Micro LED Wafer-Level Rapid Defect Location System Based on Photoluminescence

Abstract

This invention relates to the field of optical inspection technology, specifically to a micro LED wafer-level rapid defect localization system based on photoluminescence. A broadband pulse excitation module outputs pulsed laser light matching the characteristics of the material under test; a two-dimensional scanning platform carries the wafer and achieves nanometer-level precise positioning; a time-resolved acquisition module collects fluorescence signals and converts them into time-resolved photoluminescence data; a data processing and control module synchronously controls the operation of each module, receives and processes the time-resolved photoluminescence data, extracts the transient luminescence intensity and carrier lifetime of each detection point, compares the two-dimensional feature vector formed by these two parameters with a preset defect feature database, identifies the defect type of each micro LED chip, and generates a wafer-level defect fingerprint spectrum. This system can accurately distinguish different types of failure modes such as epitaxial defects, surface state defects, and poor electrode contact, significantly improving detection efficiency and defect classification accuracy.

Description

Technical Field

[0001] This invention relates to the field of optical inspection technology, and in particular to a wafer-level rapid defect location system for Micro LEDs based on photoluminescence. Background Technology

[0002] In the field of new display technology, Micro LED is regarded as the next mainstream display technology after LCD and OLED due to its significant advantages such as high brightness, low power consumption, long lifespan, and high integration. However, the miniaturization of Micro LED chips to the micrometer level (typically less than 50μm) presents enormous challenges to its manufacturing process, especially in defect control during epitaxial growth, chip transfer, and bonding.

[0003] Currently, the industry mainly relies on two methods for inspecting Micro LED chips: one is to use a probe station to perform contact electroluminescence (EL) testing on individual chips. This method has high precision, but it is a destructive or contact-based inspection and has extremely low efficiency, which cannot meet the needs of wafer-level mass inspection; the other is to use traditional microscopic imaging or photoluminescence (PL) surface scanning technology. Traditional PL scanning usually uses continuous laser as the excitation source, combined with a spectrometer or camera to perform point-by-point or line-by-line scanning imaging. However, this method has the following inherent defects: First, the point-by-point scanning speed is slow, and it takes a long time to complete the inspection of an entire wafer, making it impossible to achieve rapid online inspection; second, traditional PL inspection can usually only obtain the overall luminous intensity distribution of the wafer surface, making it difficult to effectively distinguish between defects from the epitaxial layer material and defects from the quantum well structure itself or surface processing, that is, it cannot effectively identify the type of defect; third, for electrical performance defects caused by poor electrode contact or bonding failure, traditional PL inspection often fails to find the root cause of such "dark" or "dead" chips because it does not involve carrier injection.

[0004] Therefore, existing technologies lack a wafer-level Micro LED defect detection solution that can balance speed, accuracy, and depth (defect type differentiation). Summary of the Invention

[0005] The purpose of this invention is to provide a rapid defect location system for Micro LED wafers based on photoluminescence, which enables non-contact, high-throughput online full inspection of Micro LED wafers, significantly improving inspection efficiency and defect classification accuracy.

[0006] To achieve the above objectives, the present invention provides a micro LED wafer-level rapid defect location system based on photoluminescence, comprising: A broadband pulse excitation module is used to generate pulsed lasers adapted to the properties of the Micro LED material under test to excite the chip on the wafer to emit fluorescence; A two-dimensional scanning platform is used to carry and move the wafer under test to achieve chip-by-chip positioning; The time-resolved acquisition module is used to collect the fluorescence signal emitted by each chip under stimulated emission and convert it into time-resolved photoluminescence data; The data processing and control module is communicatively connected to the broadband pulse excitation module, the two-dimensional scanning platform, and the time-resolved acquisition module, respectively. The data processing and control module is used to control the two-dimensional scanning platform to move along a preset or dynamically planned path; synchronously control the broadband pulse excitation module to emit laser pulses at each detection point; receive and process the time-resolved photoluminescence data uploaded by the time-resolved acquisition module, extract the transient luminescence intensity and carrier lifetime of each detection point; and compare the two-dimensional features of the transient luminescence intensity and carrier lifetime with a preset defect feature database to determine the defect type of each chip and generate a wafer-level fingerprint spectrum reflecting the defect type and distribution.

[0007] The broadband pulse excitation module includes: The fundamental frequency laser generating unit is used to generate ultrashort pulse seed light of a fixed wavelength; The wavelength tuning and conversion unit includes a nonlinear optical crystal and an optical parametric oscillator, which is used to convert the seed light into excitation light with continuously tunable wavelength according to the quantum well bandgap characteristics of the Micro LED material under test. The pulse width modulation unit is used to adjust the pulse width of the excitation light in the range of 300 femtoseconds to 100 picoseconds according to the carrier recombination dynamics characteristics of the material under test. The spatial beam shaping unit is used to shape the modulated laser beam into a focused spot that matches the size of a single Micro LED chip; The synchronous triggering and control unit is used to receive the triggering command from the data processing and control module and coordinate the working timing of the fundamental frequency laser generation unit, wavelength tuning and conversion unit, and pulse width modulation unit.

[0008] The two-dimensional scanning platform includes: The closed-loop feedback control unit uses a laser interferometer or grating ruler to monitor the platform position in real time and forms a closed loop with the servo controller to ensure positioning accuracy. The wafer carrier and fixing unit uses a porous ceramic vacuum chuck to flatten and fix the wafer to be tested; The platform control and communication interface unit is used to receive motion commands from the data processing and control module and to provide real-time feedback of platform position and status signals. The platform control and communication interface unit feeds back the real-time position signal calculated by the closed-loop feedback control unit to the data processing and control module, so that the data processing and control module can generate a trigger signal that is synchronized with the motion state of the scanning platform.

[0009] The time-resolved acquisition module includes: Aperture microscopic optical collection unit, used to efficiently collect fluorescence signals emitted by stimulated emission from a single Micro LED chip; The spectral filtering and splitting unit, which includes a dichroic mirror and a narrowband filter, is used to separate and purify the fluorescence signal and suppress the excitation light scattering background. The single-photon level photoelectric conversion unit uses a single-photon avalanche diode or a hybrid photomultiplier tube to convert received fluorescent photons into electrical pulse signals. The time-correlated single-photon counting unit is used to receive the synchronization trigger signal sent by the data processing and control module as the timing start point, receive the electrical pulse output by the single-photon level photoelectric conversion unit as the timing end point, measure the arrival time of each photon, and accumulate to generate a time-resolved photoluminescence curve histogram.

[0010] The data processing and control module includes: The platform motion control and path planning unit is used to generate an initial scanning path covering all chips based on the wafer layout, and dynamically generate a re-inspection path only for chips with suspected defects based on real-time detection results, and control the two-dimensional scanning platform to perform corresponding movements. The synchronous timing control unit is used to receive the position signal fed back by the two-dimensional scanning platform, and based on a high-precision clock reference, generate an acquisition start trigger signal that is precisely synchronized with the laser pulse emission time, and send it to the broadband pulse excitation module and the time-resolved acquisition module respectively. The TRPL data acquisition and fitting analysis unit is used to receive time-resolved photoluminescence data uploaded by the time-resolved acquisition module and extract the transient luminescence intensity and carrier lifetime of each detection point through a nonlinear least squares fitting algorithm. The defect classification and pattern generation unit is used to construct a two-dimensional feature vector from the transient luminescence intensity and carrier lifetime of each detection point, compare it with the preset defect feature database, determine the defect type, and generate a wafer-level defect fingerprint pattern.

