Machine learning networks for screening quantum devices

Abstract

To provide a method, a device, a system, and a computer program product for screening a quantum device.SOLUTION: In a screening environment, a computer system sends a feature extracted from an image of a quantum device to a classification neural network, specifies a set of features of the quantum device from a group of mutually exclusive features on the basis of features specified in the image of the quantum device, makes a setting so that the specified set of features will be output, and receives the set of features specified by the classification neural network for the quantum device. The set of features shows whether the quantum device is desirably working.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present disclosure relates generally to improved computer systems, and particularly to methods, apparatus, systems, and computer program products for screening quantum devices. [Background technology]

[0002] In semiconductor manufacturing, yield is a quantitative measurement of the semiconductor process. This yield is the fraction of dies on a wafer that are not discarded during the manufacturing process. The ultimate goal of the manufacturing process is to produce only working dies. In other words, this ultimate goal may be to have a 100% yield. However, in reality, manufacturing processes are imperfect. Not all dies resulting from the manufacturing process will operate as desired within the specifications of the quantum devices on the dies.

[0003] Various factors can affect the yield of die from a wafer, such as inconsistencies such as particle contamination, lithography misalignment, lateral dimension variations caused by etching, purity variations caused by implantation processes, purity variations caused by deposition processes, and other factors that can reduce the yield of the fabrication process, which can result in die with quantum devices that do not work or that work outside of specifications.

[0004] It would therefore be desirable to have methods and apparatus that take into account at least some of the problems discussed above, as well as other potential problems, such as overcoming the technical problems associated with screening quantum devices to determine whether they function as desired. Summary of the Invention

[0005] One embodiment of the present disclosure provides a device screening system including a computer system, a feature machine learning model system in the computer system, and a classification neural network in the computer system. The feature machine learning model system is configured to receive an image of a quantum device, identify features of the quantum device in the image, and output the identified features for the quantum device in the image of the quantum device. The classification neural network is configured to receive the features of the quantum device identified in the image of the quantum device, identify a set of characteristics of the quantum device from a group of mutually exclusive characteristics based on the features identified in the image of the quantum device, and output the identified set of characteristics for the quantum device identified in the image of the quantum device. The set of characteristics is indicative of whether the quantum device will perform desirably.

[0006] Another embodiment of the present disclosure provides a device screening system including a computer system and a classification machine learning model system in the computer system, wherein the classification machine learning model system is configured to receive features of a quantum device identified in an image of the quantum device, identify a set of characteristics for the quantum device based on the features identified in the image of the quantum device, and output the set of characteristics for the quantum device based on the features identified in the image of the quantum device, wherein the set of characteristics indicates whether the quantum device will perform in a desired manner.

[0007] Yet another embodiment of the present disclosure provides a method for screening quantum devices. A computer system transmits features extracted from an image of the quantum device to a classification neural network. The classification neural network is configured to identify a set of characteristics of the quantum device from a group of mutually exclusive characteristics based on the features identified in the image of the quantum device and output the identified set of characteristics for the quantum device. The computer system receives the set of characteristics identified by the classification neural network for the quantum device. The set of characteristics indicates whether the quantum device will function in a desired manner.

[0008] Another embodiment of the present disclosure provides a computer program product for screening quantum devices, the computer program product including a computer-readable storage medium having first, second, and third program codes stored thereon. The first program code is executable by the computer system to cause the computer system to receive features, the features being received in output from a feature extraction neural network configured to receive images of the quantum devices, and output features identified in the images of the quantum devices. The second program code is executable by the computer system to cause the computer system to send the features extracted from the images of the quantum devices to a classification neural network. The classification neural network is configured to identify a set of characteristics of the quantum devices based on the features identified in the images of the quantum devices. The third program code is executable by the computer system to cause the computer system to receive a set of characteristics identified by the classification neural network for the quantum devices based on the features identified in the images of the quantum devices, the set of characteristics indicating whether the quantum devices are functioning desirably.

[0009] These features and functions may be realized alone in various embodiments of the present disclosure or may be combined in yet further embodiments, further details of which can be seen with reference to the following description and drawings.

[0010] The novel features believed characteristic of the illustrative embodiments are set forth in the appended claims. However, the illustrative embodiments, together with their preferred modes of use, further objects and features thereof, will best be understood by reference to the following detailed description of illustrative embodiments of the present disclosure taken in conjunction with the accompanying drawings. [Brief explanation of the drawings]

[0011] [Figure 1] 1 is an illustration of a network of data processing systems in which illustrative embodiments may be implemented; [Figure 2] FIG. 1 is an illustration of a block diagram of an analytics environment in accordance with an illustrative embodiment. [Figure 3] FIG. 1 is an example of a block diagram of a machine learning model network, according to an illustrative embodiment. [Figure 4] FIG. 1 is an example of a block diagram of a neural network, in accordance with an illustrative embodiment; [Figure 5] FIG. 1 is an example block diagram of a process for training a classification machine learning model system, according to an illustrative embodiment. [Figure 6] FIG. 1 is an example of a flow diagram of a process for screening quantum devices, in accordance with an illustrative embodiment. [Figure 7] FIG. 1 is an example of a flow diagram of a process for screening quantum devices, in accordance with an illustrative embodiment. [Figure 8] 1 is an example of a block diagram of a data processing system in accordance with an illustrative embodiment; [Figure 9] 1 is an illustration of an example aircraft manufacturing and service method in accordance with an illustrative embodiment; [Figure 10] 1 is an example of a block diagram of an aircraft in which an illustrative embodiment may be implemented; [Figure 11] 1 is an illustration of a block diagram of a product management system in accordance with an illustrative embodiment; DETAILED DESCRIPTION OF THE INVENTION

[0012] Exemplary embodiments recognize and take into account one or more different considerations. For example, exemplary embodiments recognize and take into account that lithography and etching processes are a major source of lateral dimensional variations that cause quantum devices to operate in an undesired manner (e.g., out of specification). For example, lithography and etching processes can cause parametric and functional failures through mismatches (e.g., gate length variations, contact overlap, or other undesirable mismatches). Exemplary embodiments recognize and take into account that many of these mismatches are system-based and layout pattern-dependent.

[0013] The exemplary embodiments recognize and take into account that the above challenges increase as the number of quantum devices increases. The exemplary embodiments recognize and take into account that imperfectly formed structures can lead to unexpected defects and openings in structures (such as quantum devices) unless lithography and etching processes are carefully controlled. The exemplary embodiments recognize and take into account that some quantum devices (e.g., Josephson junctions) may not include semiconductor material. The exemplary embodiments recognize and take into account that screening for the above problems currently occurs through electrical testing performed at later stages of quantum device fabrication or after fabrication is complete. The exemplary embodiments recognize and take into account that screening for such problems could be performed earlier in the process using visual inspection performed by operating personnel. The exemplary embodiments recognize and take into account that this type of inspection is at least one of subjective, time-consuming, and / or limited to a small subset of quantum devices.

[0014] Thus, the illustrative embodiments provide methods, apparatus, systems, and computer program products for identifying inconsistencies in quantum devices. Further, the illustrative embodiments may identify inconsistencies in processing before fabrication of the quantum devices is complete. In examples, this type of screening for inconsistencies in processing may allow quantum devices to be reworked or steps in the fabrication of the quantum devices to be adjusted to reduce the number of quantum devices that may not function as desired.

[0015] In an embodiment, a neural network system is used to identify nonconformities in a more efficient and rapid manner than visual inspection performed by personnel. Additionally, images may be analyzed using a screening system at various stages of fabrication so that at least one of fabrication process modifications, quantum device rework, or other steps may be implemented to increase yield in the fabrication of quantum devices on the wafer.

[0016] Embodiments provide methods, apparatus, systems, and computer program products for screening quantum devices. A computer system transmits features extracted from an image of the quantum device to a classification neural network. The classification neural network is configured to identify a set of characteristics of the quantum device from a group of mutually exclusive characteristics based on the features identified in the image of the quantum device and output the identified set of characteristics for the quantum device. The computer system receives the set of characteristics identified by the classification neural network for the quantum device. The set of characteristics indicates whether the quantum device will function in a desired manner.

[0017] Referring now to the figures, and particularly to the figures relating to FIG. 1 , there is shown an illustration of a network of data processing systems in which exemplary embodiments may be implemented. Network data processing system 100 is a network of computers in which exemplary embodiments may be implemented. Network data processing system 100 contains network 102, which is the medium used to provide communications links between the various devices and computers connected together within network data processing system 100. Network 102 may include connections, such as wired or wireless communications links or fiber optic cables.

