Abstract: An embodiment includes training a classification model by sending data from a seed dataset to a model training interface of the classification model where the seed dataset comprises a classification and determining a first benchmark metric. The embodiment includes iteratively executing the classification model on a production dataset to determine a set of samples predicted to comprise the classification. The embodiment includes generating an annotation of the set of samples based on clustering. The embodiment includes issuing an instruction via a fine-tuning interface of the classification model to modify a configuration parameter of the classification model based on the annotation to cause the classification model to determine a second benchmark metric and establishing a checkpoint wherein the checkpoint is incremented for each iteration of the production dataset. The embodiment also includes executing an instruction whether to distill an update of the classification model.

Abstract: According to embodiments of the present disclosure, processor chips adapted for efficient massively-concurrent conditional computation are provided. In various embodiments, a chip comprises at least one processing core; a controller operatively coupled to the at least one processing core; and an instruction memory in communication with the controller. The controller is configured to: concurrently compute a plurality of relational operators on a plurality of inputs, resulting in a plurality of results; combine the plurality of results to determine an index; select an operation based on the index; and cause the at least one processing core to execute the selected operation.

Abstract: Completion of data gaps in a dataset is provided and includes retrieving first input data that includes one or more metrics associated with one or more user devices. One or more rules are applied to the first input data and the first input data is classified into a set of groups based on the application of the one or more rules. A presence of a first multiplicative relationship between a first metric of the one or more metrics and a second metric of the one or more metrics is determined and the second metric is modified based on the determination of the presence of the first multiplicative relationship. Based on the modified second metric, a first portion of a training dataset is generated and outputted.

Abstract: Methods, systems, and computer program products for self-supervision in table question answering are provided herein. A computer-implemented method includes obtaining a table comprising a plurality of entries, wherein each entry corresponds to a particular column and particular row of the table; identifying one or more of the entries in the table that correspond to a target answer of a natural language query; generating an intermediate representation of the table comprising the rows corresponding to the identified one or more entries, wherein the intermediate representation masks each of the identified one or more entries; and generating a set of natural language question and answer pairs based on the intermediate representation.

Abstract: A method of cache management for a code load is provided. In some embodiments, a method includes receiving a request to load code at a cluster complex. The cluster complex comprises a first node and a second node. The first node includes a first cache and the second node having a second cache. The method comprises, in response to the request, destaging the first cache of the first node and directing input/output requests to the second node. The method comprises, in response to the first cache being destaged, sending the request to the first node. The method comprises executing the request at the first node. The method comprises resuming input/output requests to the first node.

Abstract: A computer-implemented method, a computer system and a computer program product predict and project vehicle movements in an augmented reality (AR) environment. The method includes obtaining telemetry data from a vehicle in an environment, where the telemetry data is selected from a group consisting of: video data, audio data and text data. The method also includes determining a path for the vehicle based on the telemetry data. The method further includes identifying an augmented reality device in the path, where the augmented reality device is associated with a user proximate to the path. Lastly, the method includes displaying the path using the augmented reality device in the augmented reality (AR) environment.

Abstract: A computer-implemented method for training a lattice layer in a Deep Lattice Network includes preparing parameters of vertices, each of the parameters corresponding to each vertex of a subdivided unit hypercube defined by subdividing an S-dimensional unit hypercube by a predetermined number k with kS vertices and defining each parameter by identifying one vertex in a specific order, identifying a first set of vertices that appear before the identified vertex in the specific order, identifying a second set of vertices that appear before the identified vertex in the specific order, defining a lower bound as a maximum value among values of vertices in the first set of vertices, defining an upper bound as a minimum value among values of vertices in the second set of vertices, and defining the parameter of the identified vertex based on the lower bound, the upper bound, and a parameter corresponding to the identified vertex.