[0011] The data processing and control module further includes: The defect feature database unit is used to store two-dimensional feature distribution data of transient luminescence intensity and carrier lifetime corresponding to good products and various known defects.

[0012] The TRPL data acquisition and fitting analysis unit is also used to adaptively select a single exponential decay model or a double exponential decay model for fitting based on the decay pattern of the time-resolved photoluminescence curve; when the double exponential model is used for fitting, the fast lifetime component, the slow lifetime component and their weights are extracted, and the equivalent carrier lifetime is calculated by weighted average.

[0013] The defect classification and map generation unit is also used to map the transient luminescence intensity and carrier lifetime of the chip under test to a two-dimensional feature space, calculate the Mahalanobis distance or probability density between the chip and each defect feature region in the defect feature database unit, and determine whether the chip belongs to a good product, epitaxial dislocation defect, surface state defect or electrode contact defect; and use different colors to represent different defect types and color depth to represent the severity of defects, generating a defect classification fingerprint map superimposed on the wafer physical coordinates.

[0014] The data processing and control module is further configured to, after performing the first round of global scanning and initially identifying suspicious defective chips, have the platform motion control and path planning unit generate an optimized path that connects only the coordinates of all suspicious defective chips, and control the two-dimensional scanning platform to perform a second round of refined re-inspection scanning along the optimized path.

[0015] The synchronous timing control unit includes a high-precision digital delay generator, which is used to perform delay calibration on the acquisition start trigger signal sent to the time-resolved acquisition module based on the flight time of the laser pulse from the excitation module to the surface of the wafer under test, the flight time of the fluorescence signal from the sample to the detector, and the inherent delay of the electronic circuit, so as to ensure that the acquisition timing zero point is aligned with the time when the laser pulse arrives at the sample surface.

[0016] This invention discloses a rapid wafer-level defect localization system for Micro LEDs based on photoluminescence, comprising a broadband pulse excitation module, a two-dimensional scanning platform, a time-resolved acquisition module, and a data processing and control module. The broadband pulse excitation module outputs pulsed laser light matching the characteristics of the material under test; the two-dimensional scanning platform supports the wafer and achieves nanometer-level precise positioning; the time-resolved acquisition module collects fluorescence signals and converts them into time-resolved photoluminescence data; the data processing and control module synchronously controls the operation of each module, receives and processes the time-resolved photoluminescence data, extracts the transient luminescence intensity and carrier lifetime at each detection point, compares the two-dimensional feature vector formed by these two parameters with a preset defect feature database, identifies the defect type of each Micro LED chip, and generates a wafer-level defect fingerprint. Compared with existing technologies, this invention, by introducing a carrier lifetime parameter and establishing a dual-parameter collaborative detection mechanism, achieves non-contact, high-throughput online full inspection of MicroLED wafers. It can accurately distinguish different types of failure modes such as epitaxial defects, surface state defects, and poor electrode contact, significantly improving detection efficiency and defect classification accuracy, and providing reliable data support for process improvement and yield enhancement. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below.

[0018] Figure 1 This is a schematic diagram of the structural principle of the Micro LED wafer-level rapid defect location system based on photoluminescence provided by the present invention.

[0019] Figure 2 This is a schematic diagram of the structure of the broadband pulse excitation module provided by the present invention.

[0020] Figure 3 This is a schematic diagram of the structure of the two-dimensional scanning platform provided by the present invention.

[0021] Figure 4 This is a schematic diagram of the time-resolved acquisition module provided by the present invention.

[0022] Figure 5 This is a schematic diagram of the data processing and control module provided by the present invention.

[0023] In the diagram: 101-Broadband pulse excitation module, 102-Two-dimensional scanning platform, 103-Time-resolved acquisition module, 104-Data processing and control module, 1011-Fundamental frequency laser generation unit, 1012-Wavelength tuning and conversion unit, 1013-Pulse width adjustment unit, 1014-Spatial beam shaping unit, 1015-Synchronous triggering and control unit, 1021-Closed-loop feedback control unit, 1022-Wafer support and fixing unit, 1023-Platform control and communication interface unit, 1031-Aperture microscopic optical collection unit, 1032-Spectral filtering and beam splitting unit, 1033-Single-photon level photoelectric conversion unit, 1034-Time-correlated single-photon counting unit, 1041-Platform motion control and path planning unit, 1042-Synchronous timing control unit, 1043-TRPL data acquisition and fitting analysis unit, 1044-Defect classification and map generation unit, 1045-Defect feature database unit. Detailed Implementation

[0024] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application.

[0025] The terminology used in this application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. The singular forms “a,” “the,” and “the” used in this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0026] It should be understood that although the terms first, second, third, etc., may be used in this application to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another. For example, without departing from the scope of this application, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to determination."

[0027] Please see Figures 1-5 This invention provides a micro LED wafer-level rapid defect location system based on photoluminescence, comprising: A broadband pulse excitation module is used to generate pulsed lasers adapted to the properties of the Micro LED material under test to excite the chip on the wafer to emit fluorescence; The two-dimensional scanning platform 102 is used to carry and move the wafer under test to achieve chip-by-chip positioning; The time-resolved acquisition module 103 is used to collect the fluorescence signal emitted by each chip under stimulated emission and convert it into time-resolved photoluminescence data; The data processing and control module 104 is communicatively connected to the broadband pulse excitation module, the two-dimensional scanning platform 102 and the time-resolved acquisition module 103, respectively. The data processing and control module 104 is used to control the two-dimensional scanning platform 102 to move along a preset or dynamically planned path; synchronously control the broadband pulse excitation module to emit laser pulses at each detection point; receive and process the time-resolved photoluminescence data uploaded by the time-resolved acquisition module 103, extract the transient luminescence intensity and carrier lifetime of each detection point; and compare the two-dimensional features of the transient luminescence intensity and carrier lifetime with a preset defect feature database to determine the defect type of each chip and generate a wafer-level fingerprint spectrum reflecting the defect type and distribution.

[0028] In this embodiment, the broadband pulse excitation module is used to generate pulsed laser light of specific wavelengths and pulse widths to excite the chip on the Micro LED wafer to emit light. This module includes a tunable laser capable of outputting excitation light of different wavelengths depending on the quantum well characteristics of the Micro LED material under test. More importantly, the laser operates in pulsed mode, generating short pulses of light on the picosecond or nanosecond scale, providing a trigger signal for subsequent time-resolved measurements.

[0029] The two-dimensional scanning platform 102 is used to carry and move the entire Micro LED wafer under test. This platform has nanometer-level positioning accuracy, ensuring that the excitation spot can accurately cover every micrometer-sized chip unit. The platform's control unit is connected to the data processing unit to realize automatic planning and synchronous triggering of the scanning path.

[0030] The time-resolved acquisition module 103 is used to collect and process the fluorescence signal emitted by stimulated emission from the wafer. The core of this module is a high-sensitivity single-photon detector or streak camera coupled with a high numerical aperture microscopic optical system. This module not only acquires the intensity information of the emitted light, but more importantly, it can record the light emission decay process of each pixel after pulse excitation with picosecond-level time resolution, i.e., obtain the time-resolved photoluminescence (TRPL) curve.

[0031] The data processing and control module 104 is the core of the entire system, and it has a pre-set database of various defect features. This unit first controls the two-dimensional scanning platform 102 to move in a step-by-step manner, while simultaneously controlling the pulse excitation module to emit light pulses. At each scanning coordinate point, the TRPL data acquired by the time-resolved acquisition module 103 is transmitted to the data processing unit in real time. This unit performs fitting analysis on the TRPL curve, extracting two key parameters: transient luminescence intensity and carrier lifetime. Then, by comparing these two key parameters with features in the pre-set database, it automatically determines the defect type of the chip and generates a wafer-level distribution fingerprint map containing the defect location and type.

