Efficient corrosion method for monolithic gallium arsenide substrate

Abstract

The invention discloses an efficient corrosion method of a monolithic gallium arsenide substrate, which comprises the following steps: placing a gallium arsenide substrate wafer in a corrosive liquid and continuously moving in the soaking process to inhibit bubble attachment; the soaked wafer rotates around the axis of the wafer, mixed liquid medicine of NHOH and HO is continuously supplied to the surface of the wafer at the same time, and corrosion by-products are removed in real time by means of centrifugal force generated by rotation and liquid medicine scouring force; wafer surface images are continuously obtained in the rotary corrosion process, a gallium arsenide residual area is recognized through an image processing algorithm, and medicine liquid supply is immediately stopped when it is judged that the area of the area reaches a preset end point condition; washing with ultrapure water, washing with two fluids mixed by ultrapure water and nitrogen, and washing with ultrapure water again so as to remove residual liquid medicine and by-products on the surface; the liquid on the surface of the wafer is removed by increasing the rotation speed of the wafer in a stepped manner and cooperating with nitrogen purging, and drying is completed.

Description

TECHNICAL FIELD

[0001] The present application relates to a high-efficiency etching method of monolithic gallium arsenide substrate. BACKGROUND

[0002] Gallium arsenide (GaAs) is widely used in high-frequency radio frequency devices, optoelectronic devices and other fields as a representative of the second generation of semiconductor materials. Wet etching is a key process step in the manufacture of gallium arsenide devices, which is used to realize substrate thinning, mesa isolation, patterned structure fabrication, etc. Gallium arsenide substrate thinning usually adopts wet bench etching, that is, the wafer is immersed in a tank containing etching solution for processing. This method has many drawbacks:

[0003] The existing production line generally adopts tank-type wet etching: multiple wafers are simultaneously immersed in NH4OH-H2O2 etching solution, and the material is removed by chemical reaction. This method has the following inherent problems:

[0004] 1. Poor uniformity - the concentration and temperature of the solution in the tank show a gradient distribution, and the etching rate varies greatly between wafers and within a wafer;

[0005] 2. Low efficiency - etching by-products easily adhere to the wafer surface, hindering the contact of fresh chemicals, resulting in a decrease in rate;

[0006] 3. Poor controllability - the etching rate is significantly affected by environmental temperature and solution aging, and the process repeatability is insufficient;

[0007] 4. Poor safety - a large amount of volatile chemicals are exposed to an open environment, posing potential risks to operators and the factory.

[0008] To improve uniformity, some research has proposed the concept of single wafer spin etching: fresh chemicals are sprayed under wafer rotation, and by-products are immediately removed by centrifugal force. However, this approach is not mature in gallium arsenide technology, and the main problems are:

[0009] By-products are still not completely removed, and local "black spots" often remain at the end of etching;

[0010] End point determination relies on fixed time or manual experience, lacks real-time monitoring means, and is prone to over-etching or incomplete etching;

[0011] The process sequence is single, and does not form a complete parameterized process with pre-soaking wetting, multi-stage cleaning and drying, making it difficult to meet the production rhythm of batch production.

[0012] In summary, there is still a lack of a wet etching process method for gallium arsenide substrate with high uniformity, high efficiency, high end point accuracy and clear parameter window in the art. SUMMARY

[0013] The present application provides a single wafer GaAs substrate high-efficiency etching method to solve the problems of the prior art.