Abstract: The classifier of an artificial intelligence model is trained by mapping output values in a final run of a training phase, measuring dense data portion and sparse data portion of a data plot produced by the training phase by identifying outputs near edges of the data plot and extracting the curve patterns as linear functions in the dense areas of the output; obtaining a fractal function base set of patterns from the linear functions provided by the dense areas of the output; applying the fractal function base set of patterns from the dense areas of the output to the sparse data portion of the data plot; and training the artificial intelligence model using the data plot including the dense areas of the output and the sparse data portion of the data plot that has been fit to the data curve using the fractal function base set of patterns.

Abstract: A method of forming a nanosheet transistor device is provided. The method includes forming a segment stack of alternating intermediate sacrificial segments and nanosheet segments on a bottom sacrificial segment, wherein the segment stack is on a mesa and a nanosheet template in on the segment stack. The method further includes removing the bottom sacrificial layer to form a conduit, and forming a fill layer in the conduit and encapsulating at least a portion of the segment stack.

Abstract: A gate contact is formed within a gate cut region of a semiconductor structure to facilitate electrical routing. The gate contact includes a bottom portion extending within the gate cut region and adjoining a vertical end portion of a metal gate. A metal layer on the front side of the semiconductor structure includes signal tracks, one or more of which is vertically above the gate cut region. A signal track in the metal layer may be electrically connected to the gate contact. Selected source/drain regions within the semiconductor structure may be electrically connected to a back side power delivery network.

Abstract: A semiconductor device includes a first source/drain region, a first contact over the first source/drain region, a second source/drain region, and a lateral contact connecting the second source/drain region to a back end of line (BEOL). Portions of the first contact are recessed, and the lateral contact overlaps with the recessed portions of the first contact. The first source/drain region is formed over the second source/drain region.

Abstract: A heat sink filler to maintain proper cooling air flow through operating processing units is disclosed. The heat sink filler comprises a frame with a removable perforated sheet inside the frame, wherein the perforated sheet is configured to match an impedance of a heat sink. The heat sink filler further includes a first keying feature. The heat sink filler also includes an extended edge on a first edge of the frame and a beveled edge on the second edge of the frame.

Abstract: A semiconductor structure includes a selection transistor having a first drain-source terminal, a second drain-source terminal, at least one channel region between the first and second source-drain terminals, and a gate adjacent the at least one channel region. A first vertical contact structure is connected to the first drain-source terminal. A second vertical contact structure is connected to the second drain-source terminal. A first resistive memory cell has a lower electrode coupled to the second vertical contact structure and has an upper electrode. A third vertical contact structure is connected to the upper electrode of the first resistive memory cell. A second resistive memory cell has an upper electrode coupled to the second vertical contact structure and has a lower electrode. A fourth vertical contact structure is connected to the lower electrode of the second resistive memory cell.

Abstract: A computer implemented method optimizes source code. A processor set receives a source code in a first version. The processor set compiles the source code in the first version to generate an intermediate representation for the source codes in the first version. The processor set compares the intermediate representation for the source code in the first version to a number of optimization paths. The processor set identifies a number of matches based on the comparisons. The processor set updates at least a portion of the source code in the first version to the second version based on the number of matches to generate the source code in the second version.

Abstract: Techniques regarding the synthesis of one or more polymers through one or more ring-opening polymerizations conducted within a flow reactor and facilitated by one or more anionic catalysts are provided. For example, one or more embodiments can comprise a method, which can comprise polymerizing, via a ring-opening polymerization within a flow reactor, a cyclic monomer in the presence of one or more anionic organocatalysts.

Abstract: A memory structure that includes a dielectric stack of a ferroelectric dielectric layer and a paraelectric dielectric layer. At least the ferroelectric dielectric layer produces a negative capacitance to amplify an applied voltage. A thickness of the ferroelectric dielectric layer and the paraelectric dielectric layer results in simultaneous breakdown of a dielectric material in each of the ferroelectric dielectric layer and the paraelectric dielectric layer for the formation of conductive filaments upon being exposed to an electric field produced by the applied voltage amplified by the negative capacitance. The memory structure also includes a first electrode at a first end of the dielectric stack, and a second electrode at a second end of the dielectric stack. The applied voltage is applied to the memory structure through at least one of the first electrode and the second electrode.

