A photoelectric device for internal storage pool calculation and its preparation method

Abstract

The present invention belongs to the field of photoelectric detection technology, and discloses a photoelectric device for in-sensor storage pool calculation and a preparation method thereof. The photoelectric device for in-sensor storage pool calculation comprises: a substrate, a drain electrode is arranged on the substrate, a polymer semiconductor layer is arranged on the drain electrode, the drain electrode and the substrate, a source electrode and a gate electrode are arranged on the polymer semiconductor layer, and an electrolyte is arranged on the source electrode, the gate electrode and the polymer semiconductor layer; the material of the polymer semiconductor layer is an organic mixed ion-electron conductor material. The photoelectric device provided by the present invention can utilize the gate voltage to regulate the dynamic response range of the device, and the device exhibits a reconfigurable nonlinear relaxation characteristic dependent on the gate voltage, which can realize the sensing, memory and in-situ preprocessing functions of the optical signal, and can be widely used in the field of edge artificial vision systems.

Description

Technical Field

[0001] The invention belongs to the technical field of photoelectric detection, and in particular relates to a photoelectric device for in-sensor storage pool calculation and a preparation method thereof. Background Art

[0002] In recent years, the Internet of Things (IoT) has been widely used in the fields of self-driving cars, smart homes, video surveillance, etc., improving people's daily living standards. Edge artificial vision systems can perform real-time video analysis and object detection in complex environments. However, under the current strategy, edge artificial vision systems often need to use traditional digital image sensors based on silicon-based circuits to perceive optical images, use memory to store visual information in the form of electrical signals, and process data through computing units. Data transmission and analog-to-digital conversion between these units lead to high latency and high power consumption. In particular, the failure of Moore's Law will further slow down the development of edge artificial vision systems.

[0003] In order to solve these problems, researchers have carried out a lot of work on the development of bionic neuromorphic vision systems, hoping to achieve a sensor-integrated or near-sensor integrated sensing, storage, and computing architecture. 2 Phototransistors made of two-dimensional layered materials such as (‌MoS2) are the most widely used and have been shown to be able to achieve neuromorphic applications such as recognition, automatic encoding, denoising, and contrast enhancement. However, these works mainly involve preprocessing of static inputs, and more complex functions such as feature extraction and data compression still rely on separate computing networks, which is bound to cause additional delays and system complexity.

[0004] Compared with the above-mentioned traditional deep learning architecture, the reservoir computing (RC) system can directly nonlinearly project the time-related dynamic optical signal data into high-dimensional features, and then combine with a simple linear readout layer to achieve effective feature separation, which can significantly reduce the system training cost, improve the operating energy efficiency, and its scalability on the edge is also higher. However, for most optoelectronic devices, although high responsiveness and wide dynamic response range are achieved, the nonlinear response and short-term memory functions required for building hardware reservoir computing systems still need other devices or circuits to meet, which greatly reduces the integration and brings about processing incompatibility issues. Summary of the invention

[0005] In order to solve the problems existing in the prior art, the purpose of the present invention is to provide a photoelectric device for intra-sensory storage pool computing and a preparation method thereof, wherein the photoelectric device can achieve a wide dynamic response range and adjustable nonlinear feature transformation in the terminal, and has a simple structure and no additional hardware integration requirements. It can be used to build an edge artificial vision system based on a storage pool computing architecture, thereby realizing efficient visual information extraction and processing.

[0006] In order to achieve the above object, the present invention provides the following technical solutions:

[0007] A photoelectric device for in-sense storage pool computing, comprising a substrate, a drain electrode arranged on the substrate, a polymer semiconductor layer arranged on the drain electrode and the substrate, a source electrode and a gate electrode arranged on the polymer semiconductor layer, and an electrolyte arranged on the source electrode, the gate electrode and the polymer semiconductor layer;

[0008] The material of the polymer semiconductor layer is an organic mixed ion-electron conductor material.

[0009] Furthermore, the substrate is silicon dioxide or silicon; the drain and the source are vertically arranged.

[0010] Further, the material of the polymer semiconductor layer is poly(3',4'-bis(2-(2-(2-methoxyethoxy)ethoxy)ethoxy)-2,2':5',2''-thiophene), and the structural formula of poly(3',4'-bis(2-(2-(2-methoxyethoxy)ethoxy)ethoxy)-2,2':5',2''-thiophene) is as follows:

[0011]

[0012]

[0013] The molecular weight of poly(3',4'-bis(2-(2-(2-methoxyethoxy)ethoxy)ethoxy)-2,2':5',2''-thiophene) is 2.25 kDa.

[0014] Furthermore, the drain electrode includes a 2 nm thick chromium layer and a 60 nm thick gold layer disposed on the chromium layer.

[0015] Furthermore, the source electrode is a 50-60 nm thick gold layer, and the gate electrode is a 60 nm thick tungsten layer.

[0016] Furthermore, the polymer semiconductor layer in the overlapping region of the drain and the source is a channel region, and the length of the channel region is 100-200 nm, the width is 100-300 μm, and the thickness is 100-300 μm.

[0017] A method for preparing a photoelectric device for sensing internal storage pool calculation comprises the following steps:

[0018] The drain electrode is prepared on the substrate by using vacuum evaporation technology;

[0019] A polymer semiconductor layer is prepared on the drain electrode and the substrate by using a solution spin coating method;

[0020] A source electrode and a gate electrode are prepared on the polymer semiconductor layer by using vacuum evaporation technology;

[0021] An electrolyte is disposed on the source and gate electrodes and on the polymer semiconductor layer.

[0022] Furthermore, the drain electrode is prepared on the substrate by using vacuum evaporation technology, including the following steps:

[0023] At a vacuum of less than 6.5 × 10 -4 In an evaporation chamber of 1.5 Å / s, a 2 nm thick chromium layer was deposited on the substrate at an evaporation rate of 0.2 Å / s, and a 60 nm thick gold layer was deposited on the chromium layer at an evaporation rate of 1.5 Å / s.