[0014] The technical scheme adopted by the present application is:

[0015] A single wafer GaAs substrate high-efficiency etching method comprises:

[0016] (1) soaking treatment: the GaAs substrate wafer is placed in the etching solution and continuously moved during the soaking process to inhibit bubble adhesion;

[0017] (2) single wafer rotation etching: the soaked wafer is rotated around its axis, and NH4OH and H2O2 mixed solution is continuously supplied to the wafer surface, and the centrifugal force generated by rotation and the scouring force of the solution are used to remove the etching by-products in real time;

[0018] (3) real-time endpoint detection: the wafer surface image is continuously acquired during the rotation etching process, the residual GaAs area is identified through image processing algorithm, and the solution supply is immediately terminated when it is determined that the area reaches the preset endpoint condition;

[0019] (4) three-stage cleaning: the wafer surface is rinsed with ultrapure water, then with a two-fluid mixture of ultrapure water and nitrogen, and then with ultrapure water again to remove residual solution and by-products on the surface;

[0020] (5) stepwise rotary drying: the wafer rotation speed is increased step by step and nitrogen is blown to remove the liquid on the wafer surface to complete the drying.

[0021] Further, in step (1), the continuous movement is an up-down lifting movement, the lifting frequency is 5-15 Hz, the lifting amplitude is 10-30 mm, and the etching solution temperature is 22-25℃.

[0022] Further, in step (2), the volume ratio of NH4OH and H2O2 mixed solution is 1:5-1:7, the solution temperature is 22-40℃, the supply flow rate is 400-600 ml / min, and the wafer rotation speed is 200-500 rpm.

[0023] Further, in step (2), while the wafer is being etched, a 0.8-1.2 MHz megasonic wave is applied coaxially, the ultrasonic power density is 15-25 W / L, when the wafer rotation speed is 200-300 rpm, the megasonic wave frequency corresponds to 0.8-1.0 MHz; when the rotation speed is 300-500 rpm, the megasonic wave frequency corresponds to 1.0-1.2 MHz, and the power density remains 15-25 W / L.

[0024] Further, step (3) comprises:

[0025] Visual detection: using improved YOLO-OBB rotating target detection framework combined with HSV color screening, Kmeans clustering and circle fitting, output residual area A(t); the improved framework is to insert CARAFE up-sampling and RFCA-Conv rotating anchor frame calibration convolution at the end of the traditional YOLO-OBB backbone, to generate a rotating frame mask of the residual area of gallium arsenide in real time;

[0026] Conductivity detection: real-time acquisition of the conductivity change rate dσ / dt of the etching cavity discharge;

[0027] Weighted fusion: after normalizing the residual area A(t) and the conductivity change rate dσ / dt, calculate S(t)=α·A(t)+β·(dσ / dt), α=0.6~0.8, β=0.2~0.4, when S(t) is continuously lower than the preset threshold, it is determined as the end point.

[0028] Further, the visual detection comprises:

[0029] a) After resizing the image, the YOLO-OBB framework with CARAFE up-sampling and RFCA-Conv rotating anchor frame calibration convolution is used to output the rotating frame coordinates of the residual area of gallium arsenide;

[0030] b) Based on the rotating frame output by the improved YOLO-OBB rotating target detection framework, draw ROI and crop the image;

[0031] c) Convert the cropped image to HSV color space, generate a binary mask of the residual area of gallium arsenide by InRange screening, and calculate the area of the mask;

[0032] d) Kmeans clustering optimization of the mask with k=2 separates the residual area and noise;

[0033] e) Circle fitting of the optimized mask to get the final effective residual area A(t).

[0034] Further, in step 3, the sampling interval of conductivity detection is 0.5s-2s, the end point threshold is 0.05-0.20, and the continuous determination frame number is 2-5 frames.

[0035] Further, in step (4), in the two-fluid flushing, the mixed fluid of ultrapure water and nitrogen gas is sprayed along the tangent direction of wafer rotation, the ultrapure water flow is 100ml / min-200ml / min, the nitrogen gas flow is 60L / min-80L / min, the flushing time is 30s-60s, and the wafer rotation speed is 200rpm-500rpm.

[0036] Further, in step (5), the step-up rotation speed is to remove water at a low speed of 200 rpm to 500 rpm first, and then to dry at a high speed of 1500 rpm to 2000 rpm within 1 s to 3 s, the nitrogen blowing flow is 60 L / min to 70 L / min, and the drying time is 45 s to 60 s.