[0032] Furthermore, the broadband pulse excitation module includes: The fundamental frequency laser generating unit 1011 is used to generate ultrashort pulse seed light of a fixed wavelength; The wavelength tuning and conversion unit 1012 includes a nonlinear optical crystal and an optical parametric oscillator, used to convert the seed light into excitation light with continuously tunable wavelength according to the quantum well bandgap characteristics of the Micro LED material under test. The pulse width modulation unit 1013 is used to adjust the pulse width of the excitation light in the range of 300 femtoseconds to 100 picoseconds according to the carrier recombination dynamics characteristics of the material under test. The spatial beam shaping unit 1014 is used to shape the modulated laser beam into a focused spot that matches the size of a single Micro LED chip; The synchronous triggering and control unit 1015 is used to receive the triggering command from the data processing and control module 104 and coordinate the working timing of the fundamental frequency laser generating unit 1011, wavelength tuning and conversion unit 1012, and pulse width modulation unit 1013.

[0033] In this embodiment, the fundamental frequency laser generating unit 1011 is the source of the entire excitation light, employing a highly stable, high-beam-quality mode-locked laser as the basic light source. The technical solution involves generating an ultrashort pulse seed light of a fixed wavelength. For example, a ytterbium-doped fiber laser can be used to generate an infrared pulse with a center wavelength of 1030 nm and a pulse width in the picosecond range. The key design feature of this unit is providing an extremely stable pulse sequence, establishing a time reference for all subsequent modulation stages. Its repetition frequency can be adjusted between 100 kHz and MHz according to the scanning speed requirements.

[0034] The wavelength tuning and conversion unit 1012 is a key unit for achieving matching with the properties of the material under test. It integrates nonlinear optical crystals (such as BBO and LBO crystals) and optical parametric oscillator (OPO) technology. A 1030nm infrared pulse light from the fundamental frequency laser generation unit 1011 is incident into the OPO system. By precisely controlling the temperature and phase-matching angle of the nonlinear crystal within the OPO cavity, the incident light of a fixed wavelength can be converted into continuously tunable signal light and idler light. The MicroLED material under test (such as InGaN and AlGaInP-based quantum wells) has different bandgap widths, and its optimal photoluminescence excitation wavelength is usually located near its absorption edge. For example, for AlGaInP-based MicroLEDs emitting red light, the optimal excitation wavelength is typically in the 500nm-550nm green light band; while for InGaN-based MicroLEDs emitting blue light, the optimal excitation wavelength is in the near-ultraviolet band (such as 380nm-420nm). This unit precisely controls the angle and temperature of the OPO crystal using a servo motor, thus accurately tuning the output wavelength to match the bandgap energy of the material under test. Specifically, before testing, the system reads the process batch information of the wafer under test, automatically calls the preset wavelength parameters, and controls the OPO crystal to lock the output wavelength at the quantum well absorption peak of the material (typically 20nm-50nm shorter than its emission peak wavelength to achieve efficient excitation), thereby ensuring maximum excitation efficiency and signal-to-noise ratio.

[0035] After wavelength matching is achieved, the temporal characteristics of the excitation light also need to match the carrier dynamics of the material. The pulse width modulation unit 1013, based on dispersion management technology, consists of a pair of gratings or prisms. Ultrashort pulses exhibit dispersion as they pass through optical elements, causing pulse width broadening or compression. This unit precisely controls the pulse width by introducing controllable negative or positive dispersion through precise adjustment of the spacing and angle between the grating pairs. The radiative recombination lifetime of carriers in Micro LED quantum wells is typically between several hundred picoseconds and several nanoseconds. To accurately measure carrier lifetime, the width of the excitation pulse must be much smaller than the lifetime to be measured. This unit can continuously adjust the pulse width within the range of 300 femtoseconds (fs) to 100 picoseconds (ps). For high-quality bulk materials or quantum wells, the carrier lifetime is long (nanosecond level), allowing for the selection of wider pulse widths (e.g., 10 ps-50 ps) to reduce nonlinear effects and obtain stable excitation. However, for materials with many defects or microstructures rich in surface states, the nonradiative recombination rate is fast, resulting in an extremely short effective emission lifetime (potentially below 100 ps). In such cases, the pulse width needs to be compressed to the sub-picosecond or femtosecond level (e.g., below 500 fs) to excite the signal with extremely high instantaneous power, thereby enabling the differentiation of extremely short-lived components and capturing the rapid carrier trapping process caused by defects. This adaptive pulse width adjustment ensures that the dynamic range of time-resolved measurements can cover materials of varying qualities.

[0036] The spatial beam shaping unit 1014 is responsible for shaping the tuned pulsed laser into a spot suitable for exciting a single Micro LED chip or a microarray. It consists of a series of beam expanders, spatial filters, and a variable-focus lens group. First, the beam is expanded and collimated. Then, a pinhole filter with adjustable aperture removes stray light from the beam edges, obtaining an ideal near-Gaussian distribution spot. Finally, the beam is focused through a high numerical aperture microscope objective. After passing through this unit, the diameter of the spot focused on the wafer surface can be adjusted to between 1 μm and 10 μm. This size range covers Micro LED chip sizes from a few micrometers to tens of micrometers, ensuring that the excitation light can precisely act on a single chip area while avoiding crosstalk to adjacent chips.

[0037] The synchronous triggering and control unit 1015 incorporates a high-precision digital delay generator, receiving a global clock signal from the main control computer. It provides a drive signal to the modulator within the laser and generates two precisely delayed outputs: one to trigger the time-resolved acquisition module 103 to begin acquisition, and the other to synchronize the stepping motion of the two-dimensional scanning platform 102. By finely adjusting the delay time of these two trigger signals, it is ensured that laser pulse excitation, signal acquisition, and platform movement are synchronized with picosecond-level time accuracy.

[0038] Furthermore, the two-dimensional scanning platform 102 includes: The closed-loop feedback control unit 1021 uses a laser interferometer or grating ruler to monitor the platform position in real time and forms a closed loop with the servo controller to ensure positioning accuracy. The wafer carrier and fixing unit 1022 uses a porous ceramic vacuum chuck to flatten and fix the wafer to be tested; The platform control and communication interface unit 1023 is used to receive motion commands from the data processing and control module 104 and provide real-time feedback of platform position and status signals. The platform control and communication interface unit 1023 feeds back the real-time position signal calculated by the closed-loop feedback control unit 1021 to the data processing and control module 104, so that the data processing and control module 104 can generate a trigger signal synchronized with the motion state of the scanning platform.

[0039] In this embodiment, the physical actuator of the two-dimensional scanning platform 102 is a precision mechanical motion unit responsible for carrying the wafer to move precisely in the X and Y axes. It employs a design combining an air-bearing guide rail and a linear motor drive. The air-bearing guide rail eliminates friction and wear caused by mechanical contact, ensuring smooth movement and particle-free operation (crucial for wafer cleanliness). The linear motor directly drives the wafer, avoiding the backlash and elastic deformation inherent in traditional rotary motors with ball screws. This unit achieves a motion resolution of 1 nanometer, a repeatability better than ±50 nanometers, and a maximum movement speed of 200 mm / s. This combination of high speed and high precision allows it to maintain accurate final positioning while rapidly moving to the next detection point.