Abstract: A present invention embodiment analyzes user input via natural language processing. A natural language utterance from a user is analyzed to determine one or more computing tasks. the natural language utterance is analyzed using a knowledge base to identify one or more modifications to the natural language utterance that are based on previous user modifications to a previous user utterance. An indication that the user accepted at least one modification of the one or more modifications is received, wherein the at least one modification modifies the one or more computing tasks. The modified one or more computing tasks are executed.

Abstract: Embodiments herein describe circuitry for mimicking SRAM cells. Instead of mimicking (or modeling) the entire SRAM cell, the mimic SRAM cells can mimic only the read path, which may include a pFET and two nFETs. In one embodiment, the mimic SRAM circuitry includes multiple selectable banks of mimic cells. During testing, the number of banks can be selected that best represent the worst case scenario—e.g., the PVT corner with the worst process variation. In addition to selecting the number of banks of mimic cells, the gate voltage of the transistors in the mimic cells can be adjusted.

Abstract: Systems and methods for operating a quantum system are described. A controller of a quantum system can generate a command signal. The quantum system can include quantum hardware having a plurality of qubits. An interface of the quantum system can control the quantum hardware based on the command signal received from the controller to determine a plurality of moments of a matrix using a random state vector represented by the plurality of qubits. The controller can be further configured to output the plurality of moments of the matrix to a computing device to estimate a trace of a matrix function based on one or more selected moments among the plurality of moments. The matrix function can be a function of the matrix.

Abstract: A computer-implemented method, according to one approach, includes: generating a multi-factor authentication challenge for a user in response to receiving a request from the user, The multi-factor authentication challenge includes a dynamic combination of available authentication factors corresponding to the user. The dynamic combination of authentication factors includes (i) at least one user defined authentication factor, (ii) at least one biometric authentication factor, and (iii) at least one activity-based authentication factor. Inputs are also received for the multi-factor authentication challenge, and a determination is made as to whether the inputs received satisfy the multi-factor authentication challenge.

Abstract: Computer-based provision of recommendations includes determining an item of interest to a user and a geographic region of the user. The item, as available from different e-commerce websites (websites) of geographic regions that are different from the geographic region of the user are aggregated within blockchain. Social network data for the user is retrieved and processed using natural language processing to determine contacts of the user located in the geographic regions of the websites. A likelihood of each contact traveling to the geographic region of the user is determined based on the social networking data. A website is selected that offers the item and that corresponds to a geographic region of a selected contact. The selected contact of the user has at least a minimum likelihood of traveling to the geographic region of the user. The selected website is presented to the user as an option for obtaining the item.

Abstract: A network communication device can receive a private session key from a data processing system. A first work queue element can be received in a send queue of the network communication device. The first work queue element can indicate outbound session data to be communicated to a client device. Responsive to receiving the first work queue element, the network communication device can generate encrypted outbound session data by encrypting the outbound session data using the private session key. The network communication device can communicate, via remote directory memory access (RDMA) over a secured communication tunnel, the encrypted outbound session data to the client device.

Abstract: An embodiment configures a software robot to operate collaboratively with a plurality of platforms within a work environment. The embodiment decomposes, via the software robot, a request to perform a task received from one of the platforms, into at least one integration-action pair. An integration in this pair represents a configuration to operate on an execution platform within the plurality of platforms where a function is to be executed; an action in the pair represents the function to be performed on the execution platform. Responsive to the determination that the at least one integration-action pair does not exist in a database of integration-action pairs, the embodiment trains the software robot to perform the function on the platform.

Abstract: Processing related messages in different queues in a middleware system is provided. The middleware system receives a first message that comprises a body and a header. The header includes a first message ID and a relationship key, wherein the relationship key denotes a relationship between the first message and a correlated second message. The system reads the relationship key from the header of the first message. The middleware system stores the first message in a first queue and stores the first message ID and relationship key in a relationship object in a cache list. A consumer API retrieves the first message from the first queue and extracts the relationship key from the header. The consumer API identifies, according to the relationship key, a second message ID in the relationship object and retrieves the correlated second message from a second queue according to the second message ID.