[0037] The present application has the following beneficial effects:

[0038] (1) The two-step method of "soaking first and then single-piece rotary etching" is adopted, the upper and lower lifting movements are used to remove bubbles and pre-wet the surface in the soaking stage, and the centrifugal force and fresh chemical solution scanning and spraying are used in the rotary etching stage, so that the boundary layer is significantly thinned, and the difference between the in-pellet and inter-pellet etching rates is significantly reduced.

[0039] (2) The dynamic shear force generated by rotary spraying is combined with the coaxial acoustic wave to instantaneously strip Ga(OH)3, As2O3 and other by-products, reduce surface passivation, keep the etching rate stable, and shorten the process time.

[0040] (3) Images are collected in real time during the rotary etching process, combined with YOLO-OBB rotating frame detection and conductivity double-mode fusion criteria, to automatically identify the residual gallium arsenide area, avoid over-etching or incomplete etching, and improve batch consistency.

[0041] (4) The three-stage cleaning of "UPW→two-fluid→UPW" and the step-up rotary drying are adopted, one step is used to complete the removal of residual chemical solution, particles and water stains, reduce the spots after drying, and reduce the probability of subsequent detection and rework. BRIEF DESCRIPTION OF DRAWINGS

[0042] Figure 1 The flowchart of the present application.

[0043] Figure 2 It is the continuous collection of etching process images of the gallium arsenide substrate by the industrial camera through the quartz window on the top of the etching cavity. DETAILED DESCRIPTION

[0044] The present application will be further described below in conjunction with the drawings.

[0045] This embodiment takes a 4-inch n-type gallium arsenide substrate (initial thickness 300 μm, used for high-frequency radio frequency device substrate thinning process) as the processing object, and details the efficient etching method of the single-piece gallium arsenide substrate.

[0046] This embodiment is completed on a conventional single-piece wet etching platform, which includes:

[0047] Polytetrafluoroethylene soaking tank with programmable up-down pulling mechanism, vacuum adsorption type rotary etching cavity (with built-in coaxial megasonic generator), industrial camera (Basler acA2500-14gm, resolution 2592x1944, frame rate 30fps), online conductivity sensor (Mettler-Toledo InPro7250, detection range 0-2000 μS / cm), three-stage cleaning module (including ultrapure water supply system, nitrogen source) and stepped rotary drying module, all equipment realizes process automation connection through PLC control system.

[0048] As Figure 1 , the method of the present application comprises the following steps:

[0049] (1) soaking treatment: placing the gallium arsenide substrate wafer in the etching solution and continuously moving during the soaking process to inhibit bubble adhesion;

[0050] (2) single wafer rotary etching: rotating the soaked wafer around its axis while continuously supplying NH4OH and H2O2 mixed solution to the wafer surface, removing the etching byproducts in real time by the centrifugal force generated by rotation and the scouring force of the solution;

[0051] (3) real-time endpoint detection: continuously acquiring wafer surface images during the rotary etching process, identifying the residual gallium arsenide area through image processing algorithm, and immediately terminating the solution supply when the area reaches the preset endpoint condition;

[0052] (4) three-stage cleaning: sequentially rinsing with ultrapure water, two-fluid flushing with ultrapure water and nitrogen mixture, and again rinsing with ultrapure water to remove surface residual solution and byproducts;

[0053] (5) stepped rotary drying: removing the liquid on the wafer surface by gradually increasing the wafer rotation speed and cooperating with nitrogen blowing to complete the drying.

[0054] The following further describes each step of the method of the present application.

[0055] Soaking treatment;

[0056] (1) etching solution preparation: prepare the soaking etching solution according to the volume ratio of "25% concentration NH4OH: 30% concentration H2O2: ultrapure water (resistivity 18.2 MΩ·cm) = 1:5:10", total amount 5L, after injection into the soaking tank, maintain the temperature at 23℃ through the water bath temperature control device.

