Wavelength-selective dual-channel coupling transistor for neuromorphic vision and In-sensor logic

Abstract

Most existing hardware based on the von Neumann architecture inherently features a single readout, which limits multifunctional capabilities in advanced computing systems. Here, we report a wavelength-selective dual-channel coupling transistor based on an MoS₂ channel and ferro-photoactive CuInP₂S₆ (CIPS) gate, which simultaneously generates horizontal source–drain (I_ds) and vertical gate-source (I_gs) dual readouts under optical illumination. Under 532 nm and 405 nm illumination, the device exhibits wavelength-dependent synaptic response polarity: the I_ds current mimics the center-excitatory pathway of bipolar cells, while the I_gs current mimics the surround-inhibitory pathway of horizontal cells, enabling bio-inspired in-sensor neuromorphic perception. Moreover, the wavelength-selective dual readouts enable reconfigurable XOR and AND optical logic operations within a single device, allowing it to function as an all-optical half-adder. This reduces the component count by 91.7% compared with CMOS-based circuits. Our device integrates sensing, neuromorphic preprocessing, and arithmetic logic within a single hardware unit, paving the pathway toward compact, energy-efficient vision systems.

Graphical abstract

Keywords

ferro-photoactive gate / dual-readout / retina-neuromorphic / half-adder / dual-channel coupling

1 Introduction

The von Neumann architecture inherently suffers from energy inefficiency and latency due to the physical separation of sensors, memory, and processors [1−4], a bottleneck that is particularly prominent in visual information processing [5−9]. Biological retinas address this challenge through a parallel “center-excitation and periphery-inhibition” processing mechanism [10−12], accomplishing dynamic range compression and feature extraction with merely microwatt-scale power consumption. This provides an ideal blueprint for neuromorphic device design.

Two-dimensional (2D) materials, characterized by atomic thickness, tunable bandgaps, and strong light-matter interactions, have emerged as promising platforms for bio-inspired optoelectronic devices [13−16]. In the field of visual simulation, WSe₂-based bifurcated floating-gate device arrays have successfully emulated retinal functions [17], while MoS₂ exhibits advantages in dynamic visual adaptability advantages — such as alcohol-doped MoS₂ photonic synapses that simulate human eye adaptation mechanisms [18], and Fe₇S₈@MoS₂ heterostructures that achieve 92% recognition accuracy under low-light conditions [19]. In addition, a ReS₂/CuInP₂S₆ heterostructure realizes the emulation of multimodal synapses by leveraging the polarization switching of CIPS and the laser-induced carrier relaxation process in ReS₂ [20]. Moreover, an optoelectronic device based on a multilayer graphene/CuInP₂S₆/Au structure achieves visual perception, processing, and memory functions through photoelectrical cooperative modulation [21].

In addition, as a core component of arithmetic logic units, significant progress has been made in miniaturizing such logic elements using 2D materials — demonstrating transistors capable of XOR or AND operations, such as monolayer WSe₂ homojunction split-gate transistors [25], 2-transistor implementations based on dual-ferroelectric-gate MoS₂/MoTe₂ hetero-integration [26], and 2T2R structures leveraging 2D WSe₂/MoS₂ heterostructures [27]. However, these logic devices are typically developed in isolation from sensory components [22−24]. Most reported 2D optoelectronic synapses only perform retinal neuromorphic preprocessing without embedded logic functions, while standalone optical logic transistors require separate sensor units for signal pretreatment. Our wavelength-selective dual-channel coupling mechanism overcomes this functional segregation within a single MoS₂/CIPS heterostructure device.

Here, we materialize this vision by introducing a ferro-photoactive CIPS gate into a MoS₂ transistor, demonstrating wavelength-selective dual-channel coupling between I_ds and I_gs, wherein a single optical input elicits two distinct, antagonistic outputs, integrating bio-inspired preprocessing and logic capabilities into one single device. The I_ds mimics the bipolar cell’s center-excitatory response, while I_gs mimics the horizontal cell’s surround-inhibitory feedback, enabling the device to perform retina-like, in-sensor neuromorphic computing for real-time image enhancement. More importantly, we exploit the wavelength-dependent synergism and antagonism between these two channels to directly implement reconfigurable optical logic gates (XOR and AND), allowing a single device to function as an optical half-adder, a fundamental arithmetic unit. This work provides a tangible hardware blueprint for next-generation vision systems that are not only “smart” in perception but also capable of primitive decision-making at the very point of sensing, paving the way to overcome the von Neumann bottleneck in visual information processing.

