Micro-LED Quality Validation

A defect inspection system that uses machine learning to improve defect detection in products. The system has a learning unit that trains a model to estimate multiple non-defective versions of an image. When inspecting a product, the system captures an image, inputs it into the trained model, and receives multiple estimated non-defective versions. By comparing the actual and estimated non-defective images, the system can more accurately identify and isolate defects.

Source

Inspection apparatus for defect detection in high-resolution display devices like micro-OLEDs without using sensor defect data. The method involves capturing an image of the display with sensor pixels, locating defect candidate positions, generating contours around the candidates, and determining if they are actual defects based on contour perimeter. This allows detection without sensor defect data, mitigating issues with sensor noise and pixel defects appearing as false positives.

Source

Testing large arrays of devices like microLED displays using parallel excitation with repeated contact and interaction without damaging the probes or devices. The test tool has multiple probes that can simultaneously contact multiple device contacts. Liquid metal is applied to the probe tips before contact to act as a flexible interface. When the probes touch the device contacts, oxide pinholes form in the contact layer allowing electrical connection through the liquid metal bridge. This allows repeated contact without damaging the probes or devices. The liquid metal seals when the probes lift, preventing flow onto the contacts. This enables parallel device excitation with optical monitoring without active feedback loops to maintain probe-contact alignment.

Source

A test system for efficiently and accurately measuring the performance of small form factor LEDs like micro-LEDs and nano-LEDs. The system uses a probe head with obliquely angled probes that can contact the tiny LEDs without blocking emitted light. It also has a movable cover over the probe slot to protect the probes when not in use. The probes measure voltage and current through the LEDs while a separate light detector captures emitted light. This allows calculating the quantum efficiency based on all three parameters.

Source

Determining quality of inspection data using a machine learning model that can distinguish between defective and non-defective products without requiring labeled training data for defective products. The method involves generating training data from non-defective products, learning the ML model on that, preparing a feature spectrum from a specific layer output for non-defective training data, and using that learned model and feature spectrum to determine quality of unseen inspection data. Similarity between the inspection data feature spectrum and the known feature spectrum is calculated and a threshold is used to determine if it's non-defective or defective.

Source

Contactless probe for measuring cycles of microdevices like LEDs without needing post-processing steps. The probe has an electrode covering a dielectric to stimulate the microdevice capacitively. A switch keeps the device on after an active portion of the stimulus signal. This allows measuring cycles without contact, resets, or parasitic effects. The stimulus amplitude can also be increased after each cycle.

Source

Detecting defects in display panels using deep learning to accurately locate and identify the type of defect in a stacked structure of a display panel. The method involves collecting images of defects and layers from a database, learning the defect and layer information using a deep learning model, and then using that model to extract the location of the defect in the stacked structure. The deep learning allows detecting detailed defect locations and layer associations compared to just checking for defects in a single image.

Source

Method to monitor the growth quality of nitride semiconductor films like GaN using a template substrate with a mask during molecular beam epitaxy (MBE) growth. The technique involves growing the nitride semiconductor on a template substrate with a mask pattern. By using light with a wavelength absorbed by the nitride semiconductor at the growth temperature, fringes are observed in the reflected light due to interference between the mask and the growing nitride film. This allows real-time monitoring of the nitride film growth without requiring a flat substrate.

Source

Display device with improved reliability by adding inspection lines that overlap the cathode-power electrode connection regions to check for defects. The inspection lines are disposed on the encapsulation layer covering the display pixels. They overlap the cathode-power connections and allow capacitance measurements to determine if the connections are stable. If the measured capacitance is within a range, it indicates good connection. If outside the range, it indicates defective connection. This provides a reliability check for the cathode-power connections to prevent issues like noise leakage through poorly connected cathodes.

Source

Determining epitaxial defects in semiconductor substrates using frequency domain filtering and machine learning to improve yield and manufacturing efficiency. The method involves applying a frequency domain filter to substrate images to enhance defect visibility. Then feature detection is performed on the filtered images to count epitaxial defects. This allows automated, scalable defect classification and counting compared to manual review of raw images. The defect count can trigger corrective actions to address the underlying issues.