[0024] Furthermore, a source electrode and a gate electrode are prepared on the polymer semiconductor layer by using vacuum evaporation technology, comprising the following steps:

[0025] At a vacuum of less than 6.5 × 10 -4 In a Pa evaporation chamber, a 50-60 nm thick gold layer is deposited on the polymer semiconductor layer at an evaporation rate of 1.5 Å / s to form a source electrode, and the source electrode is arranged vertically to the drain electrode;

[0026] Then, the vacuum degree was lower than 6.5 × 10 -4 In a evaporation chamber at 1.500 Pa, a 60 nm thick tungsten layer was deposited on the polymer semiconductor layer at an evaporation rate of 0.2 Å / s to form a gate.

[0027] The distance between the gate and the drain in the horizontal direction and the distance between the gate and the source in the horizontal direction are both 100 μm.

[0028] Furthermore, the electrolyte is a sodium chloride aqueous solution.

[0029] Compared with the prior art, the present invention has at least the following beneficial effects:

[0030] In the present invention, an organic mixed ion-electron conductor material with a suitable molecular structure is used to prepare a polymer semiconductor layer, the polymer semiconductor layer is arranged on a substrate and a drain electrode, a source electrode and a gate electrode are arranged on the polymer semiconductor layer, and an electrolyte is arranged on the source electrode, the gate electrode and the polymer semiconductor layer to form a photoelectric device for in-sense storage pool calculation. Since the organic mixed ion-electron conductor material has efficient exciton dissociation characteristics and electrochemical doping characteristics, it can realize bulk exciton dissociation and construct an ion-carrier coupling system, which significantly improves the neuromorphic visual performance of the photoelectric device. Compared with phototransistors based on two-dimensional layered materials or perovskite materials, the photoelectric device of the present invention has a stronger photoelectric conversion ability, a larger photocurrent value, and can achieve better feature separation effects for different light stimulus inputs; and the photoelectric device exhibits an intrinsic short-term memory ability for the photocurrent, and can realize a good and adjustable nonlinear feature conversion process in situ without introducing additional devices or circuits, that is, each pixel only needs one transistor, and the structure and preparation process are very simple, which has great advantages in terms of integration. The operating voltage of the photoelectric device provided by the present invention is low (<1 V), and the energy consumption is significantly better than other neuromorphic visual sensors. By adjusting the gate voltage, the optoelectronic device provided by the present invention can operate under various dynamic lighting conditions, and the dynamic response range of the optoelectronic device is comparable to the most advanced silicon-based CMOS (Complementary Metal Oxide Semiconductor) technology level. The above advantages make the optoelectronic device of the present invention very suitable for building a hardware neuromorphic vision system based on a reservoir computing architecture. For complex dynamic light signal input, the optoelectronic device of the present invention can be used as an in-situ integrated sensor and reservoir computing execution unit, which is of great significance to the new generation of intelligent technology innovation of edge artificial vision systems.

[0031] Furthermore, the alkoxy side chains in the molecular structure of poly(3',4'-bis(2-(2-(2-methoxyethoxy)ethoxy)ethoxy)-2,2':5',2''-thiophene) enable the channel region to fully swell in an electrolyte environment, resulting in the presence of many electrolyte solution-polymer semiconductor contact interfaces in the channel region. The band bending at the interface can provide sufficient driving force for exciton dissociation, and the photoelectric device exhibits bulk rather than interface exciton dissociation characteristics, so the dynamic response range of the photoelectric device is large and there will be no rapid saturation of the photocurrent. When the drain voltage is positive, a large number of electrons generated by the exciton dissociation in the channel region will be conducted away by the gold electrode, while the holes will couple with the anions, which causes electrochemical doping in the channel region and thus changes its conductivity. Therefore, the drain current of the photoelectric device can be used to accurately map the characteristics of light stimulation. In addition, the polarons generated by the doping process have a long lifespan, which gives the photoelectric device a short-term memory capability, so that the photocurrent of the photoelectric device exhibits a nonlinear relaxation attenuation characteristic.

[0032] Furthermore, in the optoelectronic device provided by the present invention, adjusting the thickness of the polymer semiconductor layer (i.e., the length of the channel region) can change the spatial electric field distribution and the degree of ion swelling; adjusting the thickness of the source electrode gold layer can change the intensity of light stimulation incident on the channel region; adjusting the size of the source electrode and the drain electrode (i.e., the width and thickness of the channel region) can control the area of ​​the channel region, i.e., the effective area size of the optoelectronic device sensing light stimulation, thereby changing the number of incident photons. By changing the above four parameters, the exciton dissociation ability and electrochemical doping characteristics of the channel region of the optoelectronic device can be regulated, thereby affecting the working performance of the optoelectronic device. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a schematic diagram of the structure of a single optoelectronic device of the present invention;

[0034] Figure 2 is a schematic diagram of an optoelectronic device array;

[0035] Figure 3 The transfer characteristic curve of the photoelectric device prepared in Example 1 under different light intensities;

[0036] Figure 4 The output characteristic curve of the photoelectric device prepared in Example 1 under different light intensities;

[0037] Figure 5 The drain current variation curve of the photoelectric device prepared in Example 1 under different illumination times;

[0038] Figure 6 The drain current variation curve of the photoelectric device prepared in Example 1 under different numbers of light pulse inputs;

[0039] Figure 7 The double pulse facilitated learning curve of the optoelectronic device prepared in Example 1 for a dynamic light signal;

[0040] Figure 8 This is a diagram showing the dynamic response range of the optoelectronic device prepared in Example 1 as a function of the gate voltage;

[0041] Fig. 9 The normalized photocurrent relaxation curve of the photoelectric device prepared in Example 1 under different gate voltages and the same light stimulation conditions;

[0042] Fig.10 A diagram of the photocurrent relaxation time constant of the photoelectric device prepared in Example 1 at different gate voltages;

[0043] Fig.11 A schematic diagram of the working mechanism of the optoelectronic device prepared in Example 1;

[0044] Fig.12 The drain current variation curve of the optoelectronic device prepared in Example 2 under 4-bit light pulse input;