2 Characterization and measurements

All few-layered CuInP₂S₆ and MoS₂ flakes were exfoliated from commercially available bulk crystals (purchased from Shanghai Onway Technology Co., Ltd.). The device is fabricated via an ultraviolet maskless lithography machine (TuoTuo Technology, UV Litho-ACA). The morphology of the thin film was measured using an atomic force microscope (AFM, Bruker Dimension Icon). UV-Vis absorption spectra were measured on a PerkinElmer Lambda 1050 spectrometer. Electrical measurements were conducted using an Agilent B1500A semiconductor parameter analyzer.

3 Results and discussion

The direct pathway and lateral pathway in the biological retina constitute the core mechanism of parallel visual signal processing, and this neural circuit is composed of photoreceptor cells (rods/cones cells), horizontal cells, and bipolar cells, which collectively enable efficient encoding and preprocessing of optical information collectively [28−30]. As shown in Fig. 1(a), in the center-surround receptive field in biological vision, horizontal cells act as key inhibitory interneurons in the outer retinal network, laterally connecting multiple photoreceptors and bipolar cells. Through γ-aminobutyric acid (GABA)-ergic inhibitory neurotransmission, this feedback mechanism ensures that bipolar cells simultaneously integrate direct central inputs and modulated peripheral inputs, establishing antagonistic center-surround receptive fields [31−34]. Inspired by this structure, we design a dual-channel output phototransistor based on MoS₂ channel and ferro-photoactive CIPS gate dielectric [Fig. 1(b)]. Schematic diagrams, an optical image, and an atomic force microscopy (AFM) image of the device are presented in Figs. S1(a–c) (Supporting Information). The thicknesses of the CIPS and MoS₂ films are approximately 50.4 nm and 5.9 nm, respectively. Figure S1(d) shows the Raman spectra of CIPS and MoS₂, whose characteristic peaks are consistent with those reported in the literature [35, 36]. Because the CIPS layer combines high-κ gate control capability (Fig. S2) with moderate bandgap characteristics, it can simultaneously output the main channel current I_ds (Y₁) and the light-induced gate current I_gs (Y₂) when stimulated by different wavelengths (532 nm and 405 nm). Here, “dual output” refers to the simultaneous readout of I_ds and I_gs, while “dual-channel coupling” describes the modulation of the horizontal MoS₂ channel current by photoinduced carrier transport through the vertical CIPS/MoS₂ path. As shown in Fig. 1(b), the resulting electrical response mimics the synaptic transmission and modulation processes of neurons after receiving neurotransmitters. The horizontal channel (Y₁) emulates the direct signal transmission of bipolar cells, while the vertical channel (Y₂) reproduces the lateral inhibition mediated by horizontal cells, thereby replicating the “center-surround” antagonistic visual information processing mechanism of the retina.

This design provides an attractive insight into constructing a compact and efficient neuromorphic computing system. Under a negative gate voltage (V_gs < 0), the spectral response curves of the transistor all exhibit positive current changes. In contrast, when V_gs > 0, the device exhibits significant wavelength selectivity, 532 nm light irradiation enhances the current, while 405 nm light irradiation causes current attenuation [Fig. 2(a)]. To systematically investigate the photoresponse, we measured the I_ds−V_gs transfer characteristic curves under different irradiation conditions [Fig. 2(b)]. The observed counterclockwise hysteresis is mainly attributed to the polarization-induced electrostatic modulation of the MoS₂ channel by the ferroelectric CIPS gate. The curves show that the I_ds output exhibits typical n-type semiconductor behavior, with an on/off ratio of up to 10⁶. Meanwhile, the curves demonstrate a significant wavelength-gate voltage synergistic effect. Under positive gate voltage and 532 nm light irradiation, long-wavelength photons primarily excite electron transport in the horizontal channel of MoS_2, leading to an increase in I_ds. Under 405 nm light irradiation, short-wavelength photons simultaneously activate the vertical channel modulated by the CIPS layer, generating a significant I_gs output [Fig. 2(c)]. In addition, the output curves of the photosensor were measured and are shown in Fig. S3, demonstrating excellent gate control capability.

Notably, the subthreshold characteristics extracted from Fig. 2(b) show that the minimum subthreshold swing (SS) is 50 mV·dec⁻¹ [Fig. 2(d)], and that I_ds remains stable over four orders of magnitude, both of which reflect the device’s low power consumption and excellent switching performance. After fabricating a MoS₂/h-BN photosensor by replacing CIPS with h-BN, we found that its SS was significantly higher than that of the MoS₂/CIPS photosensor (Fig. S4). This difference indicates that the excellent performance originates from the negative capacitance effect of the CIPS ferroelectric gate, which enables it to break through the 60 mV·dec⁻¹ thermionic limit through internal amplification of the gate voltage [37−39]. The expression for SS is as follows:

(1)SS={\displaystyle \frac{\mathrm{d}{V}_{\mathrm{g}\mathrm{s}}}{\mathrm{d}\left({\log }_{10}{I}_{\mathrm{d}\mathrm{s}}\right)}}={\displaystyle \frac{\mathrm{d}{V}_{\mathrm{g}\mathrm{s}}}{\mathrm{d}{\psi }_{\mathrm{s}}}}\times {\displaystyle \frac{\mathrm{d}{\psi }_{\mathrm{s}}}{\mathrm{d}\left({\log }_{10}{I}_{\mathrm{d}\mathrm{s}}\right)}}=m\times n,

where ψ_s is the surface potential.