[0040] The closed-loop feedback control unit 1021 uses a laser interferometer or a high-resolution optical grating ruler as the position feedback element. Taking a laser interferometer as an example, it calculates the precise position of the platform in real time by measuring the optical path difference between the reflector fixed on the motion stage and the interferometer head. This position data is fed back to the servo controller at a nanosecond update rate. The servo controller compares the received actual position with the commanded target position, calculates the position deviation, and immediately generates a compensation current to drive the linear motor, forming a complete closed-loop control system that ensures the platform is always at the commanded position and resists external disturbances.

[0041] The wafer carrier and fixation unit 1022 employs a porous ceramic vacuum chuck. The chuck surface undergoes ultra-precision machining, achieving micron-level flatness to ensure the wafer is completely flat and conforms to the focal plane after being adsorbed. The porous design guarantees a uniform distribution of adsorption force, preventing damage to the wafer or Micro LED chip array caused by localized stress. Simultaneously, the chuck material possesses excellent thermal stability, reducing the impact of thermal expansion and contraction caused by ambient temperature changes on positioning accuracy.

[0042] The platform control and communication interface unit 1023 is based on a high-performance FPGA or DSP motion controller and incorporates a complex motion trajectory planning algorithm. This unit is connected to the system's synchronization trigger and control unit 1015 and data processing and control module 104 via a high-speed fiber optic cable or dedicated digital interface. It not only receives movement commands but also outputs real-time position pulse signals for synchronizing other modules. The synchronization trigger and control unit 1015 uses the real-time position information fed back by the platform, combined with a preset scanning path (target position sequence), to make predictions. When the platform is about to reach the next target detection point (e.g., 100 nanometers from the target point), the digital delay generator inside the synchronization trigger and control unit 1015 begins preparation. Once the position signal fed back by the platform confirms that it has entered the set "trigger window" (i.e., reaching the target point and stabilizing, with fluctuations less than a set threshold, such as ±10 nanometers), the synchronization trigger and control unit 1015 immediately issues a hardware trigger level. This trigger signal is sent in parallel to two destinations: to the broadband pulse excitation module 101, commanding it to immediately emit a laser pulse; and to the time-resolved acquisition module 103, commanding it to begin data acquisition.

[0043] The scanning path of the 2D scanning platform 102 is not a simple raster scan, but is dynamically and intelligently planned by the data processing and control module 104 based on real-time detection results. This reflects the system's adaptability and high efficiency. At the start of the inspection, the data processing and control module 104 generates a sparse global coordinate grid based on the layout file (GDSII file) of the wafer under test and sends it to the platform for the first round of "coarse scanning". The purpose of this stage is to quickly establish the position index of all chips on the wafer and correct for the small rotations and offsets caused by wafer loading.

[0044] The first round of rapid screening: Following the initial path planned by the data processing and control module 104, the platform performs rapid single-point excitation and data acquisition on all chips. This stage may only involve a short integration time to obtain transient luminescence intensity I0 and preliminary carrier lifetime τ data.

[0045] Real-time data analysis and path replanning: While receiving the first round of data, the data processing and control module 104 performs real-time (I0, τ) dual-parameter analysis. Once it is found that the (I0, τ) combination of a certain chip falls into the preset "defect suspicion area" or exceeds the good product threshold, the unit immediately marks the coordinates of this chip as a "point that needs to be re-inspected".

[0046] Before the system is put into formal testing, a large number of known qualified wafers from the same production line (i.e., wafers confirmed to be defect-free through traditional electroluminescence testing or manual sampling) are first subjected to standard testing. The system collects transient luminescence intensity I0 and carrier lifetime τ data of all chips on these wafers, obtaining a set of two-dimensional data points. A central region is determined using statistical methods (e.g., calculating the mean and standard deviation). Typically, the yield threshold is defined as an elliptical or rectangular region centered on the center of the data point set and extended along both I0 and τ directions by a certain multiple of the standard deviation (e.g., 3 times the standard deviation). Chips falling within this region are considered "good," and their boundaries are the "yield threshold." The defect-suspected region refers to the area located outside the yield threshold or close to the characteristic regions of various known defects. Specifically, the system simultaneously stores in the database the (I0, τ) characteristic distributions of various typical defects (such as epitaxial dislocations, surface state defects, and poor electrode contact) obtained through comparative experiments. These defect data points also form their own cluster regions. The system defines the outer boundaries of these regions (e.g., a confidence ellipse containing 95% of the samples) as "defect feature areas." Transitional zones outside the yield threshold and close to any defect feature area, or areas not yet clearly classified but deviating from the yield area by a certain distance (e.g., more than 3 standard deviations), can be designated as "suspected defect areas." This setting is pre-configured through software parameters. During the inspection process, the Euclidean or Mahalanobis distance between the (I0, τ) point of each chip and the center of the yield area is calculated in real time. When the distance exceeds a preset threshold (e.g., 3σ), it is judged as suspicious.

[0047] Intelligent re-inspection path generation: After the first round of scanning is completed, the data processing and control module 104 does not simply repeat the scan, but generates an optimized path that connects only all "points to be re-inspected" (for example, using the traveling salesman problem algorithm to solve for the shortest path). This new path is sent to the two-dimensional scanning platform 102.

[0048] Refined re-inspection execution: The platform executes this optimized re-inspection path, performing longer integration time and higher signal-to-noise ratio TRPL curve acquisition on suspicious chips to accurately classify defect types.

[0049] Furthermore, the time-resolved acquisition module 103 includes: Aperture microscopic optical collection unit 1031 is used to efficiently collect fluorescence signals emitted by stimulated emission from a single Micro LED chip; The spectral filtering and splitting unit 1032 includes a dichroic mirror and a narrowband filter, which are used to separate and purify the fluorescence signal and suppress the excitation light scattering background. The single-photon-level photoelectric conversion unit 1033 uses a single-photon avalanche diode or a hybrid photomultiplier tube to convert received fluorescent photons into electrical pulse signals. The time-correlated single-photon counting unit 1034 is used to receive the synchronous trigger signal sent by the data processing and control module 104 as the timing start point, receive the electrical pulse output by the single-photon level photoelectric conversion unit 1033 as the timing end point, measure the arrival time of each photon, and accumulate to generate a time-resolved photoluminescence curve histogram.

[0050] In this embodiment, the aperture microscopy optical collection unit 1031 employs a long working distance, high numerical aperture (NA) apochromatic microscope objective as the front-end collection element. Its numerical aperture is designed between 0.7 and 0.9 to achieve the largest possible solid angle collection efficiency. Multi-layered coated optical lenses are embedded within the objective to effectively correct chromatic aberration and spherical aberration across a broad spectral range from visible to near-infrared, ensuring precise focusing of fluorescence signals at different wavelengths. This unit is mounted on a precision Z-axis focusing mechanism driven by piezoelectric ceramics, enabling automatic focusing based on minute undulations on the wafer surface to ensure the signal is always in optimal collection condition. Fluorescence emitted from the chip on the wafer surface is first collected and collimated into parallel light by the microscope objective, and then enters the subsequent spectral filtering unit.

[0051] The spectral filtering and dispersive unit 1032 employs a combination of a dichroic mirror and an ultra-steep bandpass filter. First, a specially designed dichroic mirror is placed at a 45-degree angle in the optical path, characterized by reflecting the excitation wavelength (e.g., 405 nm) while transmitting the fluorescence wavelength (e.g., 450 nm-700 nm). After the excitation pulse excites the sample, the residual reflected light is reflected back to the light source by the dichroic mirror or captured by the absorption trap. The forward fluorescence signal is efficiently transmitted. The transmitted fluorescence then passes through one or more ultra-steep bandpass filters with an out-of-band cutoff depth (optical density OD value) greater than 6, ensuring a suppression ratio of more than one part per million for the excitation light. For multicolor detection requirements, this unit can be designed as a motorized filter wheel, incorporating multiple sets of narrowband filters for different MicroLED emission wavelengths (red, green, blue). The system can automatically switch to the corresponding channel based on the type of wafer being detected, allowing only the fluorescence of the target wavelength to pass through, further suppressing crosstalk and background noise.