[0045] Fig.13 The initial state and final output current size diagram of the five optoelectronic devices prepared in Example 2 under 4-bit light pulse input;

[0046] Fig.14 is a schematic diagram of a dynamic real-time scanning process for image recognition;

[0047] Fig.15 Graph showing the image recognition accuracy and required hardware cost with and without the participation of the optoelectronic device prepared in Example 2;

[0048] Fig.16 A confusion matrix diagram of the storage pool computing system of the optoelectronic device prepared in Example 2 after training is completed;

[0049] Fig.17 The drain current variation curve of the optoelectronic device prepared in Example 3 under 4-bit light pulse input;

[0050] Fig.18 The initial state and final output current size diagram of the five optoelectronic devices prepared in Example 3 under 4-bit light pulse input;

[0051] Fig.19 Graph showing the image recognition accuracy and required hardware cost with and without the participation of the optoelectronic device prepared in Example 2;

[0052] Fig. 20 A confusion matrix diagram of the storage pool computing system of the optoelectronic device prepared in Example 3 after training is completed;

[0053] In the figure, 1 is a substrate, 2 is a drain, 3 is a polymer semiconductor layer, 4 is a source, 5 is a gate, 6 is an electrolyte, and 7 is a channel region. DETAILED DESCRIPTION

[0054] The present invention provides a photoelectric device for sensing internal storage pool calculation and a preparation method thereof. In order to enable those skilled in the art to more clearly understand the purpose, technical solution and effect of the present invention, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0055] See also Figure 1 The present invention provides a photoelectric device for inductive storage cell computing, comprising: a substrate 1, a drain electrode 2 formed on the substrate 1, a polymer semiconductor layer 3 formed on the drain electrode 2 and on the substrate 1 not covering the drain electrode 2, a source electrode 4 and a gate electrode 5 formed on the polymer semiconductor layer 3, and an electrolyte 6 formed on the source electrode 4 and the gate electrode 5 and on the polymer semiconductor layer 3 not covering the source electrode 4 and the gate electrode 5. In the working state, the source electrode 4 is grounded, a voltage is applied at the drain electrode 2, and a current signal of a channel region 7 can be read, and a voltage is applied at the gate electrode 5, and the photoelectric response characteristics of the photoelectric device can be adjusted.

[0056] The optoelectronic device with in-sense storage pool computing performance provided by the present invention can utilize the gate voltage to regulate the dynamic response range of the device, and the optoelectronic device exhibits a reconfigurable nonlinear relaxation characteristic device that is gate voltage-dependent, which can realize the sensing, memory and in-situ preprocessing functions of optical signals, and can be widely used in the field of edge artificial vision systems.

[0057] Preferably, the substrate 1 is silicon dioxide (SiO 2 ) or silicon (Si), the substrate 1 can provide excellent mechanical stability.

[0058] Preferably, the drain 2 and the source 4 cross each other and are distributed at 90°, the polymer semiconductor layer 3 in the spatially overlapping region of the drain 2 and the source 4 is the channel region 7, the gate 5 exists near the channel region 7, and the horizontal distance between the gate 5 and the drain 2 and the horizontal distance between the gate 5 and the source 4 are both 100 μm.

[0059] Preferably, the drain electrode 2 is composed of a 2 nm thick chromium layer and a 60 nm thick gold layer disposed on the chromium (Cr) layer, the source electrode 4 is composed of a 50-60 nm thick gold (Au) layer, and the gate electrode 5 is composed of a 60 nm thick tungsten layer.

[0060] Preferably, the polymer semiconductor layer 3 is made of poly(3',4'-bis(2-(2-(2-methoxyethoxy)ethoxy)ethoxy)-2,2':5',2''-thiophene) (Pu2gT for short), an organic mixed ion-electron conductor material with efficient exciton dissociation and electrochemical doping properties, to provide a rich photoelectric behavior mechanism. The structural formula of poly(3',4'-bis(2-(2-(2-methoxyethoxy)ethoxy)ethoxy)-2,2':5',2''-thiophene) is as follows:

[0061]

[0062] .

[0063] The molecular weight of poly(3',4'-bis(2-(2-(2-methoxyethoxy)ethoxy)ethoxy)-2,2':5',2''-thiophene) is 2.25 kDa.

[0064] Preferably, the dimensions of the channel region 7 are: length (i.e., thickness of the polymer semiconductor layer 3) of 100-200 nm, width of 100-300 μm, and thickness of 100-300 μm. With this dimension of the channel region 7, a high-density visual sensor can be prepared, which is beneficial to improving the processing efficiency of optical signals.

[0065] Preferably, the dimensions of the gate 5 are: a length of 100 μm and a width of 50 μm.

[0066] The present invention also provides a method for preparing the above-mentioned optoelectronic device for in-sense storage pool calculation, comprising the steps of:

[0067] Step 1, providing a substrate 1;

[0068] Step 2, forming a patterned drain electrode 2 on the substrate 1 by using vacuum evaporation technology;

[0069] Step 3, preparing a polymer semiconductor layer 3 on the drain electrode 2 and on the substrate 1 not covered by the drain electrode 2 by using a solution spin coating method;

[0070] Step 4, preparing a patterned source electrode 4 and gate electrode 5 on the polymer semiconductor layer 3 by vacuum evaporation technology;

[0071] Step 5: preparing an electrolyte 6 on the source electrode 4 and the gate electrode 5 and on the polymer semiconductor layer 3 not covered by the source electrode 4 and the gate electrode 5 .

[0072] Preferably, in step 2, the process parameters for preparing the patterned drain electrode 2 on the substrate 1 by vacuum evaporation technology are:

[0073] At a vacuum of less than 6.5 × 10-4 In a vapor deposition chamber of 1000 nm, a 2 nm thick chromium layer is deposited on the substrate 1 at a vapor deposition rate of 0.2 Å / s, and then a 60 nm thick gold layer is deposited on the chromium layer at a vapor deposition rate of 1.5 Å / s, thereby forming a drain 2, wherein the chromium layer plays a role in strengthening the bonding force between the gold layer and the substrate 1;

[0074] Preferably, in step 4, the process parameters for preparing the patterned source 4 and gate 5 on the polymer semiconductor layer 3 by vacuum evaporation technology are:

[0075] At a vacuum of less than 6.5 × 10 -4 In the evaporation chamber of Pa, a 50-60 nm thick gold layer is deposited on the polymer semiconductor layer 3 at an evaporation rate of 1.5 Å / s to form the source 4. At a faster evaporation rate, the short circuit phenomenon caused by excessive penetration of the gold layer into the polymer semiconductor layer 3 can be effectively curbed.