By measuring the ultraviolet-visible (UV-Vis) absorption spectrum [Fig. 2(e)], it can be observed that MoS₂ exhibits strong absorption in the 405−532 nm wavelength range, while CIPS has a sharp absorption edge at ≈ 450 nm and weak absorption in the long-wavelength region. In the wavelength range of 450−800 nm, the lower absorbance of the CIPS/MoS₂ overlapping region is mainly attributed to the weak absorption of CIPS and the additional reflection and scattering losses introduced by the stacked heterostructure.

This spectral difference between CIPS and MoS₂ stems from their distinct band structures [Fig. 2(f)]. The electron affinities of CIPS and multilayer MoS₂ were taken as approximately 3.7 and 4.3 eV, respectively, based on previous reports [40, 41]. The optical bandgap of CIPS was determined to be approximately 2.9 eV from the UV-Vis absorption spectrum and Tauc analysis shown in Fig. S5, whereas the approximately 1.2 eV bandgap of multilayer MoS₂ was adopted from the literature. Therefore, Figure 2(f) represents a schematic band alignment used to illustrate the qualitative direction of photoinduced carrier transfer. The medium-wide bandgap of CIPS enables selective optical excitation of the CIPS/MoS₂ heterojunction by 405 and 532 nm lasers. Figure 2(g) illustrates the device’s photocharge transport mechanism. Under 532 nm light irradiation, since CIPS has weak absorption of visible light, the photoresponse mainly originates from MoS₂. The positive V_gs induces downward band bending in MoS₂, leading to electron accumulation and forming a horizontal current channel (Y₁). In contrast, CIPS exhibits strong absorption of 405 nm light, under bottom gate regulation, the photogenerated holes produced in CIPS migrate vertically toward the MoS₂ channel, forming a vertical gate current (I_gs) that exerts a significant inhibitory effect on the horizontal current (I_ds) [Figs. 2(h) and (i)]. The energy band diagrams under dark and light illumination conditions at negative gate voltage, as well as the photocurrent spectra under different gate voltages, are presented in Figs. S6 and S7, respectively. Through the synergistic design of band engineering (CIPS/MoS₂ heterojunction) and gate control, the device achieves co-regulation of I_ds and I_gs, breaking the traditional approach in 2D devices where performance optimization is achieved by minimizing I_gs. In summary, Figures 2(e) and (f) can explain the output characteristics in Figs. 2(h) and (i). 532 nm light is mainly absorbed by MoS₂, which enhances I_ds while the I_gs response is relatively weak, whereas 405 nm light can also excite narrow-bandgap CIPS, enabling vertical carrier transfer and generating I_gs while inhibiting I_ds. In biological neural networks (BNNs), retinal bipolar cells are responsible for transmitting all information required for vision [42, 43]. Based on axon terminal morphology and inner plexiform layer distribution, they are classified into ON/OFF types, forming synapses with corresponding ganglion cells to relay visual signals [Fig. 3(a)]. Under light stimulation, photoreceptors reduce glutamate release, inducing depolarization of ON bipolar cells and excitatory postsynaptic currents (EPSCs), while OFF bipolar cells hyperpolarize, release inhibitory neurotransmitters, and produce inhibitory postsynaptic currents (IPSCs) [44, 45]. This mechanism, which achieves efficient detection of light intensity changes through a dynamic balance between depolarization and hyperpolarization [46, 47], provides significant insights into artificial vision sensor design. In the device, optical pulses act as presynaptic stimuli, and the resulting channel-current variation is regarded as the postsynaptic current. The positive response under 532 nm illumination corresponds to EPSC, whereas the negative response under 405 nm illumination corresponds to IPSC. The transient and history-dependent current modulation corresponds to STP, with paired-pulse facilitation or depression serving as a representative manifestation. The enhanced and more persistent current modulation induced by increasing the pulse width, intensity, or number indicates the transition toward LTP.