[0018] In the illustrated example, server computer 104 and server computer 106, along with storage unit 108, are connected to network 102. In addition, client device 110 is also connected to network 102. As illustrated, client devices 110 include client computer 112, client computer 114, and client computer 116. Client devices 110 may be, for example, computers, workstations, or network computers. In the illustrated example, server computer 104 provides information (such as boot files, operating system images, and applications) to client device 110. Additionally, client device 110 may also include other types of client devices (such as mobile phone 118, tablet computer 120, and smart glasses 122). In this example, server computer 104, server computer 106, storage unit 108, and client device 110 are network devices connected to network 102, and network 102 is the communication medium for these network devices. Some or all of the client devices 110 may form an Internet of Things (IoT), in which these physical devices are connected to the network 102 and may exchange information with each other through the network 102.

[0019] In this example, client device 110 is a client to server computer 104. Network data processing system 100 may include additional server computers, client computers, and other devices not shown. Client device 110 is connected to network 102 using at least one of a wired connection, an optical fiber connection, or a wireless connection.

[0020] Program code located within network data processing system 100 may be stored on a computer-recordable storage medium and downloaded for use to a data processing system or other device. For example, program code may be stored on a computer-recordable storage medium of server computer 104 and downloaded over network 102 to client device 110 for use by client device 110.

[0021] In the depicted example, network data processing system 100 is the Internet with network 102, which is a worldwide collection of networks and gateways that use the TCP / IP (Transmission Control Protocol / Internet Protocol) suite of protocols to communicate with each other. At the heart of the Internet is a backbone of high-speed data communication lines between major nodes, or host computers, consisting of thousands of commercial, government, educational, and other computer systems that route data and messages. Of course, network data processing system 100 may also be implemented using a number of different types of networks. For example, network 102 may be comprised of at least one of the Internet, an intranet, a local area network (LAN), a metropolitan area network (MAN), or a wide area network (WAN). Figure 1 is intended as an example, and not as architectural limitations for the different illustrative embodiments.

[0022] As used herein, the term "number of," when used in reference to an item, means one or more items. For example, "a number of different types of networks" refers to one or more different types of networks.

[0023] Furthermore, the phrase "at least one of," when used in conjunction with a list of items, means that various combinations of one or more of the listed items may be used, and that only one of each listed item may be required. In other words, "at least one of" means that any combination of items and any number of items from the list may be used, and not all of the listed items are required. An item may be a specific object, article, or category.

[0024] By way of example and not limitation, "at least one of item A, item B, or item C" may include "item A," "item A and item B," or "item B." Examples may also include "item A, item B, and item C," or "item B and item C." Of course, any combination of these items may be present. In some embodiments, "at least one of" may be, by way of example and not limitation, "two items A," "one item B," and "ten items C," "four items B and seven items C," or any other suitable combination.

[0025] In this example, camera system 130 is in communication with client computer 112. As shown, camera system 130 operates to generate image 132 of quantum device 134. In this example, camera system 130 may be one or more cameras. Camera system 130 may include at least one of a scanning electron microscope, a transmission electron microscope, an infrared microscope, an optical microscope, an X-ray microscope, a scanning probe microscope, a scanning tunneling microscope, a scanning transmission electron microscope, a scanning Auger microscope, a Raman microscope, a photoluminescence microscope, an atomic force microscope, a conduction atomic force microscope, a scanning capacitance microscope, a scanning thermal microscope, a magnetic force microscope, a piezoelectric force microscope, a near-field scanning microwave optical microscope, a Kelvin probe microscope, a scanning superconducting quantum interference device (SQUID) microscope, a NV center magnetic microscope, or any other suitable type of microscope.

[0026] As shown, quantum devices 134 are disposed on a wafer 136. In this example, quantum devices 134 may be at various stages of processing on wafer 136. For example, quantum devices 134 may include at least one of completed quantum devices, partially completed quantum devices, or quantum devices in a certain phase of fabrication. In other examples, wafer 136 may be singulated into dies on which quantum devices 134 are disposed.

[0027] In this example, image 132 may be analyzed by device screening system 138. As shown, image 132 is input to device screening system 138. Device screening system 138 processes image 132 and outputs information in the form of properties 140 of quantum device 134 determined from image 132.

[0028] As shown, device screening system 138 includes a feature machine learning model system 142 and a classification neural network 144. Feature machine learning model system 142 operates to identify features found in images 132 of quantum devices 134. Such features are output to classification neural network 144, which uses such features to identify characteristics 140 of quantum devices 134. In this example, classification neural network 144 may classify the features into classes that correspond to characteristics 140 of quantum devices 134.

[0029] In this embodiment, characteristic 140 may be used to make modifications to the processing of quantum device 134. For example, if fabrication of quantum device 134 is not complete, modifications may be made so that quantum device 134 may be salvaged. For example, modifications or rework of the processing of quantum device 134 may be performed. In some embodiments, modifications may be made to future fabrications of quantum device 134, thereby avoiding undesirable characteristics of quantum device 134. For example, modifications to at least one of lithography, etch time, dopant concentration, or other modifications may be made to improve characteristic 140 of quantum device 134.

[0030] 2, a block diagram of an analysis environment is shown in accordance with an illustrative embodiment. In this example, screening environment 200 includes components that may be implemented in hardware (e.g., the hardware shown in network data processing system 100 in FIG. 1).

[0031] As shown, screening environment 200 is an environment in which device screening system 202 may analyze image 204 of quantum device 206. As shown, image 204 may be generated by camera system 207. Camera system 207 may include one or more cameras.

[0032] In this example, quantum device 206 is a structure used to at least one of process and store quantum information. Many types of quantum device 206 are semiconductor structures, although some quantum devices 206 do not use semiconductors. Quantum device 206 may be selected from at least one of a quantum dot device, a nanowire device, a quantum well device, a quantum information processing device, a quantum memory, a superconducting resonator, a Josephson junction, a nonlinear inductor, a linear inductor, a capacitor, an optical resonator, a diode, a transistor, a field effect transistor, a memory device, a quantum interference device, a topological quantum device, a waveguide, an optical resonator, or any other suitable quantum device.

[0033] In this embodiment, quantum device 206 is disposed on substrate 208. Substrate 208 may take a number of different forms. For example, substrate 208 may be selected from at least one of a semiconductor substrate, a silicon substrate, a sapphire (Al2O3) substrate, a silicon thermal oxide substrate, a silicon-on-insulator substrate, or other suitable substrate. In an embodiment, substrate 208 may take the form of a wafer.

[0034] Inconsistencies 210 may arise in the fabrication of quantum devices 206 or in the final structure of quantum devices 206. Various steps performed at various stages in the process to fabricate quantum devices 206 may result in a set of inconsistencies 210 in one or more of quantum devices 206. In addition to, or instead of, physical variations in quantum devices 206, there may also be inconsistencies 210 in the fabrication process that cause the quantum devices to not function as expected.

[0035] Mismatches can cause a quantum device to function improperly, whether the mismatch is in the final structure of the quantum device, a misalignment of a temporary structure used during the fabrication of the quantum device, or any other type of mismatch that can affect the functionality of the quantum device.

[0036] An example of a temporary structure is resist. An inconsistency in resist can be incorrect exposure or development of the resist, misalignment of the resist, incorrect thickness of the resist, or some other undesirable characteristic. Resist is a layer material used to transfer a pattern into a structure (e.g., a substrate or structure on which a quantum device is to be fabricated). Resist can be patterned by lithography to form a temporary mask that protects underlying portions of the structure in subsequent processing steps. The resist used in photolithography is photoresist.

[0037] In this embodiment, device screening system 202 may operate to analyze quantum device 206 at various stages of processing. For example, image 204 of quantum device 206 may be generated at a set of stages of fabricating quantum device 206 selected from at least one of developing a resist on the quantum device, patterning a thin film on the quantum device, or etching the quantum device.

[0038] In this example, quantum devices 206 may be processed to re-pattern the resist if incompatible, which may result in an increased yield of quantum devices 206 on substrates 208 by mitigating problems that may arise at various stages of processing of quantum devices 206.

[0039] In this example, the device screening system 202 includes a screening manager 212, a machine learning model 214, and a computer system 216. The screening manager 212 and the machine learning model 214 execute on the computer system 216.

[0040] As shown, screening manager 212 may communicate with machine learning model 214 and control the operation of machine learning model 214 to screen quantum device 206. For example, screening manager 212 may receive images 204 and input such images into machine learning model 214. Screening may be performed at one or more stages of processing of quantum device 206.