[0076] Then, the vacuum degree was lower than 6.5 × 10 -4 In the evaporation chamber of Pa, a 60 nm thick tungsten layer is deposited on the polymer semiconductor layer 3 at an evaporation rate of 0.2 Å / s to form the gate 5. At a slower evaporation rate, the density of the electrode can be guaranteed to obtain stable chemical properties, and tungsten will not excessively penetrate into the polymer semiconductor layer 3, which can effectively avoid short circuits between electrodes.

[0077] Preferably, in step 3, the specific process of the solution spin coating method is: drop the Pu2gT solution on the substrate 1, immediately rotate at a speed of 1500 rpm for 60 seconds to form a uniform polymer film, and then place the polymer film in a vacuum environment for 20 minutes to completely evaporate the solvent. The solution spin coating method can form a uniform polymer semiconductor layer 3.

[0078] Preferably, in step 5, the electrolyte is dripped onto the channel region 7 and the gate 5 using a dispensing machine.

[0079] Preferably, in step 5, the electrolyte 6 is a sodium chloride aqueous solution. Pu2gT can fully swell in the sodium chloride aqueous solution at room temperature, so that ion-carrier coupling can occur in the channel region 7, which can well exert the working characteristics of the polymer semiconductor material and thus enrich the photoelectric conversion behavior of the device.

[0080] In order to facilitate those skilled in the art to understand the present invention, more specific and detailed embodiments are provided herein.

[0081] Example 1

[0082] In this embodiment, a photoelectric device for in-sense storage pool computing is provided, comprising: a substrate 1, a drain electrode 2 formed on the substrate, a uniform polymer semiconductor layer 3 formed on the drain electrode 2 and on the substrate 1 not covered by the drain electrode 2, a source electrode 4 and a gate electrode 5 formed on the polymer semiconductor layer 3, and an electrolyte 6 (NaCl aqueous solution) formed on the source electrode 4 and the gate electrode 5 and on the polymer semiconductor layer 3 not covered by the source electrode 4 and the gate electrode 5. The polymer semiconductor layer 3 existing in the spatial cross-overlapping region of the drain electrode 2 and the source electrode 4 is a channel region 7, and the channel region 7 has a length of 150 nm, a width of 200 μm, and a thickness of 200 μm.

[0083] Among them, substrate 1 is SiO 2 .

[0084] The specific preparation method of the optoelectronic device for sensing internal storage cell computing is as follows:

[0085] Step 1, cleaning the substrate 1: placing the substrate 1 in detergent water, deionized water, acetone and isopropanol in turn and ultrasonically cleaning them for 30 minutes respectively.

[0086] Step 2, the drain 2 is formed on the clean substrate 1 by vacuum evaporation technology: first, a 2 nm thick Cr layer is deposited at an evaporation rate of 0.2 Å / s as an adhesion layer to enhance the bonding force between the Au layer and the substrate 1, and then a 60 nm thick Au layer is deposited at an evaporation rate of 1.5 Å / s to form the drain 2. The specific pattern of the drain 2 is determined by the mask.

[0087] Step 3, prepare Pu2gT solution: add Pu2gT into chloroform at room temperature and stir for 12 hours to obtain Pu2gT solution with a concentration of 17 mg / ml.

[0088] Step 4, processing the polymer semiconductor layer 3 by solution spin coating: drop the Pu2gT solution on the substrate 1 deposited with the drain electrode 2, immediately rotate it at 1500 rpm for 60 s to form a uniform polymer film, and then place it in a vacuum environment for 20 min to allow the solvent to completely evaporate to form a polymer semiconductor layer 3.

[0089] Step 5, the source electrode 4 is formed on the polymer semiconductor layer 3 by vacuum evaporation technology: a 55 nm thick Au layer is deposited on the polymer semiconductor layer 3 at an evaporation rate of 1.5 Å / s to form the source electrode 4, and the specific pattern of the source electrode 4 is determined by the mask.

[0090] Step 6: The gate 5 is formed on the polymer semiconductor layer 3 by vacuum evaporation technology: a 60 nm thick tungsten layer is deposited on the polymer semiconductor layer 3 at an evaporation rate of 0.2 Å / s to form the gate 5. The specific pattern of the gate 5 is determined by the mask.

[0091] Step 7, prepare a 0.1 mol / L NaCl aqueous solution as the electrolyte 6, and drip the NaCl aqueous solution at the intersection of the source 4 and the drain 2 by a dispensing machine, covering the channel area 7 and the gate 5, to form a photoelectric device for in-storage cell computing.

[0092] A top view of a 10 × 10 array model of optoelectronic devices used for in-sense reservoir calculations as described above can be found at Figure 2 .

[0093] The gate 5 of the optoelectronic device in this embodiment can simulate the function of horizontal cells in the retina of the human eye, and the role of the channel region 7 is similar to that of light receiving cells such as cones and rods. During the operation of the optoelectronic device, the coupling (doping) and decoupling (de-doping) process between ions and carriers is similar to the disappearance and generation of photopigments in light receiving cells of the human eye.

[0094] This embodiment focuses on the working characteristics and working mechanism of the optoelectronic device, wherein the wavelength of the optical signal used to stimulate the optoelectronic device is 455 nm, and the intensity varies with the test requirements.