[0052] The single-photon level photoelectric conversion unit 1033 employs a single-photon avalanche diode (SPAD) or a hybrid photomultiplier tube (Hybrid PMT) as its detector, which possesses extremely high sensitivity and ultrafast time response. These detectors can generate a standard current pulse with an extremely steep rise time (tens of picoseconds) upon receiving a single photon. Key performance indicators include a photon detection efficiency exceeding 50% (especially in the visible light band) and a time jitter of less than 50 picoseconds (i.e., measurement uncertainty regarding photon arrival time), which is fundamental to ensuring the time resolution of the TRPL curve.

[0053] The Time-Correlated Single-Photon Counting (TCSPC) unit is based on a time-to-amplitude converter (TAC) and an analog-to-digital converter (ADC), or a TCSPC module based on digital signal processing (DSP). Its working principle is as follows: The broadband pulse excitation module 101 emits ultrashort laser pulses at a high repetition rate (e.g., 1 MHz), each pulse instantaneously exciting a large number of electron-hole pairs in the quantum well of the Micro LED. The excited carriers emit light through radiative recombination. The photon emission time is not fixed but follows an exponentially decaying probability distribution. For high-quality materials, the recombination process is slow, and photons tend to be emitted later; for materials with many defects, non-radiative recombination dominates, carriers are rapidly trapped, and the photon emission probability decreases rapidly over time. The TCSPC unit records at most the first arriving photon in each laser pulse cycle (or processes multi-photon events using an algorithm). This is to avoid temporal overlap caused by multiple photons in a single cycle. Through the strategy of "multiple cycles, only recording the first photon," the system transforms the physical change in light intensity decay over time into a probability density distribution of photon arrival times. The arrival times Δt of all photons recorded in tens of thousands (e.g., 10 million cycles) are statistically analyzed in chronological order. The horizontal axis represents time (usually divided into channels in picoseconds), and the vertical axis represents the number of photons recorded in each time channel.

[0054] The final histogram has its ordinate proportional to the fluorescence intensity at that moment. Smoothing and fitting this histogram (e.g., using an exponential decay model I(t) = I0·exp(-t / τ)) yields a standard TRPL curve. From the fitting results, two key parameters can be precisely extracted: Transient luminescence intensity I0: The photon count at the start of the decay curve (t=0) reflects the luminescence capability immediately after stimulated emission and is related to the excitation power and radiative recombination probability.

[0055] Carrier lifetime τ: The time required for the transient emission intensity to decay to 1 / e of I0, reflecting the average existence time of the carriers before recombination, and is extremely sensitive to various defects and impurities. Here, "I0" represents the transient emission intensity, and "e" is the base of the natural logarithm (approximately 2.71828). This expression has a standard definition in time-resolved spectroscopy: In a time-resolved photoluminescence decay curve, the intensity decreases exponentially with time: I(t) = I0·exp(-t / τ). When time t equals the carrier lifetime τ, I(τ) = I0·exp(-1) = I0 / e. That is, "1 / e of I0" refers to the value obtained by dividing the transient emission intensity by the natural constant e. On the curve, the time interval from the initial intensity I0 to I0 / e is the carrier lifetime τ.

[0056] The data and control interaction between the time-resolved acquisition module 103 and the broadband pulse excitation module 101 follows a precise "excitation-acquisition" handshake protocol and is uniformly scheduled by the synchronous triggering and control unit 1015. The broadband pulse excitation module 101 acts as the "master source" for timing, and its internal synchronization triggering and control unit 1015 generates a high-precision clock reference. When the two-dimensional scanning platform 102 is ready, the synchronization triggering and control unit 1015 sends a trigger pulse, which is divided into two paths: one path activates the excitation module to emit a laser, and the other path (after precise electronic delay to compensate for optical and circuit transmission time) is sent as a "start" signal to the time-correlated single-photon counting unit 1034, resetting its internal timer and starting to wait for photons. The laser pulse (approximately picosecond wide) illuminates the Micro LED chip, exciting fluorescence. The fluorescent photons travel through the physical path of the microscopic optical collection unit and the spectral filtering and splitting unit 1032, finally reaching the detector. This optical path transmission time (typically nanoseconds) is fixed and can be subtracted through system calibration. The electrical pulse generated after the detector detects the photon is immediately sent to the TCSPC unit as a "stop" signal. The time difference Δt calculated by the TCSPC unit actually includes the fixed delay of optical path transmission and the actual fluorescence decay process. Accurate fluorescence lifetime information can be obtained by subtracting the fixed delay obtained from calibration. The TRPL curve data generated cumulatively at each measurement point, along with the platform coordinate information of that point, is packaged and transmitted to the data processing and control module 104 for subsequent (I0, τ) dual-parameter extraction and defect classification.

[0057] Furthermore, the data processing and control module 104 includes: The platform motion control and path planning unit 1041 is used to generate an initial scanning path covering all chips according to the wafer layout, and dynamically generate a re-inspection path only for chips with suspected defects according to the real-time detection results, and control the two-dimensional scanning platform 102 to perform corresponding movements. The synchronization timing control unit 1042 is used to receive the position signal fed back by the two-dimensional scanning platform 102, and generate a collection start trigger signal that is precisely synchronized with the laser pulse emission time based on a high-precision clock reference, and send it to the broadband pulse excitation module 101 and the time-resolved acquisition module 103 respectively. The TRPL data acquisition and fitting analysis unit 1043 receives time-resolved photoluminescence data uploaded by the time-resolved acquisition module 103 and extracts the transient luminescence intensity and carrier lifetime at each detection point using a nonlinear least squares fitting algorithm. The process of obtaining time-resolved photoluminescence data is as follows: First, the broadband pulse excitation module 101 emits an ultrashort laser pulse (pulse width between 300 femtoseconds and 100 picoseconds) towards the Micro LED chip under test. The chip emits fluorescence due to stimulated emission, which is collected by the high numerical aperture microscope objective in the time-resolved acquisition module 103. After filtering and dispersion, each fluorescent photon is converted into an electrical pulse by a single-photon avalanche diode. The time-correlated single-photon counting unit 1034 (TCSPC) records the arrival time of each photon, using the laser pulse emission time as the timing start point. After thousands to millions of laser pulse cycles of repeated excitation and photon counting, a histogram of photon arrival time distribution is calculated. The histogram's horizontal axis represents time (picoseconds), and the vertical axis represents photon count. Its envelope is the time-resolved photoluminescence (TRPL) curve. This curve directly reflects the decay of fluorescence intensity over time.

[0058] The defect classification and map generation unit 1044 is used to construct a two-dimensional feature vector from the transient luminescence intensity and carrier lifetime of each detection point, compare it with the preset defect feature database, determine the defect type, and generate a wafer-level defect fingerprint map. The defect feature database unit 1045 is used to store two-dimensional feature distribution data of transient luminescence intensity and carrier lifetime corresponding to good products and various known defects.