[0057] (2) wafer loading and motion control:

[0058] Put 12 pieces of GaAs substrate wafer vertically into a quartz basket, slowly immerse into the etching solution of the soaking tank (the liquid level is 10 mm higher than the uppermost wafer); start the up-and-down lifting mechanism (the lifting mechanism in this embodiment is a clean electric sliding table driven by a servo motor), set the lifting frequency to 8 Hz, the lifting amplitude to 20 mm, the lifting speed to 5 mm / s, and continue to soak for 25 min (the conventional process window is 20-30 min, taking into account the bubble peeling and pre-etching effect).

[0059] (3) Key operation purpose: By periodic up-and-down lifting, a dynamic liquid flow is formed on the wafer surface to peel off the tiny bubbles (diameter <1 μm) attached in real time, avoiding local "etching pits" caused by bubble shielding; at the same time, the 23℃ etching solution can achieve mild pre-etching (thinning thickness of about 20 μm), laying a foundation for the uniformity of subsequent rotary etching.

[0060] II. Single wafer rotary etching.

[0061] (1) Wafer transfer: After soaking, the single wafer is taken out of the quartz basket by a mechanical arm (with a vacuum chuck, suction force 0.08 MPa) and transferred to the vacuum chuck of the rotary etching cavity, and the chuck vacuum adsorption (adsorption force 0.06 MPa) is started to ensure that the wafer is completely attached to the chuck (concentricity error <0.1 mm).

[0062] (2) Etching parameter setting:

[0063] Chemical solution configuration: configure the mixed chemical solution for etching according to the volume ratio of NH4OH to H2O2 1:6, and control the temperature to 30℃ by a heating jacket.

[0064] Spraying and rotation: start the shower head to uniformly supply the mixed chemical solution to the wafer surface, and set the flow rate to 500 ml / min; at the same time, start the chuck rotation, and set the rotation speed to 350 rpm.

[0065] Megasonic wave cooperation: start the coaxial megasonic wave generator, because the current chuck rotation speed 350 rpm is in the interval of "300-500 rpm", so set the megasonic wave frequency to 1.1 MHz, the ultrasonic power density to 22 W / L (in the range of 15-25 W / L), and the megasonic wave action area covers the effective area of the wafer (diameter 100 mm).

[0066] (3) Etching process control: continue etching for 4 min, during which the PLC monitors the chemical solution flow rate, chuck rotation speed and megasonic wave power in real time to ensure that the parameter fluctuation amplitude is <±5%.

[0067] Traditional single wafer etching only relies on centrifugal force and etchant flushing to remove byproducts (such as Ga(OH)3, As2O3), but such byproducts are prone to accumulate in the concave part of the wafer surface to form a "passivation layer", resulting in the attenuation of etching rate and the deterioration of etching uniformity in the wafer. The present application adopts the design of "coupling of megasonic wave and rotation", and the core logic is to build a "multi-physical field collaborative force field".

[0068] The centrifugal force generated by the rotation provides the initial peeling power for the byproducts, ensuring the etchant renewal and the preliminary separation of the byproducts on the whole wafer surface; the megasonic wave (0.8-1.2MHz) generates micro-jet effect through the etchant conduction, which can strip the stubborn byproducts attached to the small concave area of the wafer surface, and form a complement with the centrifugal force and the etchant flushing force. The hidden danger of byproduct accumulation is fundamentally eliminated, ensuring the stability of etching rate throughout the process, while improving the uniformity of etching in the wafer, and changing the etching process from "pure time control" to "interface state control".

[0069] III. Real-time endpoint detection.