[0095] In this example, the transfer characteristic curves of the photoelectric device under various light intensities are tested. For the test results, please refer to Figure 3 It can be found that the photoelectric device exhibits obvious p-type switching behavior, and as the light intensity increases, the threshold voltage of the photoelectric device V th The positive shift continues, and the drain current I D It will also gradually increase, which indicates that the carrier density in the channel region 7 has a gain that is positively correlated with the light intensity. The above phenomenon can be attributed to the coupling of anions and holes in the channel region 7 under light. For the optoelectronic device, light stimulation acts as an equivalent negative gate voltage, enabling effective electrochemical doping in the channel region 7. When other conditions are the same, the doping density depends on the light intensity.

[0096] In this embodiment, the output characteristic curve of the photoelectric device under various light intensities is tested. For the test results, please refer to Figure 4 Similar to the transfer characteristic curve test results above, it can be found that the drain current I D This indicates that the optoelectronic device has real-time sensing performance for optical signals and can effectively distinguish the light intensity.

[0097] In this embodiment, the change of drain current of the photoelectric device under the stimulation of light pulses of different durations is tested. For the test results, please refer to Figure 5 The light intensity is 50 mW / cm 2, gate voltage V G is 0 V, the drain voltage V D When stimulated by a light signal, the photoelectric device exhibits an obvious nonlinear dynamic light response, and the drain current of the photoelectric device I D It does not saturate quickly under light stimulation, but increases continuously with the extension of the duration of the light pulse, thus achieving effective separation of different time features.

[0098] In this embodiment, the change of drain current of the photoelectric device under the stimulation of light pulses with different pulse numbers is tested. For the test results, please refer to Figure 6 . Wherein, the gate voltage V G is 0 V, the drain voltage V D The optical pulse intensity is 50 mW / cm 2 , the duration of the light pulse τ on The interval between light pulses is 200 ms. τ off It can be found that the response behavior of the photoelectric device under multiple light pulse stimulation can be regarded as the cumulative superposition of its single light pulse response behavior on a time scale. As the frequency of light stimulation increases, the drain current of the photoelectric device I D It also gradually increases, realizing the separation of features of different frequencies.

[0099] exist Figure 5 and Figure 6 In the two photoelectric conversion behavior tests shown, it can be found that the photoelectric device has a short-term memory ability for the photoelectric response behavior. After the light stimulus is removed, the drain current of the photoelectric device I D It does not return to the initial level instantly, but decays slowly in the form of nonlinear relaxation. That is, after the light is removed, it can still be based on the drain current. I D The size effectively distinguishes light inputs of different characteristics. This characteristic enables the optoelectronic device to well implement storage pool calculation of optical signals at the hardware level.

[0100] The short-term memory capacity of the optoelectronic device was evaluated by paired-pulse facilitation (PPF, short-term learning and memory mechanism of synapses in the visual system of organisms) test. The PPF factor is defined as follows:

[0101]

[0102] in, is the PPF factor, ΔI 1 and ΔI 2 The drain current after two consecutive light stimulations is I D The difference from the initial value (i.e., the photocurrent I ph ), PPF factor size and the interval between two light stimulations ∆t There is a double exponential relationship, namely:

[0103]

[0104] Among them, τ PPF is the time constant, the time constant τ PPF Can be used to indicate the short-term memory capacity of optoelectronic devices, the time constant τ PPF The larger the value, the better the memory ability. In this embodiment, the light intensity is selected to be 50 mW / cm 2 , a light pulse stimulus with a duration of 200 ms is used as input, and the interval between stimulus signals is changed by adjusting the duty cycle. ∆t For exponential relationships, see Figure 7 ,from Figure 7 Extract the time constant τ of the optoelectronic device PPF It is 0.59 s, which is comparable to the level of biological organisms and can meet the calculation needs of the storage pool.

[0105] In this embodiment, the optoelectronic device can realize the sensing, memory and in-situ preprocessing functions of the optical signal.

[0106] The following focuses on the reconfigurable characteristics of optoelectronic devices.

[0107] The light intensity recognition capability of the photoelectric device in this embodiment can be determined by applying a voltage to the gate 5. V G Define the dynamic perception range (DPR):

[0108]

[0109] in, I max and I min Indicates the maximum and minimum photocurrent generated within a specific light intensity range. For the optoelectronic device, when the drain voltage V D At 0.4 V, the drain current I DIt increases monotonically with the increase of light intensity, so it is taken as 50 mW / cm 2 Photocurrent under light stimulation I ph The value is the maximum current I max (duration is 2 s), take its dark state current I dark The value is the minimum current I min The photoelectric device has a gate voltage V G For DPR test results in the range of -0.2 V to 0.25 V, see Figure 8 , DPR reaches a maximum value of 56 dB near 0.15 V.

[0110] In addition to the dynamic response range, the optoelectronic device in this embodiment can also V G Controlling the photocurrent after light removal I ph A series of gate voltages of different magnitudes are applied to the photoelectric device. V G , extract the 2 , duration 200 ms) after I ph Attenuation process, the obtained curve can be seen in Fig. 9 When the gate voltage V G When the photocurrent of the photoelectric device gradually increases from - 0.15 V to 0.15 V, I ph The rate of nonlinear relaxation gradually increases. The stretched exponential Kohlrausch function is used to quantitatively analyze this decay process:

[0111]

[0112] Among them, 𝜏 K represents the relaxation time constant, and 𝛽 represents the stretching exponent. The relaxation time constant 𝜏 K Together with the stretching exponent 𝛽, it explains the stretching exponential behavior, representing the relaxation process with different relaxation times. The results of the exponential fitting can be found in Fig.10 , as the gate voltage V G Change, relaxation time constant 𝜏 K The gate voltage of the photoelectric device decreases gradually from 2.80 s to 0.43 s. V GControl of ion-carrier coupling dynamics.