[0059] In this embodiment, the platform motion control and path planning unit 1041 establishes a dedicated digital link with the platform control and communication interface unit 1023 via a high-speed field-programmable gate array (FPGA). This unit issues target position commands (X, Y coordinates) in real time and receives real-time position and status signals from the platform. A built-in PID (proportional-integral-derivative) control algorithm is used to adjust the servo according to the feedback deviation, ensuring the platform stabilizes at the target point in the shortest time without overshoot. Before the inspection begins, this unit reads the layout file of the wafer under test (e.g., GDSII format), parses the center coordinates of all valid chips, and generates a continuous scan path covering all chips. Path planning employs a "shortest path first" algorithm (e.g., a heuristic algorithm based on the traveling salesman problem), combined with the platform's acceleration and deceleration performance, to plan the path with the shortest total movement distance and the least running time, while avoiding sharp turns and overshoot. During the inspection process, this unit continuously receives real-time feedback from the defect classification and map generation unit 1044. Once a chip is initially identified as a "suspected defect," the unit immediately inserts the chip's coordinates into the "re-inspection queue." Once the first round of global scanning is complete, an optimized path connecting only all re-inspection points is automatically generated and executed. This process achieves an intelligent closed loop of "scanning and diagnosing simultaneously, with immediate re-inspection."

[0060] The synchronization timing control unit 1042 incorporates a high-precision digital delay generator, with a clock reference derived from a temperature-controlled crystal oscillator or atomic clock, resulting in a time jitter of less than 1 picosecond. This unit receives a "platform in position" signal from the platform motion control and path planning unit 1041 (triggered by a high-precision position provided by the platform closed-loop feedback unit). When the platform enters the set trigger window (e.g., within ±10 nanometers of the target point and with zero velocity), this unit immediately generates two synchronous trigger pulses: The first trigger pulse (photo-excitation trigger) is sent to the synchronous trigger and control unit 1015 of the broadband pulse excitation module 101, commanding it to emit a laser pulse.

[0061] The second trigger pulse (acquisition start trigger) is sent to the time-correlated single-photon counting unit 1034 of the time-resolved acquisition module 103 after a precisely programmable delay, serving as the "start" signal for time measurement. This delay value is obtained through system calibration to compensate for the flight time of the laser from the excitation module to the sample surface, the flight time of the fluorescence from the sample to the detector, and the inherent delay of the electronic circuitry, ensuring that time T0 is precisely aligned with the instant the laser pulse arrives at the sample surface.

[0062] The TRPL data acquisition and fitting analysis unit 1043 is responsible for receiving and processing the raw data from the acquisition module and extracting key physical parameters. After the 2D scanning platform 102 completes its dwell time at a detection point and triggers acquisition, the time-correlated single-photon counting unit 1034 packages the accumulated histogram data and transmits it to this unit in real time via a high-speed data bus (e.g., PCIe Gen3). Each data packet contains the three-dimensional information of that point: X-coordinate, Y-coordinate, and a one-dimensional array that records the number of photons detected within each time window (typically 512 to 4096 channels) divided in picosecond intervals, starting from the laser pulse arrival time (T0). Before formal processing, the unit first performs a data validity check. For example, it checks whether the total photon count is too low (possibly due to chip missingness or detection failure) or whether the count of a certain time channel is abnormally saturated. For data points that fail to meet quality standards, the system marks them as "acquisition anomalies" and may trigger a re-check later. All detectors have a dark count (background noise when there is no light signal). The system utilizes "dark background" data collected before or after each measurement cycle to perform channel-by-channel subtraction on the current signal histogram, eliminating the influence of dark noise. Simultaneously, it also subtracts fixed background signals caused by ambient light leakage. Although photon counting statistics themselves follow a Poisson distribution, the system employs digital filtering techniques to smooth the curve for more stable subsequent fitting. Specifically, it uses the Savitzky-Golay convolutional smoothing algorithm. This method performs local polynomial fitting on the data within a sliding window, effectively filtering out high-frequency statistical noise while better preserving the peak shape and decay trend of the original curve, avoiding a decrease in temporal resolution or distortion of lifetime information due to smoothing.

[0063] The preprocessed curve is a histogram reflecting the decay of fluorescence intensity over time. The system uses a nonlinear least squares method (such as the Levenberg-Marquardt algorithm) to mathematically fit the preprocessed TRPL curve. The fitting model typically employs a multi-exponential decay function to accommodate different material systems. The single-exponential model, I(t) = I0 * exp(-t / τ) + C, is suitable for high-quality materials with a simple carrier recombination process. The transient luminescence intensity I0 (the intensity at t=0, obtained by extrapolation from the fitted curve) and carrier lifetime τ (decay constant) are extracted from this model. For Micro LED chips with extremely high epitaxial growth quality and extremely low interface state density, the carrier recombination process is relatively simple, mainly manifested as direct radiative recombination from the bottom of the conduction band to the top of the valence band. In this case, the fluorescence decay curve should present a perfect straight line after taking logarithmic coordinates. The system will preferentially attempt to use the single-exponential model for fitting. The core idea of ​​this model is: assuming there is a group of N0 carriers at the initial time t0, they recombine and emit light with a fixed probability (i.e., the reciprocal of the lifetime) over time, then the number of remaining carriers at time t (i.e., luminescence intensity) will decrease successively according to a fixed proportion.

[0064] The double-exponential model, I(t) = A1*exp(-t / τ1) + A2*exp(-t / τ2) + C, is suitable for samples exhibiting both fast and slow recombination processes (e.g., bulk recombination and surface recombination). In this model, the system extracts parameters including two lifetime components (τ1, τ2) and their corresponding weights (A1, A2), and calculates the equivalent lifetime τ_avg = (A1τ1 + A2τ2) / (A1 + A2) using a weighted average, while I0 = A1 + A2. For samples with multiple recombination channels, such as chips with severe surface states or deep-level defects, the decay curve becomes complex, exhibiting a downward-curving curve when viewed on logarithmic coordinates. This indicates the existence of two different recombination processes: a fast process typically corresponds to nonradiative recombination where charge carriers are trapped by surface defects or dislocations, while a slow process corresponds to radiative recombination within the bulk. In this case, the system automatically switches to the double-exponential model. The idea behind this model is to view the decay curve as a linear superposition of two independent first-order decay processes, one decaying quickly (short lifetime) and the other decaying slowly (long lifetime), with the weights of the two processes representing the proportions of carriers participating in each process.

[0065] After selecting a model, the system initiates a fitting algorithm. The algorithm first provides preliminary parameter guesses based on the curve's characteristics. For example, it roughly estimates the lifetime based on the time it takes for the curve to drop to half its initial intensity, and estimates the initial intensity based on the counts of the initial few time channels. Using these guesses, the algorithm calculates the theoretical photon counts at each time point according to the selected model and compares them with the actual measured counts, calculating the difference (residual). Then, using optimization algorithms such as Levenberg-Marquardt, the model parameters (such as lifetime and initial intensity) are automatically adjusted to minimize the sum of squared residuals between the theoretical and actual curves. This process iterates repeatedly until the parameter changes are negligible, or the theoretical and actual curves almost coincide. After each fitting, the system calculates the goodness of fit. One key indicator is whether the residual distribution is random. If the residual distribution exhibits a clear regularity (e.g., wavy), it indicates that the selected model may be incorrect or that there are unconsidered attenuation components. The system will issue a warning and may suggest trying a more complex model.

[0066] After fitting, the system extracts two core physical quantities from the final model parameters: Transient luminescence intensity: This parameter is not simply the count value of the first channel of the histogram, but rather the luminescence intensity at the instant the excitation just ended (time zero) is calculated by fitting a model. It eliminates measurement errors caused by instrument response time, time window division, etc., and more accurately reflects the intrinsic luminescence capability of the chip at the instant of excitation.

