon, Y. Kim, A. M. Foudeh, A. Ehrlich, A. Gasperini, Y. Yun, B. Murmann, J. B. Tok and Z. Bao, Nature, 2018, 555, 83–88.
24. M. J. Kim, M. Lee, H. Min, S. Kim, J. Yang, H. Kweon, W. Lee, D. H. Kim, J. H. Choi, D. Y. Ryu, M. S. Kang, B. Kim and J. H. Cho, Nat. Commun., 2020, 11, 1520.
25. B. Wang, W. Huang, S. Lee, L. Huang, Z. Wang, Y. Chen, Z. Chen, L.-W. Feng, G. Wang, T. Yokota, T. Someya, T. J. Marks and A. Facchetti, Nat. Commun., 2021, 12, 4937.
26. Y. Q. Zheng, Y. Liu, D. Zhong, S. Nikzad, S. Liu, Z. Yu, D. Liu, H. C. Wu, C. Zhu, J. Li, H. Tran, J. B. Tok and Z. Bao, Science, 2021, 373, 88–94.
27. I. C. Kwak, W. H. Cho, T. H. Kwon, M. J. Kim, J. I. Lee, D. G. Roe, D. H. Ho, S. B. Jo, M. S. Kang, B. Kim and J. H. Cho, Chem. Mater., 2023, 35, 71–80.
28. J. Y. Chen, Y. C. Chiu, Y. T. Li, C. C. Chueh and W. C. Chen, Adv. Mater., 2017, 29, 1702217.
29. J. C. Chen, Y. D. Lu and J. Y. Chen, Adv. Sci., 2023, 10, e2301028.
30. Y. H. Chang, C. W. Ku, Y. H. Zhang, H. C. Wang and J. Y. Chen, Adv. Funct. Mater., 2020, 30, 2000764.
31. Y. H. Chao, J. C. Chen, D. L. Yang, Y. J. Tseng, C. H. Hsu and J. Y. Chen, Adv. Funct. Mater., 2022, 32, 2112521.
32. L. D. Wiyanto, B. J. You, L. J. Chiang, D. L. Yang and J. Y. Chen, Adv. Funct. Mater., 2022, 32, 2206040.
33. H. T. Hsu, D. L. Yang, L. D. Wiyanto and J. Y. Chen, Adv. Photonics Res., 2021, 2, 2000185.
34. I. Raifuku, Y. P. Chao, H. H. Chen, C. F. Lin, P. E. Lin, L. C. Shih, K. T. Chen, J. Y. Chen, J. S. Chen and P. Chen, Ecomat, 2021, 3, e12142.
35. N. Awwad, A. T. Bui, E. O. Danilov and F. N. Castellano, Chem, 2020, 6, 3071–3085.
36. A. Al Mousawi, C. Poriel, F. Dumur, J. Toufaily, T. Hamieh, J. P. Fouassier and J. Lalevée, Macromolecules, 2017, 50, 746–753.
37. J. Wang, Y. Zhong, X. Wang, W. Yang, F. Bai, B. Zhang, L. Alarid, K. Bian and H. Fan, Nano Lett., 2017, 17, 6916–6921.
38. S. G. Seo and S. H. Jin, Phys. Status Solidi RRL, 2019, 13, 1900317.
39. Y. Z. Yuan, S. Y. Dai, S. Q. Yan, D. Y. Bao, Y. M. Wang, J. W. Zhang, Y. X. Li, Q. P. Wang, Q. Xin and A. M. Song, IEEE Electron Device Lett., 2021, 42, 1010–1013.
40. S. S. Radhakrishnan, S. Chakrabarti, D. Sen, M. Das, T. F. Schranghamer, A. Sebastian and S. Das, Adv. Mater., 2022, 34, 2202535.
41. P. Y. Chen, B. Lin, I. T. Wang, T. H. Hou, J. Ye, S. Vrudhula, J. s. Seo, Y. Cao and S. Yu, 2015 IEEE/ACM International Conference on Computer-Aided Design (ICCAD), 2015, pp. 194–199.
42. T. P. Xiao, C. H. Bennett, B. Feinberg, M. J. Marinella and S. Agarwal, CrossSim: accuracy simulation of analog in-memory computing, https://github.com/sandialabs/cross-sim..

---

Footnotes

† Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d5mh00036j

‡ Contributed equally to this work.

---