[0113] The working principle of the optoelectronic device in this embodiment can be explained by the energy level diagram. The material of the channel region 7 of the optoelectronic device is Pu2gT. The energy level diagram of Pu2gT can be found in Fig.11 The alkoxy side chains in the Pu2gT molecular structure enable the channel region 7 to fully swell in the electrolyte 6 environment, resulting in the presence of many electrolyte solution-polymer semiconductor contact interfaces in the channel region 7. The energy band bending at the interface can provide sufficient driving force for exciton dissociation. The optoelectronic device exhibits bulk rather than interface exciton dissociation characteristics, so the optoelectronic device has a large dynamic response range and does not produce photocurrent. I ph When the drain voltage V D When the value is positive, a large number of electrons generated by the dissociation of excitons in the channel region 7 will be conducted away by the Au electrode, and the holes will couple with the anions, which will cause electrochemical doping in the channel region 7 and thus change its conductivity. I D This can achieve accurate mapping of light stimulation characteristics. The polaritons generated by the doping process have a long lifespan, which gives the optoelectronic device short-term memory capability, making the photocurrent of the optoelectronic device I ph The gate voltage can be adjusted V G The ion-carrier coupling dynamics in the channel region 7 is effectively regulated, thereby giving the optoelectronic device a reconfigurable characteristic. The above working principle and working characteristics make the optoelectronic device very suitable for constructing an edge artificial vision system with integrated sensing, storage and computing based on a storage pool computing architecture.

[0114] Example 2

[0115] In this embodiment, a photoelectric device for in-sense storage cell computing is provided, comprising: a substrate 1, a drain electrode 2 formed on the substrate, a uniform polymer semiconductor layer 3 formed on the drain electrode 2 and on the substrate 1 not covered by the drain electrode 2, a source electrode 4 and a gate electrode 5 formed on the polymer semiconductor layer 3, and an electrolyte 6 formed on the source electrode 4 and the gate electrode 5 and on the polymer semiconductor layer 3 not covered by the source electrode 4 and the gate electrode 5. The polymer semiconductor layer 3 existing in the spatial cross-overlapping region of the drain electrode 2 and the source electrode 4 is a channel region 7, and the channel region 7 has a length of 300 nm, a width of 300 μm, and a thickness of 300 μm.

[0116] Wherein, the substrate 1 is Si.

[0117] The specific preparation method of the optoelectronic device for sensing internal storage cell computing is as follows:

[0118] Step 1, cleaning the substrate 1: placing the substrate 1 in detergent water, deionized water, acetone and isopropanol in turn and ultrasonically cleaning them for 30 minutes respectively.

[0119] Step 2, the drain 2 is formed on the clean substrate 1 by vacuum evaporation technology: first, a 2 nm thick Cr layer is deposited at an evaporation rate of 0.2 Å / s as an adhesion layer to enhance the bonding force between the Au layer and the substrate 1, and then a 60 nm thick Au layer is deposited at an evaporation rate of 1.5 Å / s to form the drain 2. The specific pattern of the drain 2 is determined by the mask.

[0120] Step 3, prepare Pu2gT solution: add Pu2gT to chloroform at room temperature and stir for 12 hours to obtain Pu2gT solution with a concentration of 20 mg / ml.

[0121] Step 4, processing the polymer semiconductor layer 3 by solution spin coating: drop the Pu2gT solution on the substrate 1 deposited with the drain electrode 2, immediately rotate it at 1500 rpm for 60 s to form a uniform polymer film, and then place it in a vacuum environment for 20 min to allow the solvent to completely evaporate to form a polymer semiconductor layer 3.

[0122] In step 5, the source electrode 4 is formed on the polymer semiconductor layer 3 by vacuum evaporation technology: a 50 nm thick Au layer is deposited on the polymer semiconductor layer 3 at an evaporation rate of 1.5 Å / s to form the source electrode 4. The specific pattern of the source electrode 4 is determined by the mask.

[0123] Step 6: The gate 5 is formed on the polymer semiconductor layer 3 by vacuum evaporation technology: a 60 nm thick tungsten layer is deposited on the polymer semiconductor layer 3 at an evaporation rate of 0.2 Å / s to form the gate 5. The specific pattern of the gate 5 is determined by the mask.

[0124] Step 7, prepare a 0.1 mol / L sodium chloride aqueous solution as the electrolyte 6, and drop it at the intersection of the source 4 and the drain 2 by a dispensing machine, covering the channel area 7 and the gate 5, to form a photoelectric device for in-storage cell computing.

[0125] In this embodiment, the image recognition application of the sensory storage pool computing system based on the provided optoelectronic device is emphasized. The wavelength of the light signal used to stimulate the optoelectronic device is 455 nm, and the light intensity varies with the test requirements.

[0126] The optoelectronic device in this embodiment can effectively distinguish 16 dynamic optical pulse sequences from (0000) to (1111), and the channel current ID See the change history for details. Fig.12 According to the method of this embodiment, five different devices are prepared. The initial state and the final output channel current extracted from the five different devices are I D See also Fig.13 The gate voltage of the photovoltaic device described here is V G is 0 V, the drain voltage V D is 0.4 V, and “0” in the input dynamic light signal represents 0 mW / cm 2 , a light pulse with a duration of 2 seconds, "1" represents an intensity of 50 mW / cm 2 , a light pulse with a duration of 2 s. After the photoelectric device receives the pulse "1", the channel current is detected. I D Increase, and detect the channel current after receiving the pulse "0" I D After passing through four input nodes, the optoelectronic device exhibits clearly distinguishable final-state output currents, which indicates that the optoelectronic device has successfully achieved feature extraction and data dimensionality reduction.