[0041] At least one of the screening manager 212 or the machine learning model 214 may be implemented in software, hardware, firmware, or a combination thereof. If software is used, the processes performed by at least one of the screening manager 212 or the machine learning model 214 may be implemented in program code configured to run on hardware (such as a processor unit). If firmware is used, the processes performed by at least one of the screening manager 212 or the machine learning model 214 may be implemented in program code and data, which may be stored in persistent memory and executed on a processor unit. If hardware is used, the hardware may include circuitry operative to perform the processes in at least one of the screening manager 212 or the machine learning model 214.

[0042] In an embodiment, the hardware may take the form of at least one of a circuit system, an integrated circuit, an application specific integrated circuit (ASIC), a programmable logic device, or any other suitable type of hardware configured to perform some process. When a programmable logic device is used, the device may be configured to perform some process. The device may be later reconfigured or permanently configured to perform some process. Programmable logic devices include, for example, programmable logic arrays, programmable array logic, field programmable logic arrays, field programmable gate arrays, and other suitable hardware devices. Additionally, processes may be implemented in organic components integrated with inorganic components and may be composed entirely of non-human organic components. For example, processes may be implemented as circuits in organic semiconductors.

[0043] Computer system 216 is a physical hardware system and includes one or more data processing systems. When more than one data processing system is present in computer system 216, such data processing systems communicate with each other using a communication medium. This communication medium may be a network. The data processing systems may be selected from at least one of a computer, a server computer, a tablet computer, or any other suitable data processing system.

[0044] Examples of machine learning models that may be used to implement machine learning model 214 include at least one of a neural network, a decision tree, a support vector machine, a Bayesian network, a genetic algorithm, a cluster analysis algorithm, or any other type of machine learning model.

[0045] A machine learning model is a type of artificial intelligence model that can learn without being explicitly programmed. The machine learning model 214 may be selected from at least one of a supervised machine learning model or an unsupervised machine learning model. The supervised machine learning model may include at least one of a regression machine learning model or a classification machine learning model. This type of machine learning model is based on the type of training algorithm used to train the machine learning model.

[0046] A machine learning model may learn based on training data input to the machine learning model. The machine learning model may learn using various types of machine learning algorithms, including at least one of supervised learning, unsupervised learning, feature learning, sparse directory learning, anomaly detection, association rules, or other types of learning algorithms.

[0047] In this example, machine learning models 214 are organized to form machine learning model network 217. As shown, machine learning model network 217 includes feature machine learning model system 218 and classification machine learning model system 220. As shown, feature machine learning model system 218 receives image 204 of quantum device 206. In this example, device screening system 202 may be a two-stage machine learning model system, where feature machine learning model system 218 performs feature extraction and classification machine learning model system 220 performs classification.

[0048] In this example, a screening manager 212 may receive the image 204. The screening manager 212 may input the image 204 into a feature machine learning model system 218 for feature extraction.

[0049] As shown, feature machine learning model system 218 may identify features 222 of quantum device 206 within image 204. Additionally, feature machine learning model system 218 may output the identified features 222 for quantum device 206 within image 204 of quantum device 206.

[0050] The feature machine learning model system 218 may be implemented using several different types of machine learning models. For example, the feature machine learning model system 218 may be a convolutional neural network (CNN) trained to process images and identify features within the images. In other words, the convolutional neural network may extract features.

[0051] As shown, classification machine learning model system 220 receives features 222 of quantum device 206 identified within image 204 of quantum device 206. In an embodiment, features 222 may be sent directly from feature machine learning model system 218 to classification machine learning model system 220.

[0052] As shown, the classification machine learning model system 220 may identify a set of characteristics 224 of the quantum device 206 based on features 222 identified in the image 204 of the quantum device 206 and output the set of identified characteristics 224 for the quantum device 206.

[0053] In this example, the set of characteristics 224 indicates whether quantum devices 206 are functioning desirably. The set of characteristics 224 may be analyzed to determine whether one or more of quantum devices 206, or all of quantum devices 206, are functioning desirably. In other words, the set of characteristics 224 may indicate that a first portion of quantum devices 206 are functioning desirably, while a second portion of quantum devices 206 are not functioning desirably. In another example, the set of characteristics 224 may indicate that none of quantum devices 206 are functioning desirably. In yet another example, the set of characteristics 224 may indicate that all of quantum devices 206 are functioning desirably.

[0054] In an embodiment, the set of characteristics 224 may be identified from a group of mutually exclusive characteristics 226. As used herein, the term "group of," when used in reference to an item, means one or more items. For example, a "group of mutually exclusive characteristics 226" refers to one or more of the mutually exclusive characteristics 226.

[0055] Mutually exclusive characteristics 226 mean that multiple characteristics cannot exist simultaneously. As shown, classification machine learning model system 220 may classify features 222 into classes 228. In this example, classes 228 may include at least one of exposure, feature decay, component, etch amount, alignment, or any other suitable class.

[0056] The properties 224 in a class of the classes 228 are mutually exclusive properties 226. For example, a class can be exposure of a resist. The property of exposure can be overexposed, underexposed, or correct exposed. These properties are mutually exclusive because, for a particular feature, if one of the three properties is present, the other two cannot be present.

[0057] As another example, another class may be a component. In this class, a property may be the presence of a particle and the absence of a particle. Such two properties for a component are mutually exclusive. Thus, there exists a class 228 in which the sets of properties 224 in the class are mutually exclusive among the plurality of classes 228.

[0058] In an embodiment, the screening manager 212 may use the characteristics 224 to form a set of actions 232. In an embodiment, the characteristics 224 may be arranged in or categorized into classes 228.

[0059] The set of actions 232 may take several forms, for example, the set of actions 232 may be selected from at least one of generating a log entry, sending an alert, taking corrective action if a non-conformance exists, initiating rework of the set of wafers, releasing the set of wafers for further processing, selecting a set of quantum devices for further analysis, or any other suitable action.

[0060] Referring now to Figure 3, an example of a block diagram of a machine learning model network is shown, according to an illustrative embodiment. In examples, the same reference number may be used in more than one figure. Such repeated reference numbers in different figures represent the same element in such different figures.

[0061] As shown, machine learning model network 217 is an example of one manner in which machine learning model network 214 of Figure 2 may be organized to screen quantum device 206. In this example, machine learning model network 217 may be comprised of multiple neural networks that form sub-networks in machine learning model network 217.

[0062] As shown, machine learning model network 217 includes convolutional neural network (CNN) 300 and classification neural network 302. In this example, convolutional neural network 300 is an example of an implementation of feature machine learning model system 218 in Figure 2. Classification neural network 302 is an example of an implementation of classification machine learning model system 220 in Figure 2.

[0063] In this example, convolutional neural network 300 may have an architecture selected to perform high-throughput image interpretation. In other words, convolutional neural network 300 may be configured to process images of individual wafers, each of which may contain thousands of quantum devices for extracting features 222 of FIG. 2. Features 222 may take a variety of forms. For example, features 222 may be selected from at least one of edges, contrast differences, brightness levels, or any other suitable type of feature that convolutional neural network 300 may output.

[0064] In one embodiment, the convolutional neural network 300 may be a block structure that includes an extraction block 304. For example, a 14-block structure may be used, where each block contains the processing with independent convolution, batch formulation, and validation performed by currently used convolutional neural networks that extract features from images.

[0065] In this embodiment, each extraction block of the plurality of extraction blocks 304 may double the channels and halve the resolution. In other words, in this particular example, each extraction block may increase the number of channels and reduce the resolution. As a result, the layers of the network may identify features within the image. Subsequent layers of the network may identify larger features compared to previous layers of the network. The output of the previous layer is input to the subsequent layer. In other implementations, the extraction blocks do not necessarily reduce the channels or resolution, depending on performance specifications.

[0066] During processing of the image 204, the extraction blocks 304 in the convolutional neural network 300 distribute information among an increasing number of increasingly abstract channels. In this example, the last of the extraction blocks 304 may output features 222 that have been grouped using global pooling to form a tensor 306. The tensor 306 may be an array of components, in this example, a 1024-dimensional collection of numbers that contains information about the features 222 extracted from the image 204.

[0067] As a result, features 222 are represented by numbers in Tencels 306 in a manner that preserves local and global structure between images. Tencels 306 form the input to classification neural network 302.

[0068] In this example, the classification neural network 302 may be implemented using neurons 308. The neurons 308 may be organized as a set of subnetworks 310 of neurons 308 corresponding to multiple classes 228, with mutually exclusive sets of characteristics 224 within the classes 228. In other words, each subnetwork in the set of subnetworks 310 of neurons 308 corresponds to one class of the multiple classes 228.