[0067] Carrier lifetime: For the single-exponential model, the decay constant is directly extracted as the lifetime. For the double-exponential model, the system outputs two lifetime components and their respective weights. To facilitate subsequent comparison and classification, the system also calculates an "equivalent lifetime," which is a weighted average of the two lifetime components according to their weights. This equivalent lifetime comprehensively reflects the overall average lifetime of carriers in the chip and is a comprehensive indicator of material quality.

[0068] For each measurement point, the fitting analysis unit outputs a set of feature vectors, which at least include the coordinates (X, Y), transient luminescence intensity I0, carrier lifetime τ (or equivalent lifetime), and goodness of fit (e.g., R0). 2 (For quality assessment.)

[0069] The defect classification and map generation unit 1044 maps the (I0, τ) dual parameters of each measurement point to a two-dimensional feature space. The system calls the feature regions (usually two-dimensional Gaussian distribution clusters or polygonal regions) of various defect types pre-stored in the defect feature database unit 1045. A distance-based discrimination method (such as Mahalanobis distance) or a machine learning classifier (such as support vector machine) is used to calculate the probability that the point belongs to each defect type. Specifically, the distance from the point to the center of the "good product area" and the distance to the center of each defect feature region are calculated. The system uses Mahalanobis distance or a probability-based discrimination method, comprehensively considering the distribution density and covariance of the point relative to each feature region, to calculate the probability that it belongs to each category. The point is ultimately determined to be the defect of the type with the highest probability. The classification result is not a simple label; the system also outputs a classification confidence level (e.g., 85% certainty that it is an "extensional dislocation"). For points with low confidence or falling in unknown areas, the system will mark them as "suspected novel defect" or "pending," prompting the user to conduct manual review and in-depth analysis.

[0070] Based on the assessment results, each chip is assigned a defect type label. Possible labels include: Good quality: (I0, τ) is located at the center of the good quality region and has no significant defects. The system does not rely on subjective judgment but is based on statistical calibration of a large number of qualified wafers. For example, at least 100 wafers that have passed multiple process verifications are randomly selected from the production line, and the I0 and τ values ​​of all chips on each wafer are measured. The I0 and τ values ​​of all qualified chips are plotted as distribution histograms, which typically show an approximately normal distribution. The mean of I0 ± 3 standard deviations is taken as the good quality range for I0, and the mean of τ ± 3 standard deviations is taken as the good quality range for τ. The rectangular area formed by these two values ​​on a two-dimensional plane is the good quality region. Alternatively, the Mahalanobis distance can be calculated using the covariance matrix, and the elliptical region with a Mahalanobis distance less than a certain critical value (e.g., 6) is selected. These methods are all common statistical knowledge.

[0071] Epitaxial dislocations / bulk defects: Characterized by a significantly shortened τ (due to more non-radiative recombination centers), while I0 may decrease slightly or remain normal. "Significantly shortened" refers to a carrier lifetime τ being less than the lower limit of the good product region by a certain percentage. For example, when τ is less than the good product mean minus three standard deviations, and further less than 50% of the good product mean, it can be defined as "significantly shortened." This typically corresponds to severe bulk defects such as epitaxial dislocations.

[0072] Surface / interface state defects: manifested as a shortening of τ (surface recombination), and I0 may decrease (enhanced non-radiative recombination). "Shortening" means that τ is less than the lower limit of the good product region, but still within 50% to 80% of the good product average. This corresponds to a moderate reduction in lifetime due to surface state defects.

[0073] Electrode contact / bonding failure: This is mainly manifested as extremely low I0 (electric injection failure leads to no light emission or weak light emission), but τ may be normal (because photoexcitation can still generate charge carriers, but they cannot be electrically injected). "Extremely low" means that the transient light emission intensity I0 is much lower than the lower limit of the good product area, for example, less than 20% of the average good product value. This usually occurs in cases of poor electrode contact or complete chip failure.

[0074] Quantum well quality degradation: manifested as a decrease in I0, with τ possibly changing slightly. "Slight decrease" means that I0 is slightly lower than the average of good quality, but still above the lower limit of the good quality region, or exceeds the lower limit by no more than 10%. This corresponds to slight degradation of quantum well quality.

[0075] The classification results of all chips on the entire wafer are plotted into a two-dimensional color map according to actual physical coordinates, tightly linking the abstract classification results with specific physical locations. Each chip's classification label (such as "Good," "Epiaxial Dislocation," "Electrode Failure," etc.) is placed on a virtual canvas representing the entire wafer according to its actual X and Y coordinates. To allow observers to clearly grasp the defect distribution patterns on the wafer, the system employs rich visualization coding methods, assigning different colors to different types of defects. For example, good chips are displayed in green, epitaxial defects in red, surface defects in blue, and electrode problems in yellow. Users can immediately determine the type of major problem by glancing at the overall color tone of the map. Within the same type of defect, the intensity or grayscale of the color indicates the severity of the defect. For example, for a chip classified as "Epiaxial Dislocation," the darker the color (such as deep red), the greater the deviation of its carrier lifetime from the standard value, and the more severe the defect. The system does not generate just a single final classification map, but rather a series of maps. In addition to the defect classification map, it also includes separate "transient luminescence intensity distribution maps" and "carrier lifetime distribution maps." Users can perform more in-depth analysis by comparing these three maps. For example, if a chip shows normal performance on the intensity map but abnormal performance on the lifetime map, the root cause of the problem can be quickly identified as a defect affecting lifetime. The generated fingerprint spectrum is output in standard image formats (such as TIFF, PNG) or data file formats (such as HDF5). On the accompanying software interface, users can zoom and pan the spectrum, and click on any point to view the chip's original TRPL curve and detailed fitting parameters, achieving seamless traceability from macroscopic distribution to microscopic data. Simultaneously, the spectrum data can be directly interfaced with the manufacturing execution system on the production line to guide subsequent repair, dicing, or scrapping of specific defective areas.

[0076] Defect Feature Database Unit 1045 is the knowledge base for the system's intelligent classification, storing standard feature data for comparison. The database content includes: Theoretical standard value: The ideal (I0, τ) reference value calculated by a theoretical model based on the material parameters (band gap, quantum well structure, doping concentration, etc.) of the Micro LED under test.

[0077] Historical experimental data: A large number of samples with known failure modes are collected, calibrated by this system or other characterization methods (such as TEM, EDX), and their (I0, τ) feature vectors are stored in the database to form a "fingerprint" sample set of various defects.

[0078] Process window data: Statistical distribution data from good chips on the production line, defining the normal fluctuation range (I0, τ) of good chips (e.g., mean ± 3σ).

[0079] This database is not a simple table, but contains multiple "feature regions" defined by historical data and theoretical models.

[0080] Good Product Area: The database stores a large number of data points (strength, lifetime) collected from known qualified wafers on the production line. These points cluster in two-dimensional space to form a high-density area. The system uses statistical algorithms to depict the boundary of this area (e.g., defined as an ellipse or a polygon), and this area is the "good product area". Any chip falling within this area is directly judged as good.

[0081] Various Defect Feature Regions: For each known defect type (such as epitaxial dislocations, surface contamination, poor electrode contact, etc.), the database also stores the corresponding (intensity, lifetime) feature distribution. These features were obtained through extensive comparative experiments: a batch of failed chips whose defect types were clearly identified by other high-precision methods (such as transmission electron microscopy, focused ion beam cutting, etc.) were selected, and their (intensity, lifetime) parameters were measured using this system, thereby delineating the typical distribution region of this type of defect in two-dimensional space. For example, bulk defects usually reduce both intensity and lifetime simultaneously, and their points will fall in the lower left region; while simple poor electrode contact, because photoexcitation can still generate charge carriers but cannot inject them electrically, its lifetime may be close to normal but its intensity is extremely low, and its points will fall in the left region close to normal lifetime but with very low intensity.