Source

Pixel circuit for display devices that allows simultaneous driving and sensing of electrical characteristics of multiple driving transistors for a pixel to improve reliability, efficiency, and HDR contrast. The circuit has two driving transistors connected to the pixel's light-emitting element and a sensing part to simultaneously drive one transistor while sensing the electrical characteristics of the other transistor. This allows real-time monitoring of transistor performance while simultaneously emitting light. It enables selection of the best performing transistor for driving the light-emitting element and detects deterioration early. The simultaneous emission of light from both transistors increases luminance with less current and improves display performance.

Source

A non-destructive method to analyze and monitor defects in semiconductor structures using femtosecond laser beams and terahertz waves. The method involves injecting femtosecond lasers into the semiconductor to create electron excitations, then irradiating terahertz waves onto the semiconductor while the excitations recombine. The terahertz wave transmittance/reflectance is measured and analyzed to determine defect densities and distributions in the semiconductor. The analysis involves tracking the recombination time constants of the excited carriers. This allows non-contact, non-destructive analysis of defects in complex 3D semiconductor structures.

Source

Self-monitoring method for displays to detect abnormalities without manual intervention. The method involves inputting a test image, extracting display parameters for each frame of the displayed test image, comparing the parameters between frames to find variations, and determining if display abnormalities are present based on the parameter comparisons. By feeding back test images with known variations, the display can learn normal parameter ranges and detect aberrations.

Source

Efficiently monitoring production line conformance using categorical validation machine learning models trained on a subset of related categories for a target category. The models are trained using training images from similar categories to the target category. This reduces the number of training images needed compared to using a large, diverse set of images. By tailoring the training data to the target category, the models are better at distinguishing between the target category and similar categories during inference. This allows more reliable and efficient production line monitoring.

Source

Detecting defective pixels in an image array using statistical analysis on sets of surrounding pixels. For each pixel, a cell with neighboring pixels is analyzed statistically. If the statistical distance of the pixel from the cell exceeds a threshold, it indicates a defective pixel. This provides a way to detect defective pixels without relying on calibration or edge detection, as the statistical outlier can be identified on-the-fly during image processing.

Source

Predicting defects in assembly units using machine learning techniques to enable real-time yield protection in optical inspection systems. The method involves extracting features from inspection images, generating vectors, labeling clusters based on inspection outcomes, and propagating defect labels upstream to detect anomalous regions in new assembly units. It uses machine learning to identify common features associated with proper function versus defects, and can assist in confirming defects by comparing against these feature sets.

Source

Display apparatus with improved diagnostics for microLED displays. It allows detecting panel issues during vertical blank periods when the display is off. The display has multiple microLED drivers with timings controlled by separate clock signals. During blanking, all clock signals go low to disable emitting in all pixels. This lets detecting abnormalities like stuck pixels.

Source

A method for non-destructively detecting LEDs bonded to a display substrate using a light source and a sensor. The method involves irradiating the LEDs with light from the source and measuring the emitted light with the sensor. By analyzing the spectral characteristics of the emitted light, the quality and performance of the LEDs can be assessed without physically contacting or damaging them. The method enables high-precision LED detection without requiring expensive equipment like lasers or probes.

Source

Display manufacturing apparatus that removes foreign substances from micro LEDs before assembly to prevent defects in the finished display. The apparatus has two modules for filtering out non-magnetic and magnetic foreign substances from the micro LEDs. The first module removes non-magnetic contaminants using filtration. The second module removes magnetic foreign substances using a magnetic field. This ensures the micro LEDs are clean before assembly on the display substrate to prevent issues like misalignment, electrical shorts, and lighting defects.

Source

Contactless micro LED inspection system that uses pulsed lasers to detect defective micro LEDs without direct electrical testing. The system emits pulsed lasers at the LEDs to generate photovoltaic radio frequency signals when radiated. An antenna receives these signals, which are amplified and analyzed by a processor to determine if the LED is functioning or defective. This allows high-efficiency, contactless testing of micro LED arrays without needing to electrically test each individual tiny LED.

Source