[0069] The classification neural network 302 may receive the features 222 of the quantum device 206 identified in the image 204 of the quantum device 206 of Figure 2 in a tensor 306 output by the convolutional neural network 300. Additionally, the classification neural network 302 may identify a set of properties 224 of the quantum device 206 from the group of mutually exclusive properties 226 of Figure 2 and output the set of properties 224. In this example, the set of properties 224 identified for the group of mutually exclusive properties 226 of Figure 2 is based on the features 222 identified in the image 204 of the quantum device 206, and the features 222 are among the information disposed in the tensor 306. The set of properties 224 indicates whether the quantum device 206 is functioning as desired.

[0070] In this example, classification neural network 302 may be a three-layer network. In this example, each layer may be 1,024 neurons wide. In other examples, other numbers of layers of neurons may be used, depending on the implementation.

[0071] Additionally, classification neural network 302 may classify features 222 represented by information in tensor 306 into classes 228. Each class in multiple classes 228 may be a mutually exclusive characteristic 226 of FIG.

[0072] Referring now to Figure 4, an example block diagram of a neural network is shown in accordance with an illustrative embodiment. In this example, neural network 400 is an example of one of subnetworks 310 in classification neural network 302 in Figure 3. In this depicted example, neural network 400 classifies features into multiple characteristics per class.

[0073] As shown, neural network 400 has two layers (layer 404 and layer 406). In this example, each layer contains four neurons. Layer 404 contains neuron 408, neuron 410, neuron 412, and neuron 414. In this example, layer 406 contains neuron 416, neuron 418, neuron 420, and neuron 422. Each neuron in layer 404 has an output that is connected to a neuron in layer 406. These neurons represent a subnetwork for a class that encompasses four mutually exclusive properties.

[0074] As shown, neural network 400 receives tensor 424. In this example, tensor 424 contains features extracted by a feature machine learning model system. Each feature in tensor 424 is sent to one neuron in layer 404. In this example, tensor 424 contains four features. In this example, characteristic probabilities 426 are output from layer 406. In this example, each neuron in layer 406 outputs the probability that a particular characteristic is present in the image.

[0075] In this example, four features are output from layer 406 for one class, and within this class, the features are mutually exclusive. Neural network 400 is a simplified example of an implementation of one of the subnetworks 310 of FIG. 3, with a small number of neurons shown to illustrate the features in this example in a manner that avoids obscuring them by illustrating three layers 1,024 neurons wide that are grouped into multiple subnetworks for the multiple classes that encompass the features. This simplified example is shown to illustrate the inputs and outputs for a neural network (e.g., a subnetwork) trained to classify features, with probabilities for mutually exclusive features being the output of one class.

[0076] Referring now to FIG. 5, an example block diagram for training a classification machine learning model system is shown in accordance with an illustrative embodiment. In this example, training data 500 includes images 502. Images 502 may be any images of quantum devices at one or more stages of processing. In one example, images 502 may include quantum devices having at least one of undesirable characteristics or desirable characteristics of such quantum devices in images 502. In this example, images 502 may be comprised of thousands of scanning electron microscope (SEM) images of quantum devices.

[0077] In this example, the training data 500 may also include metadata 504. The metadata 504 may be, for example, labels that indicate characteristics to be identified within a particular image. The metadata 504 may be used in conjunction with a set of machine learning algorithms 506 to train the classification machine learning model system 220.

[0078] This training may be performed by a training manager 508 using a set of machine learning algorithms 506. In this example, the training manager 508 executes on the computer system 216 of Figure 2. In this example, the set of machine learning algorithms 506 is a set of supervised machine learning algorithms. As shown, the feature machine learning model system 218 has already been trained to extract features 222.

[0079] In this example, when the classification machine learning model system 220 is implemented using a classification neural network 302, overfitting can be reduced by the training manager 508 using at least one of dropout 510 or regularization 512.

[0080] In this example, dropout 510 involves disabling a set of neurons 308 during training of classification neural network 302 so that overfitting is reduced. Disabling a neuron, in this example, involves disconnecting the neuron from at least one of its inputs or outputs. In an example, this disabling may be performed randomly.

[0081] As shown, regularization 512 is a technique used to reduce the likelihood of overfitting the classification neural network 302. Overfitting can occur when training a neural network. The neural network may predict the output values ​​of training data very well, with very high accuracy and little error. However, when the trained network receives new, never-before-seen data, the neural network may produce poor predictions.

[0082] In this example, the regularization 512 may be an L1 regularization, in which a term may be added to an error function used by the set of machine learning algorithms 506. This error function is configured to penalize large magnitude weight values ​​generated during training. For example, the L1 regularization may be used by the training manager 508 in conjunction with the machine learning algorithms 506, such that the set of neurons 308 have a set of amplitudes that are reduced during training of the classification neuron network 302 so that overfitting is reduced.

[0083] After training is complete, the classification neural network 302 may then be used to determine characteristics 224 in classes 228 in Figures 2-3. Additionally, validation may be performed by the training manager 508 to ensure that the classification neural network 302 is performing as desired. A portion of the training data 500 may be saved for validation. For example, 80 percent of the training data 500 may be used for training, while 20 percent of the training data 500 may be used for validation.

[0084] Thus, compared to current techniques, the machine learning model network 217 having two stages, the feature machine learning model system 218 and the classification machine learning model system 220, may identify the characteristics 224 of the quantum device 206 of Figure 2 more accurately and with a desirable level of speed. Furthermore, the use of at least one of the mutually exclusive characteristics 226, dropout 510, or regularization 512 of Figure 2 may improve the performance of the classification machine learning model system 220. This improved performance may include improved speed in addition to maintaining or improving accuracy in determining the characteristics 224 of the quantum device 206 of Figure 2.

[0085] In one embodiment, one or more technical solutions exist that overcome the technical problem of screening quantum devices to determine whether they function desirably. As a result, the one or more technical solutions may provide a technical effect of enabling screening of quantum devices at one or more stages of processing of the quantum devices on a substrate (e.g., a wafer) in a manner that enables identification of characteristics that indicate that the quantum devices will not function desirably. Such characteristics, in some embodiments, may also be referred to as failure modes.

[0086] In an embodiment, the one or more technical solutions provide a technical effect of enabling device screening of quantum devices to be performed in-line after a processing step (e.g., lithographic patterning), for example, at approximately 1000 per minute. In an embodiment, this technical effect may reduce the feedback learning cycle from months to hours (or even shorter). Currently, when testing the functionality of quantum devices, months are required to identify problems. This time may be reduced by using the machine learning model network 217 of Figures 2-3 and 5 to perform screening at various processing steps to identify problems. This may allow rework steps to be performed in a timely manner, which may in turn improve the yield of quantum devices.

[0087] In embodiments, the machine learning model network 217 of Figures 2-3 and 5 may automatically perform root cause analysis on a class of characteristics that may be used to identify problems. Such problems may be the presence of particles, missing gates, missing leads, or other issues that may cause a quantum device to not function as desired. These types of problems may result in electrical defects (such as defects or openings) that are currently not identified until the completed device is subjected to electrical testing.

[0088] 2 may be configured to perform at least one of the steps, processes, or actions described in various embodiments using software, hardware, firmware, or a combination thereof, such that computer system 216 operates as a special-purpose computer system that enables screening of quantum device 206 to identify characteristics 224 in a manner that enables machine learning model network 217 in computer system 216 to identify problems or conditions in which quantum device 206 may not function as desired.

[0089] In particular, machine learning model network 217 transforms computer system 216 into a special-purpose computer system compared to currently available general-purpose computer systems that do not have machine learning model network 217. In an embodiment, use of machine learning model network 217 in computer system 216 provides a practical application for screening quantum devices 206 during fabrication in a manner that improves the performance of computer system 216 in screening quantum devices 206 from images 204 of quantum devices 206. In other words, machine learning model network 217 in computer system 216 is directed to a practical application of a process that includes sending features 222 extracted from images of quantum devices to a classification neural network 302 incorporated in machine learning model network 217 in computer system 216. A set of characteristics 224 identified by classification neural network 302 for quantum device 206 is received. The set of characteristics 224 indicates whether the quantum device will function in a desired manner. Use of machine learning model network 217 in computer system 216 may improve at least one of the speed and accuracy with which computer system 216 may identify characteristics 224, execute set of actions 232, or perform other processing related to analyzing images 204 to obtain characteristics 224 that may be used to improve yield of quantum devices 206. In this manner, at least one of screening manager 212 or machine learning model network 217 in computer system 216 provides practical application of analyzing images 204 for use in screening quantum devices 206, thereby improving the operation of computer system 216.