[0082] The database has self-learning capabilities. When users confirm new defect types or correct existing classifications through other means (such as electrical testing or failure analysis), they can feed the new (I0, τ) samples or corrected labels back to the database. The system then updates the classification model through an incremental learning algorithm, making subsequent detections increasingly accurate.

[0083] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the disclosure herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0084] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope.

Claims

1. A wafer-level rapid defect location system for Micro LEDs based on photoluminescence, characterized in that, include: A broadband pulse excitation module is used to generate pulsed lasers adapted to the properties of the Micro LED material under test to excite the chip on the wafer to emit fluorescence; A two-dimensional scanning platform is used to carry and move the wafer under test to achieve chip-by-chip positioning; The time-resolved acquisition module is used to collect the fluorescence signal emitted by each chip under stimulated emission and convert it into time-resolved photoluminescence data; The data processing and control module is communicatively connected to the broadband pulse excitation module, the two-dimensional scanning platform, and the time-resolved acquisition module, respectively. The data processing and control module is used to control the two-dimensional scanning platform to move along a preset or dynamically planned path; synchronously control the broadband pulse excitation module to emit laser pulses at each detection point; receive and process the time-resolved photoluminescence data uploaded by the time-resolved acquisition module, extract the transient luminescence intensity and carrier lifetime of each detection point; and compare the two-dimensional features of the transient luminescence intensity and carrier lifetime with a preset defect feature database to determine the defect type of each chip and generate a wafer-level fingerprint spectrum reflecting the defect type and distribution.

2. The Micro LED wafer-level rapid defect location system based on photoluminescence as described in claim 1, characterized in that, The broadband pulse excitation module includes: The fundamental frequency laser generating unit is used to generate ultrashort pulse seed light of a fixed wavelength; The wavelength tuning and conversion unit includes a nonlinear optical crystal and an optical parametric oscillator, which is used to convert the seed light into excitation light with continuously tunable wavelength according to the quantum well bandgap characteristics of the MicroLED material under test. The pulse width modulation unit is used to adjust the pulse width of the excitation light in the range of 300 femtoseconds to 100 picoseconds according to the carrier recombination dynamics characteristics of the material under test; The spatial beam shaping unit is used to shape the modulated laser beam into a focused spot that matches the size of a single Micro LED chip; The synchronous triggering and control unit is used to receive the triggering command from the data processing and control module and coordinate the working timing of the fundamental frequency laser generation unit, wavelength tuning and conversion unit, and pulse width modulation unit.

3. The Micro LED wafer-level rapid defect location system based on photoluminescence as described in claim 1, characterized in that, The two-dimensional scanning platform includes: The closed-loop feedback control unit uses a laser interferometer or grating ruler to monitor the platform position in real time and forms a closed loop with the servo controller to ensure positioning accuracy. The wafer carrier and fixing unit uses a porous ceramic vacuum chuck to flatten and fix the wafer to be tested; The platform control and communication interface unit is used to receive motion commands from the data processing and control module and to provide real-time feedback of platform position and status signals. The platform control and communication interface unit feeds back the real-time position signal calculated by the closed-loop feedback control unit to the data processing and control module, so that the data processing and control module can generate a trigger signal that is synchronized with the motion state of the scanning platform.

4. The Micro LED wafer-level rapid defect location system based on photoluminescence as described in claim 1, characterized in that, The time-resolved acquisition module includes: Aperture microscopic optical collection unit, used to efficiently collect fluorescence signals emitted by stimulated emission from a single Micro LED chip; The spectral filtering and splitting unit, which includes a dichroic mirror and a narrowband filter, is used to separate and purify the fluorescence signal and suppress the excitation light scattering background. The single-photon level photoelectric conversion unit uses a single-photon avalanche diode or a hybrid photomultiplier tube to convert received fluorescent photons into electrical pulse signals. The time-correlated single-photon counting unit is used to receive the synchronization trigger signal sent by the data processing and control module as the timing start point, receive the electrical pulse output by the single-photon level photoelectric conversion unit as the timing end point, measure the arrival time of each photon, and accumulate to generate a time-resolved photoluminescence curve histogram.

5. The Micro LED wafer-level rapid defect location system based on photoluminescence as described in claim 1, characterized in that, The data processing and control module includes: The platform motion control and path planning unit is used to generate an initial scanning path covering all chips based on the wafer layout, and dynamically generate a re-inspection path only for chips with suspected defects based on real-time detection results, and control the two-dimensional scanning platform to perform corresponding movements. The synchronous timing control unit is used to receive the position signal fed back by the two-dimensional scanning platform, and based on a high-precision clock reference, generate an acquisition start trigger signal that is precisely synchronized with the laser pulse emission time, and send it to the broadband pulse excitation module and the time-resolved acquisition module respectively. The TRPL data acquisition and fitting analysis unit is used to receive time-resolved photoluminescence data uploaded by the time-resolved acquisition module and extract the transient luminescence intensity and carrier lifetime of each detection point through a nonlinear least squares fitting algorithm. The defect classification and pattern generation unit is used to construct a two-dimensional feature vector from the transient luminescence intensity and carrier lifetime of each detection point, compare it with the preset defect feature database, determine the defect type, and generate a wafer-level defect fingerprint pattern.

6. The Micro LED wafer-level rapid defect location system based on photoluminescence as described in claim 5, characterized in that, The data processing and control module also includes: The defect feature database unit is used to store two-dimensional feature distribution data of transient luminescence intensity and carrier lifetime corresponding to good products and various known defects.

7. The Micro LED wafer-level rapid defect location system based on photoluminescence as described in claim 5, characterized in that, The TRPL data acquisition and fitting analysis unit is also used to adaptively select a single exponential decay model or a double exponential decay model for fitting based on the decay pattern of the time-resolved photoluminescence curve; when using the double exponential model for fitting, the fast lifetime component, the slow lifetime component and their weights are extracted, and the equivalent carrier lifetime is calculated by weighted average.

8. The Micro LED wafer-level rapid defect location system based on photoluminescence as described in claim 6, characterized in that, The defect classification and map generation unit is also used to map the transient luminescence intensity and carrier lifetime of the chip under test to a two-dimensional feature space, calculate the Mahalanobis distance or probability density between the chip and each defect feature region in the defect feature database unit, and determine whether the chip belongs to a good product, epitaxial dislocation defect, surface state defect or electrode contact defect; and use different colors to represent different defect types and color depth to represent the severity of defects, generating a defect classification fingerprint map superimposed on the wafer physical coordinates.

9. The Micro LED wafer-level rapid defect location system based on photoluminescence as described in claim 5, characterized in that, The data processing and control module is also used to generate an optimized path that connects only the coordinates of all the suspected defective chips after the first round of global scanning and preliminary identification of the suspected defective chips, and to control the two-dimensional scanning platform to perform a second round of refined re-inspection scanning along the optimized path.

10. The Micro LED wafer-level rapid defect location system based on photoluminescence as described in claim 5, characterized in that, The synchronous timing control unit includes a high-precision digital delay generator, which is used to perform delay calibration on the acquisition start trigger signal sent to the time-resolved acquisition module based on the flight time of the laser pulse from the excitation module to the surface of the wafer under test, the flight time of the fluorescence signal from the sample to the detector, and the inherent delay of the electronic circuit, so as to ensure that the acquisition timing zero point is aligned with the time when the laser pulse arrives at the sample surface.

Images