[0127] Based on the above working characteristics, this embodiment uses the photoelectric device to realize the handwritten digit recognition function. The MNIST (Mixed National Institute of Standards and Technology) data set is used to construct the image recognition task, which contains 60,000 handwritten digit images for training and 10,000 handwritten digit images for testing. First, the 28 × 28 pixel grayscale image is preprocessed to convert it into a binary image, where the pixels "0" and "1" represent the light intensity of 0 mW / cm 2 and 50 mW / cm 2 In order to convert the spatial information of the image into temporal optical information, the present embodiment divides the binary image into 7 equal intervals of 4 time steps, each time step is 2 s, and uses the photoelectric device array of 28 × 1 scale to construct the hardware reservoir layer, where each photoelectric device processes 7 intervals, i.e. 28 time steps of dynamic light pulse input. For a specific example, please refer to Fig.14 The channel current of each device at the end of each interval is I D Sampled and recorded as virtual node state, channel current I DThe feature information extracted from the dynamic optical input is then fed into a fully connected layer of size 196 × 10 for training and testing. The simulation training results and confusion matrix can be found in Fig.15 , Fig.16 , Fig.15 In the above, with RC layer is an image recognition system with the participation of the optoelectronic device prepared in Example 2, single ANN is an image recognition system without the participation of the optoelectronic device prepared in Example 2, Accuracy is the image recognition accuracy after training, Number of weights is the hardware cost required for the image recognition system, Fig.16 In the figure, Correct Out Didit is the correct output digit, and inferred Output Digit is the inferred output digit. The optoelectronic device in this embodiment can converge the recognition accuracy to about 86% after 7 training epochs. Compared with the image recognition system without the optoelectronic device prepared in this embodiment, the image recognition system with the optoelectronic device prepared in this embodiment has a smaller full-layer array scale (1960 vs 78410) and a faster convergence speed (7 epochs (training steps) vs 15 epochs) without the need to integrate additional sensor units. It can significantly reduce the training cost and improve the operating energy efficiency without affecting the recognition accuracy (86% vs 90%), thanks to the excellent feature extraction capability of the optoelectronic device.

[0128] Example 3

[0129] In this embodiment, a photoelectric device for inductive storage cell computing is provided, comprising: a substrate 1, a drain electrode 2 formed on the substrate, a uniform polymer semiconductor layer 3 formed on the drain electrode 2 and on the substrate 1 not covered by the drain electrode 2, a source electrode 4 and a gate electrode 5 formed on the polymer semiconductor layer 3, and an electrolyte 6 formed on the source electrode 4 and the gate electrode 5 and on the polymer semiconductor layer 3 not covered by the source electrode 4 and the gate electrode 5. The polymer semiconductor layer 3 existing in the spatial cross-overlapping region of the drain electrode 2 and the source electrode 4 is a channel region 7, and the channel region 7 has a length of 100 nm, a width of 100 μm, and a thickness of 100 μm.

[0130] Among them, substrate 1 is SiO 2 .

[0131] The specific preparation method of the optoelectronic device for sensing internal storage cell computing is as follows:

[0132] Step 1, cleaning the substrate 1: placing the substrate 1 in detergent water, deionized water, acetone and isopropanol in turn and ultrasonically cleaning them for 30 minutes respectively.

[0133] Step 2, the drain 2 is formed on the clean substrate 1 by vacuum evaporation technology: first, a 2 nm thick Cr layer is deposited at an evaporation rate of 0.2 Å / s as an adhesion layer to enhance the bonding force between the Au layer and the substrate 1, and then a 60 nm thick Au layer is deposited at an evaporation rate of 1.5 Å / s to form the drain 2. The specific pattern of the drain 2 is determined by the mask.

[0134] Step 3, prepare Pu2gT solution: add Pu2gT to chloroform at room temperature and stir for 12 hours to obtain Pu2gT solution with a concentration of 10 mg / ml.

[0135] Step 4, processing the polymer semiconductor layer 3 by solution spin coating: drop the Pu2gT solution on the substrate 1 deposited with the drain electrode 2, immediately rotate it at 1500 rpm for 60 s to form a uniform polymer film, and then place it in a vacuum environment for 20 min to allow the solvent to completely evaporate to form a polymer semiconductor layer 3.

[0136] In step 5, the source electrode 4 is formed on the polymer semiconductor layer 3 by vacuum evaporation technology: a 60 nm thick Au layer is deposited on the polymer semiconductor layer 3 at an evaporation rate of 1.5 Å / s to form the source electrode 4. The specific pattern of the source electrode 4 is determined by the mask.

[0137] Step 6: The gate 5 is formed on the polymer semiconductor layer 3 by vacuum evaporation technology: a 60 nm thick tungsten layer is deposited on the polymer semiconductor layer 3 at an evaporation rate of 0.2 Å / s to form the gate 5. The specific pattern of the gate 5 is determined by the mask.

[0138] Step 7, use 0.1 mol / L sodium chloride aqueous solution as electrolyte 6, and drip it at the intersection of source 4 and drain 2 by dispensing machine, covering channel area 7 and gate 5, to form a photoelectric device for in-storage cell computing.

[0139] Similar to Example 2, this example focuses on the image recognition application of the sensory storage pool computing system based on the provided optoelectronic device. The wavelength of the light signal used to stimulate the optoelectronic device is 455 nm, and the light intensity varies with the test requirements.

[0140] The optoelectronic device in this embodiment can effectively distinguish 16 dynamic optical pulse sequences from (0000) to (1111), and the channel current I D See the change history for details. Fig.17 , the initial state and the final output channel current extracted from the five different devices prepared in this embodiment I D See also Fig.18As in Example 2, the gate voltage of the photoelectric device described here is V G is 0 V, the drain voltage V D is 0.4 V, and “0” in the input dynamic light signal represents 0 mW / cm 2 , a light pulse with a duration of 2 s, “1” represents an intensity of 50 mW / cm 2 , a light pulse with a duration of 2 s. After passing through 4 input nodes, the optoelectronic device also successfully achieves feature extraction and data dimension reduction through the clearly distinguishable final state output current. It should be noted that the photocurrent of the optoelectronic device in this embodiment is I ph The level is significantly lower than that of the photoelectric device in Example 2, and the photocurrent I ph The relaxation time constant 𝜏 K This is because the channel region 7 of the optoelectronic device in this embodiment has smaller length, width and thickness than the optoelectronic device in Example 2, but the Au layer thickness of the source electrode 4 is larger. Therefore, compared with the optoelectronic device in Example 2, the channel region 7 of the optoelectronic device in this embodiment swells more weakly in the electrolyte 6 environment, and the number of ions entering the channel region 7 is less; and when subjected to the same light stimulus, the number of photons incident on the channel region 7 of the optoelectronic device in this embodiment is less, that is, the number of dissociated excitons is less. These factors affect the ion-carrier coupling dynamics process, and thus change the photoelectric response behavior of the device.