[0090] The illustration of screening environment 200 in Figure 2 is not meant to imply physical or architectural limitations to the manner in which an illustrative embodiment may be implemented. Other components in addition to or in place of the illustrated components may be used. Some components may not be required. Also, blocks are presented to illustrate some functional components. When implemented in an illustrative embodiment, one or more of these blocks may be combined, divided into different blocks, or combined and divided.

[0091] For example, the device screening system 202 may be applied to analyze images of other types of structures other than quantum devices. For example, the device screening system 202 may be used to screen semiconductor structures, such as transistors, field effect transistors, capacitors, diodes, or other types of semiconductor structures.

[0092] As another example, the features 222 may be indirectly transmitted to the classification machine learning model system 220. For example, the features 222 may be output from the feature machine learning model system 218 to the screening manager 212. The screening manager 212 may then transmit the features 222 to the classification machine learning model system 220.

[0093] Referring now to Figure 6, an example flow diagram of a process for screening quantum devices is shown, according to an illustrative embodiment. The process of Figure 6 may be implemented in hardware, software, or both. If implemented in software, the process may take the form of program code executed by one or more processor units located in one or more hardware devices in one or more computer systems. For example, the process may be implemented in screening manager 212 in computer system 216 of Figure 2.

[0094] The process begins by sending features extracted from an image of a quantum device to a classification neural network (operation 600). In operation 600, the classification neural network is configured to identify a set of properties of the quantum device from a group of mutually exclusive properties based on the features identified in the image of the quantum device, and output this set of identified properties for the quantum device.

[0095] The process receives a set of characteristics identified by the classification neural network for the quantum device (operation 602), which indicates whether the quantum device will function in a desired manner.

[0096] The process performs a set of actions based on the set of characteristics (operation 604). The process then terminates. The set of actions may be selected from at least one of: releasing the wafer for further processing, generating a log entry, sending an alert, taking corrective action if a non-conformance exists, initiating rework of the set of wafers, or any other suitable action. As another example, the set of actions may include modifying a process for fabricating quantum devices based on the set of characteristics identified by the classification neural network.

[0097] Referring now to Figure 7, an example flow diagram of a process for screening quantum devices is shown, according to an illustrative embodiment. The process of Figure 7 may be implemented in hardware, software, or both. If implemented in software, the process may take the form of program code executed by one or more processor units located in one or more hardware devices in one or more computer systems. For example, the process may be implemented in training manager 508 of Figure 5.

[0098] The process begins by identifying training data (operation 700), which may be images of quantum devices, semiconductor devices, or other devices or structures on which training is being performed.

[0099] The process trains a classification machine learning model system using training data and a set of machine learning algorithms (operation 702). In operation 702, the training data can be sent to an already trained feature machine learning model system. The feature machine learning model system outputs features, which are sent to a classification neural network.

[0100] The process disables a set of neurons in the set of sub-networks of neurons during training of the classification neural network to reduce overfitting (operation 704). The process reduces the amplitude of this set of neurons in the set of sub-networks of neurons during training of the classification neural network to reduce overfitting (operation 706). The process then terminates.

[0101] The flow diagrams and block diagrams in the various illustrated embodiments illustrate the architecture, functionality, and processing of some possible implementations of apparatuses and methods in an example embodiment. In this regard, each block in a flow diagram or block diagram may represent at least one of a module, a segment, a function, or a portion of a process or step. For example, one or more of the blocks may be implemented as program code, hardware, or a combination of program code and hardware. If implemented in hardware, the hardware may take the form of, for example, an integrated circuit that is manufactured or configured to perform one or more processes in the flow diagram or block diagram. If implemented as a combination of program code and hardware, the implementation may take the form of firmware. Each block in a flow diagram or block diagram may be implemented using a dedicated hardware system performing various processes or various combinations of dedicated hardware and program code executed by the dedicated hardware.

[0102] In some alternative implementations of the illustrative embodiments, one or more functions noted in the blocks may occur out of the order noted in the figures. For example, in some cases, two blocks shown in succession may be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending on the functionality involved. Also, other blocks may be added in addition to the blocks noted in the flow diagrams or block diagrams.

[0103] Referring now to Figure 8, an example of a block diagram of a data processing system is shown in accordance with an illustrative embodiment. Data processing system 800 may be used to implement server computer 104, server computer 106, and client device 110 in Figure 1. Data processing system 800 may also be used to implement computer system 216 in Figure 2. In this example, data processing system 800 includes a communications framework 802. Communications framework 802 provides communications between a processor unit 804, a memory 806, a persistent storage device 808, a communications unit 810, an input / output (I / O) unit 812, and a display 814. In this example, communications framework 802 takes the form of a bus system.

[0104] Processor unit 804 is responsible for executing instructions for software that may be loaded into memory 806. Processor unit 804 includes one or more processors. For example, processor unit 804 may be selected from at least one of a multi-core processor, a central processing unit (CPU), a graphics processing unit (GPU), a physical processing unit (PPU), a digital signal processor (DSP), a network processor, or any other suitable type of processor. Furthermore, processor unit 804 may be implemented using one or more heterogeneous processor systems, in which both a primary processor and a secondary processor reside on a single chip. As another example, processor unit 804 may be a symmetric multiprocessor system that includes multiple processors of the same type on a single chip.

[0105] Memory 806 and persistent storage 808 are examples of storage device(s) 816. A storage device is any hardware capable of storing information, such as, but not limited to, data, program code in a functional form, or other suitable information, on a temporary, permanent, or temporary and permanent basis. Storage device 816, in these examples, may also be referred to as a computer-readable storage device. Memory 806, in these examples, may be, for example, a random access memory or any other suitable volatile or non-volatile storage device. Persistent storage 808 may take various forms, depending on the particular implementation.

[0106] For example, persistent storage 808 may encompass one or more components or devices. For example, persistent storage 808 may be a hard drive, a solid-state drive (SSD), a flash memory, a rewritable optical disk, a rewritable magnetic tape, or some combination thereof. The media used by persistent storage 808 may also be removable. For example, a removable hard drive may be used for persistent storage 808.

[0107] In these examples, communications unit 810 provides for communications with other data processing systems or devices, hi these examples, communications unit 810 is a network interface card.

[0108] Input / output unit 812 allows for input and output of data to and from other devices that may be connected to data processing system 800. For example, input / output unit 812 may provide a connection for user input through at least one of a keyboard, a mouse, or some other suitable input device. Further, input / output unit 812 may send output to a printer. Display 814 provides a mechanism for displaying information to a user.

[0109] Instructions for at least one of the operating system, applications, or programs may be located in storage devices 816, which are in communication with processor unit 804 through communications framework 802. The processes of the different embodiments may be executed by processor unit 804 using computer-implemented instructions, which may be located in a memory (such as memory 806).

[0110] Such instructions may be referred to as program code, computer usable program code, or computer readable program code, which may be read and executed by a processor in processor unit 804. In different embodiments, the program code may be embodied in different physical or computer readable storage media, such as memory 806 or persistent storage 808.

[0111] Program code 818 may be functionally disposed on computer readable media 820 that is selectively removable and may be loaded onto or transmitted over to data processing system 800 for execution by processor unit 804. In these illustrative examples, program code 818 and computer readable media 820 form computer program product 822. In the illustrative example, computer readable media 820 is computer readable storage medium 824.

[0112] In these examples, computer readable storage medium 824 is a physical or tangible storage device used to store program code 818, rather than a medium that propagates or transmits program code 818. As used herein, computer readable storage medium 824 should not be construed as, per se, a transitory signal such as an electric wave or other freely propagating electromagnetic wave, an electromagnetic wave propagating through a waveguide or other transmission medium (such as a light pulse passing through a fiber optic cable), or an electrical signal transmitted over a wire.

[0113] Alternatively, program code 818 may be transmitted to data processing system 800 using a computer-readable signal medium. The computer-readable signal medium may be, for example, a propagated data signal embodied with program code 818. For example, the computer-readable signal medium may be at least one of an electromagnetic signal, an optical signal, or any other suitable type of signal. Such signals may be transmitted over a connection, such as a wireless connection, an optical fiber cable, a coaxial cable, a wire, or any other suitable type of connection.

[0114] Additionally, as used herein, "computer-readable medium 820" may refer to either the singular or the plural. For example, program code 818 may be located in computer-readable medium 820 in the form of a single storage device or storage system. In another example, program code 818 may be located in computer-readable medium 820 distributed across multiple data processing systems. In other words, some of the instructions in program code 818 may be located in one data processing system, while other instructions in program code 818 may be located in another data processing system. For example, one portion of program code 818 may be located in computer-readable medium 820 in a server computer, while another portion of program code 818 may be located in computer-readable medium 820 located within a set of client computers.