Abstract: An embodiment performs, by an attribute value substitution engine, key-value extraction on infrastructure as code. The embodiment assigns, by the attribute value substitution engine, an attribute in a plurality of attributes to a value in the extracted key-value. The embodiment computes, by the attribute value substitution engine, a plurality of similarities between the plurality of attributes and a cloud resource. The embodiment groups, based on the plurality of similarities, the plurality of attributes into a group of attributes. The embodiment substitutes, by the attribute value substitution engine, the attribute with another attribute, the attribute and the another attribute belonging to the group of attributes.

Abstract: An example operation includes one or more of receiving a natural language input via a software application, executing a machine learning model on the natural language input to determine a plurality of sub-queries corresponding to the natural language input, searching a tree structure for field values that correspond to the plurality of sub-queries, relationships between the field values, and computations to be performed, generating a structured query language (SQL) command based on the field values, relationships between the field values, and the computations to be performed, and executing the SQL command on a database to generate query results and return the query results to the software application.

Abstract: One or more computer processors create a fully convolution network (FCN) comprising a plurality of 1×1 convolutions. The one or more computer processors append linear mapping layer (LM) to created FCN. The one or more computer processors capture a plurality of features utilizing multi-scale dilated convolutional kernels from the linear mapped FCN (LM-FCN). The one or more computer processors apply an average pool layer to the captured plurality of features along a temporal axis of a dilated convolutional kernel within the LM-FCN. The one or more computer processors predict a classification for subsequent time-series data utilizing the pooled plurality of features.

Abstract: A set of input features, each feature having a value, can be processed to determine pairwise correlations between features of the set. The features can be arranged into groups based on correlations with one another. Each feature can also be analyzed to determine a predictive value. A representative feature of each group can be selected based on the predictive value.

Abstract: A semiconductor structure is provided that includes a self-aligned offset frontside gate contact structure and a direct backside source/drain contact structure. The presence of the off-centered frontside gate contact structure is attractive since it mitigates the risk of gate contact-to-source/drain contact shorts and it also improves the metal line, M1, spacing within the overlying frontside back-end-of-the-line (BEOL) structure.

Abstract: A semiconductor device includes a first-type transistor, a second-type transistor, and a t-shaped backbone. The first-type transistor includes a first-type source/drain and the second-type transistor includes a second-type source/drain. The t-shaped backbone includes a wall and a sub-wall. The wall separates the first-type transistor and the second-type source/drain. The sub-wall extends from the wall and into the second-type source/drain.

Abstract: A semiconductor structure including a first stacked transistor structure adjacent to a second stacked transistor structure, and a first conductive structure in direct contact with and electrically connecting a bottom gate conductor of the first stacked transistor structure and a top gate conductor of the second stacked transistor structure.

Abstract: Techniques for sidewall passivation and removal of redeposited materials and processing damage from phase change memory materials are provided. In one aspect, a phase change memory device includes: one or more phase change memory cells, where each of the phase change memory cells includes a phase change material between a bottom electrode and a top electrode; and a carbon and oxygen-containing passivation layer on sidewalls of the phase change material. An ovonic threshold switch can also be present between the bottom and top electrodes, in series with the phase change material, and the carbon and oxygen-containing passivation layer can also be present on sidewalls of the ovonic threshold switch. A method of fabricating the present phase change memory devices is also provided.

Abstract: A method, system, and computer program product are configured to: determine a location in an industrial system and an amount of mass based on predicting, using a digital twin of the industrial system, that adding the amount of mass to the location will reduce a vibration in the industrial system; and control a robot to position a movable object at the location, the movable object having a mass equal to the amount of mass.

Abstract: A method comprises receiving a workload for a computer system; sweeping at least one parameter of the computer system while executing the workload; monitoring one or more characteristics of the computer system while sweeping the at least one parameter, the one or more characteristics including total power consumption of the computer system; generating a power profile for the workload that indicates a respective selected value for the at least one parameter based on analysis of the monitored total power consumption of the computer system while sweeping the at least one parameter; and executing the workload based on the respective selected value of the at least one parameter.