Paired-pulse facilitation (PPF), a typical short-term plasticity (STP) process in biological synapses, refers to postsynaptic signal enhancement induced by successive presynaptic stimulations. Under 532 nm laser irradiation (pulse width = 1 s, pulse interval = 1 s; Fig. 3(b), the conductance enhancement induced by the second pulse is significantly higher than that induced by the first, exhibiting positive PPF behavior. This arises from incomplete electrode collection of photogenerated carriers from the first pulse, which improves the carrier injection efficiency of the second pulse. In contrast, under 405 nm laser irradiation (pulse width = 1 s, pulse interval = 0.3 s). Figure 3(c) exhibits negative PPF behavior due to the injection of photogenerated holes from CIPS into the channel. This wavelength-dependent PPF behavior not only emulates the short-term plasticity and dynamic information processing of biological synapses but also provides novel insights into developing spectrally selective photoelectric neuromorphic devices. Memory retention depends on learning intensity, duration, and frequency, repeated learning induces the STP-LTP transition, which facilitates brain information processing and storage. Human memory and forgetting are dynamically balanced, with memory mediated via synaptic potentiation and forgetting via synaptic depression. At the device level, regulating the light pulse width, intensity, and number modulates EPSC/IPSC amplitudes to achieve the STP-LTP transition, emulating memory formation and forgetting. Figures 3(d−f) show positive photoresponse behavior under 532 nm pulse irradiation, where EPSC amplitude increases long-term with pulse width, intensity, and number, mimicking long-term memory via calcium-mediated synaptic potentiation in biological synapses. In contrast, 405 nm pulses elicit negative photoresponse, with enhanced IPSC suppression as pulse parameters increase, simulating inhibitory neurotransmitter-regulated long-term forgetting [Figs. 3(g)−(i)]. Figure S8 further presents the device’s emulation of color images perception, memory, and forgetting functions. The postsynaptic current variation (ΔPSC) increases with the light pulse number and decays post-stimulation [Figs. S8(a, b)], calculated as

(2)\mathrm{\Delta }PSC=I_{\text{light}}-I_{\text{dark}}.

Here, I_dark represents the initial current, and I_light denotes the current after the application of optical pulse stimulation.

As shown in Fig. S8(c), increasing the number of light pulses improves image clarity and anti-forgetting capability, while the image gradually blurs after light stimulation ceases. Benefiting from the ΔPSC differences under different light irradiations, this multi-parameter regulation mechanism lays a crucial foundation for efficient and reconfigurable neuromorphic computing.

In the retinal neural circuit, horizontal cells establish a center-surround antagonistic receptive field by forming lateral connections with multiple photoreceptor cells and bipolar cells [Fig. 4(a)]. The central region directly receives photoreceptor inputs, while the surround region integrates indirect inputs from adjacent photoreceptors via horizontal cells. This center-surround antagonistic mechanism enables parallel signal processing and lateral regulation, enhancing central contrast while suppressing the uniform background to optimize visual information processing [48, 49]. The dual-channel photosensor’s I_ds and I_gs output channels mimic the direct conduction pathway of bipolar cells and the lateral modulation pathway regulated by horizontal cells, respectively. Specifically, the increase in I_ds under 532 nm illumination represents center excitation, whereas the activation of I_gs accompanied by the suppression of I_ds under 405 nm illumination represents surround activation and inhibitory feedback. We first measured the time-dependent current response of the photosensor under different irradiations [Figs. 4(b, c)]. Under 532 nm light, the I_ds channel exhibits a positive response, corresponding to ON-pathway activation in the central region of bipolar cells, while the I_gs channel shows a negligible response, analogous to the weak feedback of horizontal cells in the center. In contrast, 405 nm laser irradiation induces a positive I_gs response mimicking peripheral horizontal cell activation, as well as a negative I_ds response, which is analogous to horizontal cells’ negative feedback regulation of bipolar cell peripheral responses via inhibitory neurotransmitters (e.g., γ-aminobutyric acid, GABA) to suppress peripheral bipolar cell activity.

Rational design of the device array enables real-time image preprocessing using wavelength-dependent positive/negative photoconductive responses. We simulated a 28 × 28 array for image denoising, with the input of a noisy image containing a green “3” pattern and blue background noise [left in Fig. 4(b)]. The target pattern area generates a positive photocurrent under 532 nm light, while 405 nm laser irradiation triggers a negative photoconductive response in the background. Furthermore, the generated dual-channel outputs (I_ds/I_gs) correspond to the direct pathway/lateral pathway of the retina, respectively [right in Fig. 4(b)]. Analysis of Figure 4(c) yields the quantitative correlation between pulse number and photocurrent variation (Fig. S9). Experimental results show that with an increase in the number of pulses, the Weber contrast significantly increases to up to 266.4% under 405 nm and 532 nm irradiation, thereby producing a clear target pattern [inset in Fig. 4(c)].