[0115] The different components illustrated for data processing system 800 are not meant to provide architectural limitations to the manner in which different embodiments may be implemented. In some examples, one or more of the components may be incorporated into or otherwise form part of another component. For example, in some examples, memory 806, or portions thereof, may be incorporated within processor unit 804. Different illustrative embodiments may be implemented in a data processing system including components in addition to or in place of the components illustrated for data processing system 800. Other components illustrated in FIG. 8 may differ from the illustrated example. Different embodiments may be implemented using any hardware device or system capable of running program code 818.

[0116] An exemplary embodiment of the present disclosure may be described in conjunction with aircraft manufacturing and service method 900, as shown in Figure 9, and aircraft 1000, as shown in Figure 10. Referring initially to Figure 9, an example of an aircraft manufacturing and service method is illustrated, in accordance with an illustrative embodiment. Aircraft manufacturing and service method 900 may include specification and design 902 and material sourcing 904 of aircraft 1000, as shown in Figure 10, during pre-production.

[0117] During production, component and subassembly manufacturing 906 and system integration 908 of aircraft 1000 of Figure 10 occurs. In an example, component and subassembly manufacturing 906 may include fabrication of quantum devices. Aircraft 1000 of Figure 10 may then proceed through certification and delivery 910 and be placed into service 912. While in service 912 by a customer, aircraft 1000 of Figure 10 is scheduled for routine maintenance and service 914, which may include modification, reconfiguration, refurbishment, and other maintenance and service.

[0118] Each process of aircraft manufacturing and service method 900 may be implemented or performed by a system integrator, a third party, an operator, or any combination thereof. In the above example, the operator may be a customer. As used herein, a system integrator may include, but is not limited to, any number of aircraft manufacturers and major system subcontractors; a third party may include, but is not limited to, any number of suppliers, subcontractors, and suppliers; and an operator may be an airline, a leasing company, a military organization, a service organization, etc.

[0119] Referring now to Figure 10 , an example of an aircraft in which illustrative embodiments may be implemented is illustrated. In this example, aircraft 1000 is manufactured by aircraft manufacturing and service method 900 in Figure 9 and may include an airframe 1002 having a number of systems 1004 and an interior 1006. Example systems 1004 include one or more of a propulsion system 1008, an electrical system 1010, a hydraulic system 1012, and an environmental system 1014. Any number of other systems may be included. While an aerospace example is shown, various illustrative embodiments may be applied to other industries, such as the automotive industry.

[0120] Apparatus and methods embodied herein may be used during at least one stage of aircraft manufacturing and service method 900 of FIG.

[0121] In one illustrative example, components or subassemblies produced in component and subassembly production 906 of FIG. 9 may be produced or manufactured in a manner similar to components or subassemblies produced while aircraft 1000 is in service 912 of FIG. 9 . In yet another example, one or more apparatus embodiments, method embodiments, or a combination thereof may be utilized during production stages, such as component and subassembly production 906 of FIG. 9 and system integration 908. One or more apparatus embodiments, method embodiments, or a combination thereof may be utilized during aircraft 1000's operation 912 of FIG. 9 , during maintenance and service 914, or both. The use of several different illustrative embodiments may significantly streamline the assembly of aircraft 1000, reduce the cost of aircraft 1000, or both streamline the assembly of aircraft 1000 and reduce the cost of aircraft 1000. For example, quantum device fabrication may be performed in a manner that improves the yield of quantum devices on a wafer when using a screening system (such as that shown in one or more examples). Process modifications for rework can be implemented in a manner that salvage or reduces the waste of quantum devices formed by processing of quantum devices on a wafer.

[0122] 11 , an example of a block diagram of a product management system is shown in accordance with an illustrative embodiment. Product management system 1100 is a physical hardware system. In this example, product management system 1100 includes at least one of a manufacturing system 1102 or a maintenance system 1104.

[0123] Manufacturing system 1102 is configured to manufacture products (such as aircraft 1000 in FIG. 10 ). As shown, manufacturing system 1102 includes manufacturing equipment 1106. Manufacturing equipment 1106 includes at least one of fabrication equipment 1108 or assembly equipment 1110.

[0124] Fabrication equipment 1108 is equipment used to fabricate components for parts used to form aircraft 1000 in FIG. 10 . For example, fabrication equipment 1108 may include machines and tools. Such machines and tools may be at least one of drills, hydraulic presses, furnaces, molds, composite tape layers, vacuum systems, lathes, or other suitable types of equipment. Fabrication equipment 1108 may be used to fabricate at least one of metal parts, composite parts, semiconductors, circuits, fasteners, ribs, skin panels, spars, antennas, or other suitable types of parts.

[0125] In connection with the fabrication of semiconductor components, the fabrication equipment 1108 may include at least one of an epitaxial reactor, an oxidation system, a diffusion system, an etching machine, a cleaning machine, a bonding machine, a singulation machine, a wafer saw, an ion implanter, a physical vapor deposition system, a chemical vapor deposition system, a photolithography system, an electron beam lithography system, a plasma etcher, a die attach machine, a wire bonder, a die overcoat system, a molding machine, a hermetic sealer, an electrical tester, a burn-in oven, a holding bake oven, a UV eraser, or any other suitable type of equipment that may be used in the manufacture of quantum devices.

[0126] Assembly equipment 1110 is equipment used to assemble parts to form aircraft 1000 of FIG. 10 . In particular, assembly equipment 1110 may be used to assemble components and parts to form aircraft 1000 of FIG. 10 . Assembly equipment 1110 may also include machines and tools. Such machines and tools may be at least one of a robotic arm, a crawler, a fastener installation system, a rail-based drilling system, or a robot. Assembly equipment 1110 may be used to assemble parts (e.g., seats, horizontal stabilizers, wings, engines, engine housings, landing gear systems, and other parts for aircraft 1000 of FIG. 10 ).

[0127] In this illustrative example, maintenance system 1104 includes maintenance equipment 1112. Maintenance equipment 1112 may include any equipment necessary to perform maintenance on aircraft 1000 in FIG. 10. Maintenance equipment 1112 may include tools for performing various operations on parts of aircraft 1000 in FIG. 10. Such operations may include at least one of part disassembly, part refurbishment, part inspection, part rework, replacement part manufacturing, or other operations for performing maintenance on aircraft 1000 in FIG. 10. Such operations may be for routine maintenance, inspection, upgrade, refurbishment, or other types of maintenance operations.

[0128] In this example, maintenance equipment 1112 may include ultrasonic inspection devices, x-ray imaging systems, vision systems, drills, crawlers, and other suitable devices. In some cases, maintenance equipment 1112 may include fabrication equipment 1108, assembly equipment 1110, or both to produce and assemble parts needed for maintenance.

[0129] The product management system 1100 also includes a control system 1114. The control system 1114 is a hardware system and may also include software or other types of components. The control system 1114 is configured to control the operation of at least one of the manufacturing system 1102 or the maintenance system 1104. In particular, the control system 1114 may control the operation of at least one of the fabrication equipment 1108, the assembly equipment 1110, or the maintenance equipment 1112.

[0130] The hardware in control system 1114 may be implemented using hardware that may include computers, circuits, networks, and other types of equipment. The control may take the form of direct control of manufacturing equipment 1106. For example, control system 1114 may control robots, computer-controlled machines, and other equipment. In other examples, control system 1114 may manage the processes performed by personnel 1116 in manufacturing or performing maintenance on aircraft 1000 in FIG. 10 . For example, control system 1114 may assign tasks, provide instructions, display models, or perform other operations to manage the processes performed by personnel 1116. In such examples, device screening system 202 may be implemented in product management system 1100. For example, device screening system 202 may be part of manufacturing equipment 1106 used to screen quantum devices and determine quantum device yield. In this example, device screening system 202 may communicate with control system 1114 in a manner that enables control system 1114 to manage at least one of the manufacturing and maintenance of aircraft 1000 in FIG. 10 . For example, images may be generated after developing photoresist on the quantum devices and / or etching the quantum devices, or after other stages in the fabrication of the quantum devices. In one example, an image of the developed photoresist may enable correction of problems in lithography. Imaging after etching the quantum devices may enable screening of semiconductors for undesirable characteristics.

[0131] In various illustrative examples, work personnel 1116 may operate or interact with at least one of production equipment 1106, maintenance equipment 1112, or control system 1114. This interaction may occur to produce aircraft 1000 in Figure 10 .