[0070] The real-time endpoint detection system is started at the same time as the etching starts, and the specific operation is as follows:

[0071] (1) Image acquisition and etching process visualization;

[0072] For example, Figure 2 The industrial camera continuously acquires the etching process images of the gallium arsenide substrate through the quartz window at the top of the etching chamber, forming a dynamic sequence:

[0073] Top left: in the early stage of etching, the center area is left with gallium arsenide (light gray), and the edge area has completed etching (pink), at this time the residual area is large;

[0074] Top right: in the middle stage of etching, the center residual area is reduced (blue tone), and the etching interface uniformly advances towards the center;

[0075] Bottom right: in the late stage of etching, only a small residual area remains in the center (an obvious blue dot can be seen);

[0076] Bottom left: etching is completed, and the substrate surface has no obvious residual area, showing a uniform post-etching state.

[0077] The image sequence directly reflects the dynamic change of the residual area from "large area" to "complete disappearance", providing a clear identification object for the visual detection algorithm of the present application.

[0078] (2) Visual detection;

[0079] Image acquisition: The industrial camera continuously acquires the wafer surface image through the etching cavity top quartz window (transmittance > 95%), and the image is transmitted to the image processing unit (GPU model NVIDIA RTX4090) through HDMI.

[0080] Improved YOLO-OBB rough detection: Resize the original image to 640x640 pixels, input the improved YOLO-OBB framework with "backbone ResNet50, end inserted CARAFE up-sampling (kernel size 3x3, sampling rate 2) and RFCA-Conv rotating anchor box calibration convolution (number of convolution kernels 256)", and output the rotating box coordinates (format xywhr: center x / y coordinates, width / high, rotation angle) of the gallium arsenide residual area with a confidence level ≥0.8, to realize accurate positioning of the residual area in each stage. Figure 2

[0081] ROI cropping and HSV screening: Based on the rotating box coordinates, draw the region of interest (ROI), and after cropping, convert the image from BGR color space to HSV space. By setting the screening range (H channel 40-80, S channel 100-255, V channel 50-255) through the `cv2.inRange` function, a binary mask is generated, and the mask area A1 is calculated.

[0082] Kmeans clustering and circular fitting: Kmeans clustering (using `cv2.KMEANS_PP_CENTERS` initialization) with k=2 is performed on the HSV screened mask to separate "residual area" and "liquid spot noise". Least squares method is used to fit a circle to the clustered residual area mask, and the fitting circle center coordinates (Px, Py) and radius R (R=1.5mm in this embodiment) are output. A circular mask1 matching the wafer size is created, and mask1 is pixel-wise operated with the clustered mask to obtain the final effective residual area A(t).

[0083] In the traditional visual detection of gallium arsenide residue identification, it is easy to be disturbed by liquid reflection and local stains, and the detection accuracy of small area and irregular shaped residual targets is insufficient, which may lead to end point misjudgment. The present application inserts CARAFE up-sampling and RFCA-Conv rotating anchor box calibration convolution in the YOLO-OBB framework, which breaks through the limitation from the technical principle:

[0084] CARAFE up-sampling can accurately restore the detailed features of small area residual areas, avoiding the blurring of traditional bilinear interpolation on small targets;

[0085] ​RFCA-Conv can dynamically calibrate the angle and position of the anchor frame to address the irregular rotational shape of gallium arsenide residues, significantly reducing missed detections and false detections caused by target rotation.

[0086] The improved YOLO-OBB rotating target detection framework, combined with HSV color screening, Kmeans clustering, and circle fitting, makes visual inspection more targeted in identifying gallium arsenide residues. It provides a stable and reliable physical dimension basis for subsequent endpoint determination, avoiding the process risks caused by "noise misjudgment and missed detection of small targets" from the source.

[0087] (3) Conductivity detection;

[0088] Sampling settings: The online conductivity sensor is installed in the drainage pipe of the etching chamber (10cm from the chamber outlet), and the sampling interval is set to 1s (within the range of 0.5-2s) to collect the conductivity value σ of the drainage in real time;

[0089] Rate of change calculation: The rate of change of conductivity dσ / dt is calculated using the built-in algorithm of the image processing unit (formula: dσ / dt=(σ n -σ n₋1 ) / 1s, where σ n Let σ be the conductivity value at the nth second. n₋1 (This is the conductivity value at the (n-1)th second).