In addition, the denoised image preprocessed by the array is input into a convolutional neural network (CNN) for image recognition (Fig. S10), achieving a significantly improved accuracy of 98.4%. Therefore, the device emulates retinal visual signal preprocessing for sensor-level noise suppression and contrast enhancement. Compared with traditional retinal-mimetic designs [Fig. 4(d)], it uses the fewest transistors, a reduction of approximately 95.7% [6, 50−52]. Synergistic wavelength selectivity and dual-channel antagonistic mechanisms enable precise replication of the retina’s “center-excitation, surround-inhibition” core computation, opening new avenues for 2D semiconductor applications in neuromorphic engineering and advanced intelligent image sensors. Having demonstrated the device’s bio-inspired perception capability, we further exploit its programmable photoresponse to achieve digital logic, which is absent in biology. Recent studies show that a single biological neuron can perform nonlinear Boolean operations (e.g., XOR operation) [53]. However, existing neuromorphic systems usually separate logical computation from sensory simulation functions, which relies on complex hybrid circuit designs. Additionally, conventional CMOS half-adder circuits typically require at least 12 transistors to realize the logic functions [as shown in Fig. 5(a)], which not only increases power consumption but also limits high-density device integration. By contrast, based on the photosensitive dielectric gate regulation mechanism, this study designs a photosensor with both a horizontal I_ds channel and a vertical I_gs channel [Fig. 5(b)]. Through the independent control of the two channels via selective optical inputs, the logic function of an all-optical half-adder can be realized, reducing the transistor count by 91.7%, significantly lowering power consumption, and enhancing neuromorphic information processing capabilities.

As one of the most fundamental arithmetic logic units in digital circuits, the half-adder is used to perform the addition of two single-bit binary numbers. It consists of two inputs A and B, one carry output C₀, and a sum terminal S. Specifically, the XOR logic gate is responsible for calculating the sum, while the AND logic gate is responsible for generating the carry [Fig. 5(d)]. Figures 5(c) and (e), respectively, show the I_ds and I_gs output curves of the device under both 532 nm and 405 nm light irradiation, with bias conditions set at V_ds = 1 V and V_gs = 1 V. The red dashed lines in the figures represent the threshold levels of the current response, which are used to distinguish the logic states of the device. We define the irradiation states of 532 nm and 405 nm light as the logic input signals, with the light-on state representing logic “1” and the light-off state representing logic “0”. Concurrently, the responses of I_ds and I_gs are defined as the logic outputs, where a significant current response corresponds to logic “1” (positive or negative photoresponse), and a weak current response corresponds to logic “0”. This experiment uses a dual-ended fiber to simultaneously illuminate the device surface with two light beams at 405 nm and 532 nm, enabling precise periodic alignment of the optical signals. The logical measurement employs a fixed timing sequence to apply both optical inputs, generating the (1,1) input state through temporal overlap of the two pulses. Results show that the transistor can implement XOR logic at the I_ds output and AND logic at the I_gs output. Figure 5(f) illustrates the equivalent circuit structure of the half-adder, and the accompanying truth table in the inset further validates the correctness of its logical functions. Additionally, the half-adder logic based on the dual-channel output photosensor enables image encryption and decryption, as shown in Fig. S11 (Supporting Information). Through a dual verification mechanism, it significantly enhances the security of image transmission and storage.

4 Conclusions

In summary, we demonstrate a dual-readout transistor employing n-type MoS₂ as the channel and CIPS as the ferro-photoactive gate dielectric. The dual readouts consist of the horizontal source-drain current I_ds and the vertical gate-source current I_gs. Through the collaborative design of wavelength selectivity and the dual-readout antagonism mechanism, real-time contrast enhancement analogous to the biological retinal mechanism is physically constructed, enabling in-sensor preprocessing of visual information. Furthermore, by leveraging the cross-modulation characteristics of the dual current channels, the logical operation of an all-optical half-adder can be simultaneously realized, which will be of great significance for the construction of high-efficiency neuromorphic systems. Future work will focus on broadening the response spectrum beyond 405/532 nm for broadband in-sensor computing.

References

[1] Y. Chai , In-sensor computing for machine vision, Nature 579(7797), 32 (2020)

[2] G. D. Zhou , J. Li , Q. L. Song , L. D. Wang , Z. J. Ren , B. Sun , X. F. Hu , W. H. Wang , G. B. Xu , X. D. Chen , L. Cheng , F. C. Zhou , and S. K. Duan , Full hardware implementation of neuromorphic visual system based on multimodal optoelectronic resistive memory arrays for versatile image processing, Nat. Commun. 14(1), 8489 (2023)

[3] D. Ielmini and H. S. P. Wong , In-memory computing with resistive switching devices, Nat. Electron. 1(6), 333 (2018)