[0132] Of course, product management system 1100 may be configured to manage other products besides aircraft 1000 of Figure 10. Although product management system 1100 is described in connection with aerospace manufacturing, product management system 1100 may be configured to manage products for other industries. For example, product management system 1100 may be configured to manufacture products for the automotive industry and any other suitable industry.

[0133] Thus, illustrative embodiments provide methods, apparatus, systems, and computer program products for identifying characteristics of a quantum device that may cause it to operate in an undesirable manner. For example, such characteristics may indicate conditions when a semiconductor structure does not have an intended or desired function. In one example, a computer system transmits features extracted from an image of the quantum device to a classification neural network. The classification neural network is configured to identify a set of characteristics of the quantum device from a group of mutually exclusive characteristics based on the features identified in the image of the quantum device, and output the set of characteristics identified for the quantum device. The computer system receives the set of characteristics identified by the classification neural network for the quantum device. The set of characteristics indicates whether the quantum device is functioning in a desired manner.

[0134] In one embodiment, screening quantum devices at one or more stages of their processing can identify characteristics that indicate that the quantum device will not function as desired. In an embodiment, by using a machine learning model network to screen quantum devices at various stages of processing to identify problems, screening time can be reduced. This can allow rework processes to be performed in a timely manner, which in turn can improve quantum device yield. In an embodiment, the machine learning model network can automatically perform root cause analysis on a class of characteristics that can be used to identify problems. Such problems can be the presence of particles, missing gates, missing loads, or other issues that can cause the quantum device to not function as desired. These types of problems can result in electrical defects (e.g., defects or openings) that currently are not identified until the quantum device is subjected to electrical testing.

[0135] The description of various exemplary embodiments is presented for purposes of illustration and description and is not intended to be exhaustive or to be limited to the disclosed forms of embodiments. Components that perform actions or processes are described by various examples. In an example, a component may be configured to perform the described action or process. For example, the component may have a configuration or structural design that provides the component with the ability to perform the action or process described in the example as being performed by the component. Furthermore, to the extent that the terms "includes," "has," "contains," and variations thereof are used herein, such terms, like the term "comprises," are intended to be inclusive, as open-ended terms that do not exclude any additional or other elements.

[0136] The present disclosure further includes embodiments according to the following clauses:

[0137] Clause 1. A device screening system (202), comprising: a computer system (216); a feature machine learning model system (142, 218) in a computer system (216), the feature machine learning model system (142, 218) configured to receive an image (132, 204, 502) of a quantum device (134, 206), identify features (222) of the quantum device (134, 206) in the image (132, 204, 502), and output the identified features (222) for the quantum device (134, 206) in the image (132, 204, 502) of the quantum device (134, 206); A classification neural network (144, 302) in a computer system (216) receives characteristics (222) of the quantum devices (134, 206) identified in an image (132, 204, 502) of the quantum devices (134, 206), and selects a quantum device (134, 206) from a group of mutually exclusive characteristics (226) based on the characteristics (222) identified in the image (132, 204, 502) of the quantum devices (134, 206). a classification neural network (144, 302) configured to identify a set of characteristics (140, 224) of the quantum device (134, 206) identified in the image (132, 204, 502) of the quantum device (134, 206) and output the identified set of characteristics (140, 224) for the quantum device (134, 206), wherein the set of characteristics (140, 224) indicates whether the quantum device (134, 206) is functioning in a desired manner; Device screening system (202).

[0138] Clause 2. The device screening system (202) of clause 1, wherein the classification neural network (144, 302) includes a set of subnetworks (310) of neurons (308) corresponding to a plurality of classes (228), and the sets of characteristics (140, 224) within the classes of the plurality of classes (228) are mutually exclusive.

[0139] Clause 3. The device screening system (202) of clause 2, wherein the class (228) includes at least one of exposure, feature destruction, component, etch amount, or alignment.

[0140] Clause 4. The device screening system (202) of clause 2, wherein a set of neurons (308) in a sub-network (310) of neurons (308) are disabled during training of the classification neural network (144, 302) so that overfitting is reduced.

[0141] Clause 5. A device screening system (202) as described in clause 2, wherein a set of neurons (308) in a sub-network (310) of neurons (308) have a set of amplitudes that are reduced during training of the classification neural network (144, 302) so that overfitting is reduced.

[0142] Clause 6. The device screening system (202) of any one of clauses 1 to 5, wherein the feature machine learning model system (142, 218) is a convolutional neural network (300) trained to extract features (222) of the quantum device (134, 206).

[0143] Clause 7. The convolutional neural network (300) 10. A device screening system (202) as described in clause 6, comprising multiple layers of neurons (308) that identify features (222) in an image (132, 204, 502), with subsequent layers of neurons (308) identifying larger features compared to previous layers of neurons (308), and the output of the previous layer being input to the subsequent layer.

[0144] Clause 8. The device screening system (202) of clause 7, wherein each layer of the multiple layers of neurons (308) increases the number of channels and decreases the image resolution.

[0145] Clause 9. The device screening system (202) of any one of clauses 1 to 8, wherein the quantum device (134, 206) is selected from one of a quantum dot device, a nanowire device, a quantum well device, a quantum information processing device, a quantum memory, a superconducting resonator, a Josephson junction, a nonlinear inductor, a linear inductor, a capacitor, an optical resonator, a diode, a transistor, a field effect transistor, a memory device, a quantum interference device, a topological quantum device, a waveguide, or an optical resonator.

[0146] Clause 10. A device screening system (202) described in any one of clauses 1 to 9, wherein an image (132, 204, 502) of a quantum device (134, 206) is generated during a set of steps for fabricating the quantum device (134, 206) selected from at least one of developing a resist on the quantum device, patterning a thin film on the quantum device, or etching the quantum device.

[0147] Clause 11. A device screening system (202), comprising: a computer system (216); a classification machine learning model system (220) in the computer system (216), the classification machine learning model system (220) configured to receive features (222) of the quantum devices (134, 206) identified in the images (132, 204, 502) of the quantum devices (134, 206), identify a set of characteristics (140, 224) of the quantum devices (134, 206), and output the set of characteristics (140, 224) identified for the quantum devices (134, 206) based on the features (222) identified in the images (132, 204, 502) of the quantum devices (134, 206), wherein the set of characteristics (140, 224) indicates whether the quantum devices (134, 206) will perform in a desired manner; Device screening system (202).

[0148] Clause 12. The device screening system (202) of clause 11, wherein the classification machine learning model system (220) is a classification neural network (144, 302) configured to identify a set of properties (140, 224) of the quantum device (134, 206) from a group of mutually exclusive properties (226).

[0149] Clause 13. A device screening system (202) according to clause 11 or 12, wherein in identifying a set of characteristics (140, 224) of the quantum device (134, 206), the classification machine learning model system (220) is configured to identify the set of characteristics (140, 224) of the quantum device (134, 206) from a group of mutually exclusive characteristics (226).

[0150] Clause 14. The device screening system (202) of any one of clauses 11 to 13, wherein the classification machine learning model system (220) is selected from at least one of a neural network, a decision tree, a support vector machine, a Bayesian network, a genetic algorithm, or a cluster analysis algorithm.

[0151] Clause 15. The device screening system (202) of clause 12, wherein the classification neural network (144, 302) includes a subnetwork (310) of neurons (308) corresponding to a plurality of classes (228), and the sets of characteristics (140, 224) within the classes of the plurality of classes (228) are mutually exclusive.

[0152] Clause 16. The device screening system (202) of clause 15, wherein the class (228) includes at least one of exposure, feature destruction, component, etch amount, or alignment.

[0153] Clause 17. The device screening system (202) of clause 15, wherein a set of neurons (308) in a sub-network (310) of neurons (308) are disabled during training of the classification neural network (144, 302) so that overfitting is reduced.

[0154] Clause 18. A device screening system (202) as described in clause 15, wherein a set of neurons (308) in a sub-network (310) of neurons (308) have a set of amplitudes that are reduced during training of the classification neural network (144, 302) so that overfitting is reduced.

[0155] Clause 19. A device screening system (202) described in any one of clauses 11 to 18, wherein the feature (222) is received at output from a feature extraction neural network configured to receive an image (132, 204, 502) of the quantum device (134, 206), identify features (222) of the quantum device (134, 206) in the image (132, 204, 502), and output the identified features (222) for the quantum device (134, 206) in the image (132, 204, 502) of the quantum device (134, 206).