[0090] (4) Weighted fusion and endpoint determination;

[0091] Normalization: Normalize A(t) (unit: mm²) according to "(A(t)-0) / (A0-0)" (A0 is... Figure 2 The initial residual area in the upper left stage (A0 = 78.5 mm² in this embodiment) is normalized to dσ / dt (unit: μS / (cm・s)) using the formula "(dσ / dt-dσ_min) / (dσ_max-dσ_min)" (dσ_min is...). Figure 2 The minimum rate of change in the lower left stage, dσ_max, is Figure 2 (Maximum rate of change in the upper left phase).

[0092] Fusion calculation: The fusion score is calculated according to the formula S(t)=α·A(t)+β·(dσ / dt), where α=0.7 and β=0.3;

[0093] Endpoint trigger: Set the endpoint threshold to 0.15 (within the range of 0.05-0.20). When S(t) is lower than 0.15 for 3 consecutive frames (corresponding to 100ms, 2-5 frames), the PLC immediately sends a signal to stop the supply of liquid and megasonic waves, thus terminating the corrosion.

[0094] The traditional endpoint determination method has obvious limitations: single visual detection is easily disturbed by light and liquid reflection; fixed time control is not consistent with the actual corrosion progress, which may lead to over-corrosion or under-corrosion. The present application innovatively adopts a weighted fusion strategy of conductivity detection and visual detection, and the core value lies in "chemical-physical dual-dimension collaborative verification":

[0095] Conductivity detection (monitoring the change rate of effluent conductivity): reflects the strength and progress of the corrosion reaction from the chemical dimension, is not affected by the physical environment, and can accurately capture the dynamic changes of the reaction.

[0096] Visual detection (improved YOLO-OBB recognition residual area): intuitively presents the morphology and area of the residual gallium from the physical dimension, and provides visual basis for endpoint judgment.

[0097] Weighted fusion (S(t)=α·A(t)+β·(dσ / dt)): balance the reliability of the two detection methods (visual detection as the main, conductivity detection as the auxiliary) through α and β coefficients, avoid misjudgment caused by environmental interference of visual detection, and make up for the lack of conductivity detection in residual shape perception. This design upgrades the endpoint determination from "experience-driven" to "data-driven", ensuring the accuracy and batch consistency of the corrosion endpoint, and effectively eliminating the product quality risk caused by endpoint misjudgment.

[0098] Four, three-stage cleaning.

[0099] After the corrosion is stopped, keep the chuck speed at 350 rpm, and enter the three-stage cleaning process with a total time of 165 s.

[0100] (1) First-stage ultrapure water flushing: start the ultrapure water supply pump, spray ultrapure water (resistivity 18.2 MΩ·cm) on the wafer surface, flow rate 300 ml / min, last for 60 s, remove more than 80% of the residual mixed liquid on the surface;

[0101] (2) Second-stage two-fluid flushing: switch to "ultrapure water + nitrogen" mixed spraying mode, set parameters: ultrapure water flow rate 150 ml / min (100-200 ml / min range), nitrogen flow rate 70 L / min (60-80 L / min range), mixed fluid is sprayed along the tangent direction of wafer rotation, lasts for 45 s, uses the impact force of nitrogen to strip the attached byproduct particles (such as Ga(OH)3, As2O3, diameter <0.3 μm);

[0102] (3) Third-stage ultrapure water flushing: switch to ultrapure water spraying again, flow rate 300 ml / min, lasts for 60 s, remove trace amounts of salt and nitrogen-borne impurities remaining after two-fluid flushing.