[0141] Based on the above working characteristics, this embodiment uses the photoelectric device to realize the handwritten digit recognition function. As in Example 2, the MNIST data set is used to construct the image recognition task. First, the 28 × 28 pixel grayscale image is converted into a binary image through preprocessing, where the pixels "0" and "1" represent the light intensity of 0 mW / cm 2 and 50 mW / cm 2 In order to convert the spatial information of the image into temporal optical information, this embodiment divides the binary image into 7 equal intervals of 4 time steps, each time step is 2 s, and uses a 28 × 1 optoelectronic device array to construct a hardware storage pool layer, where each optoelectronic device processes 7 intervals, i.e. 28 time steps of dynamic optical pulse input. The channel current of each device at the end of each interval is I D Sampled and recorded as virtual node state, channel current I DThe feature information extracted from the dynamic optical input is then fed into a fully connected layer of size 196 × 10 for training and testing. The simulation training results and confusion matrix can be found in Fig.19 , Fig. 20 , the optoelectronic device in this embodiment can converge the recognition accuracy to about 85% after 10 training epochs. Compared with the image recognition system without the optoelectronic device prepared in this embodiment, the image recognition system with the optoelectronic device prepared in this embodiment has a smaller full-layer array scale (1960 vs 78410) and a faster convergence speed (10 epochs vs 15epochs) without the need to integrate additional sensor units, and can significantly reduce the training cost and improve the operating energy efficiency without affecting the recognition accuracy (85% vs 90%), thanks to the excellent feature extraction ability of the optoelectronic device. It should be pointed out that due to the differences in key parameters such as the size of the channel region 7 and the thickness of the Au layer of the source 4, the accuracy and convergence speed of the image recognition system provided in this embodiment are slightly worse than those of Example 2.

[0142] The more complex the recognition task to be processed, the greater the advantages of the storage pool computing system based on the optoelectronic devices in the present invention in terms of energy efficiency and other aspects. This makes it likely to become an effective way to implement a new generation of intelligent edge artificial vision systems, which is of great significance to the further development of the Internet of Things.

[0143] The specific implementation methods described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A photoelectric device for sensing internal storage pool calculation, characterized in that: The invention comprises a substrate (1), a drain electrode (2) being arranged on the substrate (1), a polymer semiconductor layer (3) being arranged on the drain electrode (2) and the substrate (1), a source electrode (4) and a gate electrode (5) being arranged on the polymer semiconductor layer (3), and an electrolyte (6) being arranged on the source electrode (4), the gate electrode (5) and the polymer semiconductor layer (3); The material of the polymer semiconductor layer (3) is an organic mixed ion-electron conductor material; The substrate (1) is silicon dioxide or silicon; the drain (2) and the source (4) are arranged vertically; The polymer semiconductor layer (3) in the overlapping region of the drain electrode (2) and the source electrode (4) is a channel region (7); The material of the polymer semiconductor layer (3) is poly(3',4'-bis(2-(2-(2-methoxyethoxy)ethoxy)ethoxy)-2,2':5',2''-thiophene), and the structural formula of poly(3',4'-bis(2-(2-(2-methoxyethoxy)ethoxy)ethoxy)-2,2':5',2''-thiophene) is as follows: The molecular weight of poly(3',4'-bis(2-(2-(2-methoxyethoxy)ethoxy)ethoxy)-2,2':5',2''-thiophene) is 2.25 kDa.

2. The optoelectronic device for in-sense storage pool computing according to claim 1, characterized in that: The drain electrode (2) comprises a 2 nm thick chromium layer and a 60 nm thick gold layer disposed on the chromium layer.

3. The optoelectronic device for in-sense storage pool computing according to claim 1, characterized in that: The source electrode (4) is a 50-60 nm thick gold layer, and the gate electrode (5) is a 60 nm thick tungsten layer.

4. The optoelectronic device for in-sense storage pool computing according to claim 1, characterized in that: The channel region (7) has a length of 100 - 200 nm, a width of 100 - 300 μm, and a thickness of 100 - 300 μm.

5. A method for preparing a photoelectric device for in-sense storage cell computing according to any one of claims 1 to 4, characterized in that: The following steps are involved: A drain electrode (2) is prepared on a substrate (1) by using a vacuum evaporation technique; A polymer semiconductor layer (3) is prepared on the drain electrode (2) and the substrate (1) by using a solution spin coating method; A source electrode (4) and a gate electrode (5) are prepared on the polymer semiconductor layer (3) by using a vacuum evaporation technique; An electrolyte (6) is arranged on the source electrode (4) and the gate electrode (5) and on the polymer semiconductor layer (3).

6. The method for preparing an optoelectronic device for in-sense storage cell computing according to claim 5, characterized in that: A drain electrode (2) is prepared on a substrate (1) using a vacuum evaporation technique, comprising the following steps: At a vacuum of less than 6.5 × 10 -4 In a vapor deposition chamber at 1.5 Å / s, a 2 nm thick chromium layer is deposited on the substrate (1) at a vapor deposition rate of 0.2 Å / s, and a 60 nm thick gold layer is deposited on the chromium layer at a vapor deposition rate of 1.5 Å / s.

7. The method for preparing an optoelectronic device for in-sense storage cell computing according to claim 5, characterized in that: A source electrode (4) and a gate electrode (5) are prepared on a polymer semiconductor layer (3) by using a vacuum evaporation technique, comprising the following steps: At a vacuum of less than 6.5 × 10 -4 In a vapor deposition chamber of Pa, a 50-60 nm thick gold layer is deposited on the polymer semiconductor layer (3) at a vapor deposition rate of 1.5 Å / s to form a source electrode (4), and the source electrode (4) is arranged vertically with the drain electrode (2); Then, the vacuum degree was lower than 6.5 × 10 -4 In a vapor deposition chamber of Pa, a 60 nm thick tungsten layer is deposited on the polymer semiconductor layer (3) at an evaporation rate of 0.2 Å / s to form a gate (5).

8. The method for preparing an optoelectronic device for in-sense storage cell computing according to claim 5, characterized in that: The electrolyte (6) is an aqueous solution of sodium chloride.

Images