[0156] Clause 20. A device screening system (202) according to any one of clauses 11 to 19, wherein the quantum device (134, 206) is selected from one of a quantum dot device, a nanowire device, a quantum well device, a quantum information processing device, a quantum memory, a superconducting resonator, a Josephson junction, a nonlinear inductor, a linear inductor, a capacitor, an optical resonator, a diode, a transistor, a field effect transistor, a memory device, a quantum interference device, a topological quantum device, a waveguide, or an optical resonator.

[0157] Clause 21. A method for screening a quantum device (134, 206), comprising: transmitting (600), by a computer system (216), features (222) extracted from the image of the quantum device to a classification neural network (144, 302), the classification neural network (144, 302) configured to identify a set of characteristics (140, 224) of the quantum device from a group of mutually exclusive characteristics (226) based on the features (222) identified in the image of the quantum device, and output the set of identified characteristics (140, 224) for the quantum device; receiving, by the computer system (216), a set of characteristics (140, 224) identified by the classification neural network (144, 302) for the quantum device, the set of characteristics (140, 224) indicating whether the quantum device will function as desired; method.

[0158] Article 22. 22. The method of claim 21, further comprising modifying a process for fabricating the quantum device based on the set of characteristics (140, 224) identified by the classification neural network (144, 302).

[0159] Clause 23. The method of clause 21, wherein the classification neural network (144, 302) includes a set of sub-networks (310) of neurons (308) corresponding to multiple classes (228), and the sets of characteristics (140, 224) within a class are mutually exclusive.

[0160] Clause 24. The method of clause 23, wherein the class (228) includes at least one of exposure, feature collapse, component, etch amount, or alignment.

[0161] Article 25. 24. The method of claim 23, further comprising disabling (704) a set of neurons (308) in a set of sub-networks (310) of neurons (308) during training of the classification neural network (144, 302) so that overfitting is reduced.

[0162] Article 26. 24. The method of claim 23, further comprising reducing (706) amplitudes of a set of neurons (308) in a set of sub-networks (310) of neurons (308) during training of the classification neural network (144, 302) so that overfitting is reduced.

[0163] Article 27. 22. The method of claim 21, further comprising receiving features (222), wherein the features (222) are received at output from a feature extraction neural network configured to receive an image of a quantum device and output features (222) identified for the quantum device in the image of the quantum device.

[0164] Clause 28. The method of clause 21, wherein the image of the quantum device is generated during a step of fabricating the quantum device selected from exposing a photoresist on the quantum device, patterning a thin film on the quantum device, and etching the quantum device.

[0165] Clause 29. A computer program product (822) for screening quantum devices (134, 206), comprising: a computer-readable storage medium (824); first program code stored on a computer-readable storage medium (824), the first program code executable by the computer system (216) to cause the computer system (216) to receive features (222), the features (222) received in output from a feature extraction neural network configured to receive an image of a quantum device and output features (222) identified in the image of the quantum device; second program code stored on a computer-readable storage medium (824), the second program code executable by the computer system (216) to cause the computer system (216) to transmit features (222) extracted from the image of the quantum device to a classification neural network (144, 302), the classification neural network (144, 302) configured to identify a set of characteristics (140, 224) of the quantum device (134, 206) based on the features (222) identified in the image of the quantum device; and third program code stored on a computer-readable storage medium (824), the third program code executable by the computer system (216) to cause the computer system (216) to receive a set of characteristics (140, 224) identified by the classification neural network (144, 302) for the quantum device (134, 206) based on features (222) identified in the image of the quantum device, the set of characteristics (140, 224) being indicative of whether the quantum device is functioning in a desired manner. Computer program products (822).

[0166] Numerous modifications and variations will be apparent to those skilled in the art. Moreover, various exemplary embodiments may provide features that differ from other preferred embodiments. The selected embodiment or embodiments have been chosen and described in order to best explain the principles and practical applications of the embodiments, and so that others skilled in the art, together with the disclosure of the various embodiments, may appreciate various modifications that are suitable for the particular use envisioned.

Claims

1. A device screening system (202), comprising: a computer system (216); a classification machine learning model system (220) in the computer system (216), the classification machine learning model system (220) being configured to receive features (222) of a quantum device (134, 206) identified in an image (132, 204, 502) of the quantum device (134, 206), the quantum device being a structure used to at least one of process and store quantum information; identify a set of characteristics (140, 224) of the quantum device (134, 206); and output the set of characteristics (140, 224) identified for the quantum device (134, 206) based on the features (222) identified in the image (132, 204, 502) of the quantum device (134, 206), the set of characteristics (140, 224) indicating whether the quantum device (134, 206) will perform in a desired manner; the classification machine learning model system (220) is a classification neural network (144, 302) configured to identify the set of properties (140, 224) of the quantum device (134, 206) from a group of mutually exclusive properties (226); the classification neural network (144, 302) includes a plurality of sub-networks (310) of neurons (308) corresponding to a plurality of classes (228), and the sets of characteristics (140, 224) within a class of the plurality of classes (228) are mutually exclusive; Device screening system (202).

2. 2. The device screening system (202) of claim 1, wherein in identifying a set of characteristics (140, 224) of the quantum device (134, 206), the classification machine learning model system (220) is configured to identify the set of characteristics (140, 224) of the quantum device (134, 206) from a group of mutually exclusive characteristics (226).

3. 3. The device screening system of claim 1, wherein the classification machine learning model system is selected from at least one of a neural network, a decision tree, a support vector machine, a Bayesian network, a genetic algorithm, or a cluster analysis algorithm.

4. The device screening system of claim 1 , wherein the class includes at least one of exposure, feature destruction, component, etch amount, or alignment.

5. 2. The device screening system of claim 1, wherein a set of neurons in a sub-network of neurons is disabled during training of the classification neural network to reduce overfitting.

6. 2. The device screening system of claim 1, wherein a set of neurons in a sub-network of neurons has a set of amplitudes that are reduced during training of the classification neural network to reduce overfitting.

7. 7. The device screening system of claim 1, wherein the features are received at output from a feature extraction neural network, the feature extraction neural network being configured to receive the image of the quantum device, identify the features of the quantum device in the image, and output the identified features for the quantum device in the image.

8. 8. The device screening system (202) of any one of claims 1 to 7, wherein the quantum device (134, 206) is selected from one of a quantum dot device, a nanowire device, a quantum well device, a quantum information processing device, a quantum memory, a superconducting resonator, a Josephson junction, a nonlinear inductor, a linear inductor, a capacitor, an optical resonator, a diode, a transistor, a field effect transistor, a memory device, a quantum interference device, a topological quantum device, a waveguide, or an optical resonator.

9. A method for screening a quantum device (134, 206), which is a structure used to process and / or store quantum information, comprising: transmitting (600), by a computer system (216), features (222) extracted from an image of a quantum device to a classification neural network (144, 302), the classification neural network (144, 302) configured to identify a set of characteristics (140, 224) of the quantum device from a group of mutually exclusive characteristics (226) based on the features (222) identified in the image of the quantum device, and output the set of identified characteristics (140, 224) for the quantum device; receiving, by the computer system, the set of characteristics identified by the classification neural network for the quantum device, the set of characteristics indicating whether the quantum device is functioning as desired; the classification neural network (144, 302) includes a set of sub-networks (310) of neurons (308) corresponding to a plurality of classes (228), and the sets of characteristics (140, 224) within a class are mutually exclusive; method.

10. 10. The method of claim 9, further comprising modifying a process for fabricating the quantum device based on the set of characteristics identified by the classification neural network.

11. The method of claim 9, wherein the class (228) includes at least one of exposure, feature destruction, component, etch amount, or alignment.

12. 10. The method of claim 9, further comprising disabling (704) a set of neurons (308) in the set of sub-networks (310) of neurons (308) during training of the classification neural network (144, 302) so that overfitting is reduced.

13. 10. The method of claim 9, further comprising reducing (706) amplitudes of a set of neurons (308) in a set of sub-networks (310) of neurons (308) during training of the classification neural network (144, 302) so as to reduce overfitting.

14. 10. The method of claim 9, further comprising receiving the features (222) received at output from a feature extraction neural network, the feature extraction neural network configured to receive the image of the quantum device and output the features (222) identified for the quantum device in the image of the quantum device.

15. 10. The method of claim 9, wherein the image of the quantum device is generated during a step of fabricating the quantum device selected from exposing a photoresist on the quantum device, patterning a thin film on the quantum device, and etching the quantum device.

16. A computer-readable storage medium (824) having stored thereon program code executable by a computer system (216) to cause the computer system (216) to perform the method of any one of claims 9 to 15.

Images