[0103] Traditional single cleaning can only remove the visible residual liquid on the wafer surface, and the cleaning ability of sub-μm particles and liquid salt is insufficient. These residues will directly affect the performance and yield of subsequent device preparation. The three-stage cleaning design of the present application follows the logic of "stages, targeted decontamination":

[0104] First stage (UPW1 flushing): quickly dilute and flush most of the residual liquid on the wafer surface, reduce the load of subsequent fine cleaning, and lay the foundation for deep cleaning;

[0105] Second stage (two-fluid flushing): ultra-pure water and nitrogen gas are mixed to form a high-speed jet, which uses the impact force of gas-liquid cooperation to strip μm-level or even sub-μm-level particles attached to the wafer surface, solving the traditional cleaning "particle residue" problem;

[0106] Third stage (UPW2 flushing): remove the salt and trace impurities that may remain after two-fluid flushing, and ensure that the wafer surface cleanliness meets the process requirements of device preparation.

[0107] This design realizes full-dimensional cleaning from "macroscopic liquid residue" to "microscopic particles and salt", providing a defect-free substrate surface for subsequent processes.

[0108] Five, ladder type rotary drying.

[0109] After cleaning, start the ladder drying program, the specific operation is as follows:

[0110] (1) Low-speed water removal stage: keep the chuck speed at 350 rpm, start nitrogen blowing (flow rate 65 L / min), last for 10 s, remove most of the free water on the wafer surface;

[0111] (2) High-speed drying stage: control the chuck speed by PLC to uniformly increase from 350 rpm to 1800 rpm within 1.5 s, maintain nitrogen blowing flow rate at 65 L / min, last for 40 s, and use centrifugal force to strip residual trace water stains;

[0112] (3) Drying termination: after 50 s of total drying time, first reduce the chuck speed to 500 rpm, turn off the nitrogen blowing, then stop the chuck rotation and release the vacuum adsorption, and transfer the dried wafer to a clean storage box by a mechanical arm.

[0113] The traditional drying process has two major problems: direct high-speed rotation can easily cause the wafer surface free water to splash and form "water mark" residues; pure low-speed drying is inefficient and difficult to remove small water stains. The gradient wetting and drying design of the present application essentially controls the "force and time in the drying process".

[0114] The first stage (low speed rotation: 200-500 rpm): gently remove most of the free water on the wafer surface, avoid water film splashing due to instantaneous high-speed rotation, and reduce the possibility of water mark generation from the source;

[0115] The second stage (step speed-up: increase to 1500-2000 rpm within 1-3 s): on the basis of most of the water on the surface being removed, the residual water stains are stripped by high-speed centrifugal force, and the gradual speed-up process greatly reduces the stress load of the thin gallium arsenide substrate.

[0116] This "stable first and fast later" gradient strategy takes into account both drying efficiency and avoids water stain residue and wafer deformation, and is especially suitable for thin substrates required by high-frequency devices, ensuring the flatness and cleanliness of the substrate after drying.

[0117] The above is only the preferred embodiment of the present application, it should be noted that for those skilled in the art, without departing from the principles of the present application, a number of improvements can also be made, which should be considered as the protection scope of the present application.

Claims

1. A method for high efficiency etching of monolithic gallium arsenide substrates, characterized by: The method comprises: (1) soaking treatment: placing the GaAs substrate wafer in the etching solution and continuously moving during the soaking process to inhibit bubble adhesion; (2) single wafer rotary etching: rotating the soaked wafer around its axis while continuously supplying the mixed solution of NH4OH and H2O2 to the wafer surface, and removing the etching by-products in real time by the centrifugal force generated by rotation and the scouring force of the solution; (3) real-time endpoint detection: continuously acquiring wafer surface images during the rotary etching process, identifying the residual GaAs area through image processing algorithms, and immediately terminating the solution supply when it is determined that the area reaches the preset endpoint condition; (4) three-stage cleaning: sequentially rinsing with ultrapure water, two-fluid flushing with a mixture of ultrapure water and nitrogen, and again rinsing with ultrapure water to remove residual solution and by-products on the surface; (5) stepwise rotary drying: removing the liquid on the wafer surface by stepwise increasing the wafer rotation speed and cooperating with nitrogen blowing to complete drying.