[4] A. Sebastian , M. Le Gallo , R. Khaddam-Aljameh , and E. Eleftheriou , Memory devices and applications for in-memory computing, Nat. Nanotechnol. 15(7), 529 (2020)

[5] G. Indiveri and R. Douglas , Neuromorphic vision sensors, Science 288(5469), 1189 (2000)

[6] F. Y. Liao , Z. Zhou , B. J. Kim , J. W. Chen , J. L. Wang , T. Q. Wan , Y. Zhou , A. T. Hoang , C. Wang , J. F. Kang , J. H. Ahn , and Y. Chai , Bioinspired in-sensor visual adaptation for accurate perception, Nat. Electron. 5(2), 84 (2022)

[7] P. Y. Huang , B. Y. Jiang , H. J. Chen , J. Y. Xu , K. Wang , C. Y. Zhu , X. Y. Hu , D. Li , L. Zhen , F. C. Zhou , J. K. Qin , and C. Y. Xu , Neuro-inspired optical sensor array for high-accuracy static image recognition and dynamic trace extraction, Nat. Commun. 14(1), 6736 (2023)

[8] Y. R. Wang , S. M. Nie , S. S. Liu , Y. F. Hu , J. W. Fu , J. Y. Ming , J. Liu , Y. Q. Li , X. He , L. Wang , W. Li , M. D. Yi , H. F. Ling , L. H. Xie , and W. Huang , Dual-adaptive heterojunction synaptic transistors for efficient machine vision in harsh lighting conditions, Adv. Mater. 36(32), 2404160 (2024)

[9] Y. K. Yang , C. Pan , Y. X. Li , X. Yang dong , P. F. Wang , Z. A. Li , S. Wang , W. T. Yu , G. Y. Liu , B. Cheng , Z. F. Di , S. J. Liang , and F. Miao , In-sensor dynamic computing for intelligent machine vision, Nat. Electron. 7(3), 2520 (2024)

[10] S. Strauss , M. M. Korympidou , Y. L. Ran , K. Franke , T. Schubert , T. Baden , P. Berens , T. Euler , and A. L. Vlasits , Center-surround interactions underlie bipolar cell motion sensitivity in the mouse retina, Nat. Commun. 13(1), 5574 (2022)

[11] J. A. Gaynes , S. A. Budoff , M. J. Grybko , J. B. Hunt , and A. Poleg-Polsky , Classical center-surround receptive fields facilitate novel object detection in retinal bipolar cells, Nat. Commun. 13(1), 5575 (2022)

[12] M. H. Turner , G. W. Schwartz , and F. Rieke , Receptive field center-surround interactions mediate context-dependent spatial contrast encoding in the retina, eLife 7, e38841 (2018)

[13] S. Y. Wang , X. X. Liu , and P. Zhou , The road for 2D semiconductors in the silicon age, Adv. Mater. 34(48), 2106886 (2022)

[14] B. Radisavljevic , A. Radenovic , J. Brivio , V. Giacometti , and A. Kis , Single-layer MoS₂ transistors, Nat. Nanotechnol. 6(3), 147 (2011)

[15] Y. Liu , X. D. Duan , H. J. Shin , S. Park , Y. Huang , and X. F. Duan , Promises and prospects of two-dimensional transistors, Nature 591(7848), 43 (2021)

[16] K. S. Novoselov , A. Mishchenko , A. Carvalho , and A. H. C. Neto, , 2D materials and van der Waals heterostructures, Science 353(6298), aac9439 (2016)

[17] Z. R. Peng , L. Tong , W. H. Shi , L. L. Xu , X. Y. Huang , Z. Li , X. X. Yu , X. H. Meng , X. He , S. J. Lv , G. C. Yang , H. Hao , T. Jiang , X. S. Miao , ang L. Ye , Multifunctional human visual pathway-replicated hardware based on 2D materials, Nat. Commun. 15(1), 8650 (2024)

[18] X. Liu , M. Huang , X. F. Zou , W. Ali , S. U. Rehman , J. Li , Z. W. Li , L. Xiang , and A. L. Pan , Alcohol-sensitive MoS₂ optoelectronic synapses for mimicking human-like visual adaptation, InfoMat 7(8), e70019 (2025)

[19] Y. Deng , S. H. Liu , M. S. Li , N. Zhang , Y. M. Feng , J. B. Han , Y. Kapitonov , Y. Li , and T. Y. Zhai , Large-area growth of synaptic heterostructure arrays for integrated neuromorphic visual perception chips, Chip 3(2), 100088 (2024)