Abstract: Described are techniques for automated software function code generation. The techniques include generating a list of algorithms configurable with a requested software code function provided by a client device, where the list of algorithms is based on evaluating a plurality of software policies. The techniques further include generating the requested code function based on a selected algorithm of the list of algorithms and using a repository of reusable code samples in a plurality of programming languages. The techniques further include transmitting the requested code function to the client device.

Abstract: An example operation may include one or more of storing a hierarchical data set, receiving a plurality of predicted outputs from a plurality of nodes in a distributed computing environment, respectively, wherein each predicted output is generated by a different node via execution of a time-series forecasting model on a different subset of lowest level data in the hierarchical data set, combining the plurality of predicted outputs via bottom-up aggregation to generate one or more additional predicted outputs for the time-series forecasting model based on one or more levels above the lowest level in the hierarchical time-series data set, determining error values for the time-series forecasting model at each level of the hierarchical data set based on the received and the one or more additional generated predicted outputs, and modifying a parameter of the time-series forecasting model based on the determined error values.

Abstract: A fully-aligned via interconnect structure is provided in which a first etch stop layer is formed on a first interconnect dielectric material layer containing an electrically conductive line structure to protect the interconnect dielectric material from eroding during metallization used in providing a combined via/line electrically conductive structure in a second interconnect dielectric material layer that is formed above the first interconnect dielectric material layer. The interconnect structure has low resistance due to the maximized contact between the via portion of combined via/line electrically conductive structure and the underlying electrically conductive line structure. Moreover, no bowing or metal fangs are formed, and no metal residue is introduced into the first interconnect dielectric material layer during metallization.

Abstract: A computer implemented method for authentication using a stylus. The stylus generates a stylus token using a key stored in the stylus and a current timestamp, wherein the stylus token is valid for a selected period of time. The stylus transmits the stylus token to a computer system. The stylus repeats generating the stylus token using the key stored in the stylus and the current timestamp and transmitting the stylus token to the computer system in response to the stylus token expiring after the selected period of time.

Abstract: Provided are a computer program product, system, and method for generating instrumentation for data integrity of function calls. Instrumentation is generated to determine whether argument values passed to functions have changed since written to memory locations during execution of the computer program. A reference monitor that runs during execution of the computer program compares an argument value, for a function, written to a memory location to an argument value passed to the function to determine whether the argument value has changed. The reference monitor permits execution of the function with the passed argument value in response to determining that the argument value has not changed.

Abstract: Controlling compute process access to shared data and compute resources includes, responsive to a compute process determining that access to at least one of shared resources and shared data is necessary to perform a compute task, creating, by the compute process, a ticket file belonging to the compute process in a ticket queue directory. The compute process is allowed to proceed performing the compute task upon determining that the ticket file is first in line in a ticket queue of the ticket queue directory, according to a ticket ordering algorithm independently applied by the compute process. Subsequent to completing the compute task, the compute process removes the ticket from the ticket queue directory.

Abstract: A processor set may generate a set of selection predicates based on one or more alerting rules for security risk detection in remote data from a remote system. The processor set may determine one or more data metrics based on sample data from the remote system. The processor set may remove one or more selection predicates, from the set of selection predicates, to generate a one or more filter predicates. The processor set may transmit, to the remote system, a request that is based on the one or more selection predicates and the one or more filter predicates. The processor set may receive, from the remote system, filtered data based on the request. The processor set may store the second data in a local system.

Abstract: Techniques are provided for causal learning and prediction. In one embodiment, the techniques involve determining a first attribute value based on an output of a generative model trained on a first data set, generating a second attribute value based on a change of a latent space dimension of the generative model, generating a sensitivity map based on a difference between the first attribute value and the second attribute value; and generating a causal graph based on the sensitivity map.