2. The method of claim 1, wherein the high efficiency etching of monolithic gallium arsenide substrates is characterized by: In step (1), the continuous motion is an up-and-down lifting motion, the lifting frequency is 5-15 Hz, the lifting amplitude is 10-30 mm, and the etching solution temperature is 22-25°C.

3. The method of claim 1, wherein the high efficiency etching of monolithic gallium arsenide substrates is characterized by: In step (2), the volume ratio of the mixed solution of NH4OH and H2O2 is 1:5-1:7, the solution temperature is 22-40°C, the supply flow rate is 400-600 ml / min, and the wafer rotation speed is 200-500 rpm.

4. The method of claim 3, wherein the high efficiency etching of monolithic gallium arsenide substrates is characterized by: In step (2), while rotary etching, a 0.8-1.2 MHz megasonic wave is applied coaxially, the ultrasonic power density is 15-25 W / L, when the wafer rotation speed is 200-300 rpm, the megasonic wave frequency corresponds to 0.8-1.0 MHz; when the rotation speed is 300-500 rpm, the megasonic wave frequency corresponds to 1.0-1.2 MHz, and the power density remains unchanged at 15-25 W / L.

5. The efficient etching method for a monolithic gallium arsenide substrate as described in claim 1, characterized in that: Step (3) comprises: visual detection: using an improved YOLO-OBB rotating target detection framework combined with HSV color selection, Kmeans clustering and circle fitting to output the residual area A(t); the improved framework inserts CARAFE up-sampling and RFCA-Conv rotating anchor frame calibration convolution at the end of the traditional YOLO-OBB backbone to generate a rotating frame mask of the residual GaAs area in real time; conductivity detection: continuously collecting the conductivity change rate dσ / dt of the etching cavity discharge; weighted fusion: normalizing the residual area A(t) and the conductivity change rate dσ / dt and calculating S(t)=α·A(t)+β·(dσ / dt), α=0.6-0.8, β=0.2-0.4, and determining the endpoint when S(t) continuously falls below the preset threshold.

6. The method of claim 5, wherein the high efficiency etching of monolithic gallium arsenide substrates is characterized by: The visual detection comprises: a) after resizing the image, using the YOLO-OBB framework with CARAFE up-sampling and RFCA-Conv rotating anchor frame calibration convolution to output the rotating frame coordinates of the residual GaAs area; b) drawing an ROI based on the rotating frame output by the improved YOLO-OBB rotating target detection framework and cropping the image; c) The cropped image is converted to HSV color space, and a binary mask of the GaAs residue area is generated by InRange filtering, and the area of the mask is calculated; d) The mask is optimized by Kmeans clustering with k = 2 to separate the residue area from noise; e) The optimized mask is fitted with a circle to obtain the final effective residue area A(t).

7. The method of claim 5, wherein the high efficiency etching of monolithic gallium arsenide substrates is characterized by: In step 3, the sampling interval of conductivity detection is 0.5s-2s, the end-point threshold is 0.05-0.20, and the continuous determination frame number is 2-5 frames.

8. The method of claim 1 wherein: the single crystal gallium arsenide substrate is a high efficiency etch. 5 In step (4), in the two-fluid flushing, the mixed fluid of ultrapure water and nitrogen gas sprays along the tangent direction of wafer rotation, the ultrapure water flow is 100ml / min-200ml / min, the nitrogen gas flow is 60L / min-80L / min, the flushing time is 30s-60s, and the wafer rotation speed is 200rpm-500rpm.

9. The method of claim 1 wherein: the single crystal gallium arsenide substrate is a high efficiency etch. 5 In step (5), the stepwise increase in rotation speed is to first remove water at a low speed of 200rpm-500rpm, and then increase to a high speed of 1500rpm-2000rpm within 1s-3s, the nitrogen gas purge flow is 60L / min-70L / min, and the drying time is 45s-60s.

Images