[20] K. T. Zhang , Z. W. Huo , Y. F. Wang , H. Yan , J. R. Yu , L. Q. Cheng , Y. X. Fu , C. Liu , Z. H. Li , Z. Liu , Z. L. Wang , Q. J. Sun , and X. D. Lin , Neuromorphic tactile-visual perception based on 2D ReS₂/CIPS heterojunction artificial synapse, Nano Energ. 144, 111399 (2025)

[21] Y. Q. Liu , Y. H. Wu , H. J. Han , Y. Wang , R. X. Peng , K. Liu , D. Yi , C. W. Nan , and J. Ma , CuInP₂S₆-based electronic/optoelectronic synapse for artificial visual system application, Adv. Funct. Mater. 34(1), 2306945 (2023)

[22] Z. Y. Lin , Y. Liu , U. Halim , M. N. Ding , Y. Y. Liu , Y. L. Wang , C. C. Jia , P. Chen , X. D. Duan , C. Wang , F. Song , M. F. Li , C. Z. Wan , Y. Huang , and X. F. Duan , Solution-processable 2D semiconductors for high-performance large-area electronics, Nature 562(7726), 254 (2018)

[23] X. X. Sun , C. G. Zhu , J. L. Yi , L. Xiang , C. Ma , H. W. Liu , B. Y. Zheng , Y. Liu , W. X. You , W. J. Zhang , D. L. Liang , Q. Shuai , X. L. Zhu , H. G. Duan , L. Liao , Y. Liu , D. Li , and A. L. Pan , Reconfigurable logic-in-memory architectures based on a two-dimensional van der Waals heterostructure device, Nat. Electron. 5(11), 752 (2022)

[24] K. K. Gao , B. Sun , Z. L. Cao , X. G. Duan , W. T. Yan , G. Q. Tong , G. D. Zhou , X. L. Chen , and J. Y. Shao , Biomaterial/organic heterojunction based memristor for logic gate circuit design, data encryption, and image reconstruction, Adv. Funct. Mater. 34(36), 2401132 (2024)

[25] X. P. Li , Z. J. Wang , X. J. Tang , P. Z. Yuan , L. Li , C. H. Shen , Y. R. Jiang , X. H. Song , and C. X. Xia , Logic computing field-effect transistors based on a monolayer WSe₂ homojunction for the semi-adder and decoder, Nano Lett. 24(35), 11132 (2024)

[26] Z. D. Luo , S. Q. Zhang , Y. Liu , D. W. Zhang , X. T. Gan , J. Seidel , Y. Liu , G. Q. Han , M. Alexe , and Y. Hao , Dual-ferroelectric-coupling-engineered two-dimensional transistors for multifunctional in-memory computing, ACS Nano 16(2), 3362 (2022)

[27] H. W. Chen , X. Y. Xue , C. S. Liu , J. B. Fang , Z. Wang , J. L. Wang , D. W. Zhang , W. D. Hu , and P. Zhou , Logic gates based on neuristors made from two-dimensional materials, Nat. Electron. 4(6), 399 (2021)

[28] J. J. Nassi and E. M. Callaway , Parallel processing strategies of the primate visual system, Nat. Rev. Neurosci. 10(5), 360 (2009)

[29] R. H. Masland , The neuronal organization of the retina, Neuron 76(2), 266 (2012)

[30] B. R. Conway , Color vision, cones, and color-coding in the cortex, Neuroscientist 15(3), 274 (2009)

[31] J. I. Toyoda and K. Tonosaki , Effect of polarisation of horizontal cells on the on-centre bipolar cell of carp retina, Nature 276(5686), 399 (1978)

[32] H. Qian and J. E. Dowling , Novel GABA responses from rod-driven retinal horizontal cells, Nature 361(6408), 162 (1993)

[33] W. K. Stell , A. T. Ishida , and D. O. Lightfoot , Structural basis for on-and off-center responses in retinal bipolar cells, Science 198(4323), 1269 (1977)

[34] J. E. Dowling , J. E. Brown , and D. Major , Synapses of horizontal cells in rabbit and cat retinas, Science 153(3744), 1639 (1966)

[35] F. C. Liu , L. You , K. L. Seyler , X. B. Li , P. Yu , et al. Room-temperature ferroelectricity in CuInP₂S₆ ultrathin flakes, Nat. Commun. 7(1), 12357 (2016)

[36] H. Li , Q. Zhang , C. C. R. Yap , B. K. Tay , T. H. T. Edwin , A. Olivier , and D. Baillargeat , From bulk to monolayer MoS₂, Adv. Funct. Mater. 22(7), 1385 (2012)

[37] X. W. Wang , P. Yu , Z. D. Lei , C. Zhu , X. Cao , F. C. Liu , L. You , Q. S. Zeng , Y. Deng , C. Zhu , J. D. Zhou , Q. D. Fu , J. L. Wang , Y. Z. Huang , and Z. Liu , Van der Waals negative capacitance transistors, Nat. Commun. 10, 3037 (2019)