Abstract: One or more systems, devices, computer program products and/or computer-implemented methods of use provided herein relate to automatic sentence condition matching using natural language processing (NLP). The computer-implemented system can comprise a memory that can store computer-executable components and a processor that can execute the computer-executable components, wherein the computer-executable components can comprise an extraction module that can use a probabilistic relevance weighting model to retrieve a first sentence from a document by computing a normalized relevance score of the first sentence based on a relevance weighting score of a second sentence from a dictionary of query sentences. The computer-executable components can further comprise a resolution module that can use a set of NLP rules and a linguistic dictionary to automatically identify whether the first sentence and the second sentence have a same meaning based on the normalized relevance score being above a defined threshold.

Abstract: Disclosed herein are methods, systems, and computer program products for selecting an artificial intelligence (AI) model. Aspects include receiving, by a multi-armed bandit agent, candidate entities from multiple set expansion models and selecting a subset of the candidate entities for a first set expansion task, wherein a first candidate entity of the selected subset of candidate entities is selected from a first model by a subject matter expert. Aspects also include selecting, by the multi-armed bandit agent, based on the first candidate entity selected by the subject matter expert, the first model from the set expansion models for generating further candidate entities.

Abstract: Embodiments of present invention provide a semiconductor structure, which includes a first row of transistor array; and a second row of transistor array, the second row of transistor array being adjacent to and parallel to the first row of transistor array, where gates of the first row of transistor array are not aligned with gates of the second row of transistor array. A method of forming the same is also provided.

Abstract: Embodiments of the present invention are directed to processing methods and resulting structures for integrated circuits having backside programmable gate arrays. In a non-limiting embodiment, a front end of line structure having an array of transistors is formed such that each transistor of the array of transistors includes one or more placeholder backside vias. A first portion of the backside vias defines one or more placeholder backside vias and a second portion of the one or more backside vias defines one or more programmed backside vias. A back end of line structure is formed on a first surface of the front end of line structure. A backside structure is formed on a second surface of the front end of line structure opposite the first surface. The backside structure includes a backside metallization layer in direct contact with the one or more programmed backside vias.

Abstract: Disclosed embodiments provide methods, systems, and computer program products for implementing intelligent power shifting in an uninterruptable power supply (UPS) system for enhanced system efficiency. Disclosed embodiments implement intelligent power shifting to decrease UPS power consumption during normal system operation in the presence of a main power signal, increasing system efficiency, while providing redundancy in the absence or a disruption of the main power signal. Power shifting is based on an intelligent algorithm for controlling current through power supply units (PSUs) of the UPS system to shift a substantial portion of the main power signal through PSUs that are connected to power distribution units (PDU) not powered through the UPS, reducing power consumption by the UPS.

Abstract: According to the embodiment of the present invention, a semiconductor device includes a first nanodevice comprised of a plurality of first transistors and a second nanodevice comprised of a plurality of second transistors. The second nanodevice is located adjacent to and parallel to the first nanodevice along an x-axis. A first backside signal line and a second backside signal line are located at a cell boundary of the first nanodevice. A first gap exists between the first backside signal line and the second backside signal line.

Abstract: Embodiments of the present disclosure are directed to co-integrating multiple fin heights with bottom dielectric isolations for effective width tuning. In a non-limiting embodiment, a semiconductor structure includes a first shallow trench isolation (STI) region formed over a first portion of a substrate and a second STI region formed over a second portion of the substrate. A topmost surface of the second STI region is coplanar to a topmost surface of the first STI region. A first fin is formed in the first STI region and a second fin is formed in the second STI region. A topmost surface of the second fin is coplanar to a topmost surface of the first fin. A bottom dielectric isolation (BDI) region is positioned between the first fin and the substrate, thereby providing a first effective width and a second effective width for the first fin and second fin, respectively.

Abstract: A semiconductor device includes at least a direct backside contact between a source line and a source epitaxial growth and/or a drain line and a drain epitaxial growth. A clock signal line contact via can connect a gate to a backside clock signal line. The clock signal line contact via is surrounded by a deep STI fill to prevent shorting between the clock signal line and the source and/or drain lines in the backside power rail.