[38] D. Daw , H. Bouzid , M. Jung , D. Suh , C. Biswas , and Y. Hee Lee , Ultrafast negative capacitance transition for 2D ferroelectric MoS₂/graphene transistor, Adv. Mater. 36(13), 2304338 (2024)

[39] H. Chen , Y. F. Long , S. Y. Zhang , K. Liu , M. F. Chen , J. X. Zhao , M. W. Si , and L. Wang , Van der Waals ferroelectric CuCrP₂S₆-enabled hysteresis-free negative capacitance field-effect transistors, Adv. Mater. 37(18), 2419125 (2025)

[40] J. B. Wu , H. Y. Chen , N. Yang , J. Cao , X. D. Yan , F. X. Liu , Q. B. Sun , X. Ling , J. Guo , and H. Wang , High tunnelling electroresistance in a ferroelectric van der Waals heterojunction via giant barrier height modulation, Nat. Electron. 3(8), 466 (2020)

[41] A. P. Nayak , S. Bhattacharyya , J. Zhu , J. Liu , X. Wu , T. Pandey , C. Q. Jin , A. K. Singh , D. Akinwande , and J. F. Lin , Pressure-induced semiconducting to metallic transition in multilayered molybdenum disulphide, Nat. Commun. 5(3731), 3731 (2014)

[42] T. Euler , S. Haverkamp , T. Schubert , and T. Baden , Retinal bipolar cells: elementary building blocks of vision, Nat. Rev. Neurosci. 15(8), 507 (2014)

[43] J. C. Hsiang , N. Shen , F. Soto , and D. Kerschensteiner , Distributed feature representations of natural stimuli across parallel retinal pathways, Nat. Commun. 15(1), 1920 (2024)

[44] H. M. Schreyer and T. Gollisch , Nonlinear spatial integration in retinal bipolar cells shapes the encoding of artificial and natural stimuli, Neuron 109(10), 1692 (2021)

[45] S. H. DeVries , Bipolar cells use Kainate and AMPA receptors to filter visual information into separate channels, Neuron 28(3), 847 (2000)

[46] Q. L. Lin , C. Q. Li , H. G. Xiong , M. Zhang , J. W. Qiao , J. P. Wu , L. Yang , S. Wang , H. Chen , Y. N. Wei , D. Zheng , G. H. Lu , X. T. Hao , D. H. Yu , Y. H. Cai , A. Facchetti , and H. Huang , Event-driven retinomorphic photodiode with bio-plausible temporal dynamics, Nat. Nanotechnol. 20(9), 1213 (2025)

[47] Z. Han , Y. C. Zhang , Q. Mi , J. You , N. N. Zhang , Z. Y. Zhong , Z. M. Jiang , H. Guo , H. Y. Hu , L. M. Wang , and Z. M. Zhu , Reconfigurable homojunction phototransistor for near-zero power consumption artificial biomimetic retina function, ACS Nano 18(43), 29968 (2024)

[48] S. Strauss , M. M. Korympidou , Y. L. Ran , K. Franke , T. Schubert , T. Baden , P. Berens , T. Euler , and A. L. Vlasits , Center-surround interactions underlie bipolar cell motion sensitivity in the mouse retina, Nat. Commun. 13(1), 5574 (2022)

[49] R. M. Balboa and N. M. Grzywacz , The role of early retinal lateral inhibition: More than maximizing luminance information, Vis. Neurosci. 17(1), 77 (2000)

[50] X. Q. Li , L. Yi , X. M. Yin , J. F. Cheng , Q. Xin , and A. M. Song , Fully screen-printed paper-based ZnO synaptic transistor arrays for visual perception and neuromorphic computing, npj Flexible Electron. 9(1), 57 (2025)

[51] C. Y. Wang , S. J. Liang , S. Wang , P. F. Wang , Z. Li , Z. R. Wang , A. Y. Gao , C. Pan , C. Liu , J. Liu , et al. Gate-tunable van der Waals heterostructure for reconfigurable neural network vision sensor, Sci. Adv. 6(26), eaba6173 (2020)

[52] X. Wan , T. Tsuruoka , and K. Terabe , Neuromorphic system for edge information encoding: Emulating retinal center-surround antagonism by Li-ion-mediated highly interactive devices, Nano Lett. 21(19), 7938 (2021)

[53] A. Gidon , T. A. Zolnik , P. Fidzinski , F. Bolduan , A. Papoutsi , P. Poirazi , M. Holtkamp , I. Vida , and M. E. Larkum , Dendritic action potentials and computation in human layer 2/3 cortical neurons, Science 367(6473), 83 (2020)

Rights & permissions

Higher Education Press