Abstract: Techniques that facilitate model support in deep learning are provided. In one example, a system includes a graphics processing unit and a central processing unit memory. The graphics processing unit processes data to train a deep neural network. The central processing unit memory stores a portion of the data to train the deep neural network. The graphics processing unit provides, during a forward pass process of the deep neural network that traverses through a set of layers for the deep neural network from a first layer of the set of layers to a last layer of the set of layers that provides a set of outputs for the deep neural network, input data for a layer from the set of layers for the deep neural network to the central processing unit memory.

Abstract: Techniques that facilitate model support in deep learning are provided. In one example, a system includes a graphics processing unit and a central processing unit memory. The graphics processing unit processes data to train a deep neural network. The central processing unit memory stores a portion of the data to train the deep neural network. The graphics processing unit provides, during a forward pass process of the deep neural network that traverses through a set of layers for the deep neural network from a first layer of the set of layers to a last layer of the set of layers that provides a set of outputs for the deep neural network, input data for a layer from the set of layers for the deep neural network to the central processing unit memory.

Abstract: Techniques that facilitate model support in deep learning are provided. In one example, a system includes a graphics processing unit and a central processing unit memory. The graphics processing unit processes data to train a deep neural network. The central processing unit memory stores a portion of the data to train the deep neural network. The graphics processing unit provides, during a forward pass process of the deep neural network that traverses through a set of layers for the deep neural network from a first layer of the set of layers to a last layer of the set of layers that provides a set of outputs for the deep neural network, input data for a layer from the set of layers for the deep neural network to the central processing unit memory.

Abstract: Processing a neural network data flow graph having a set of nodes and a set of edges. An insertion point is determined for a memory reduction or memory restoration operation. The determination is based on computing tensor timing slacks (TTS) for a set of input tensors; compiling a candidate list (SI) of input tensors, from the set of input tensors, using input tensors having corresponding TTS values larger than a threshold value (thTTS); filtering the SI to retain input tensors whose size meets a threshold value (thS); and determining an insertion point for the operation using the SI based on the filtering. A new data flow graph is generated or an existing one is modified using this process.

Abstract: A computer-implemented method includes identifying a noise cluster, representing the noise cluster according to a variational model, projecting the variational model onto one or more corners to yield a projected noise cluster, and determining a computed noise for the projected noise cluster. Optionally, the noise cluster includes one or more noise cluster elements, and each of the noise cluster elements are expressed as one or more circuit element terms, according to a canonical form. Optionally, at least one of the corners is a bounding corner. For the bounding corner, the projected noise cluster is generated by maximizing the circuit element terms for those noise cluster elements that tend to increase noise, and by minimizing the circuit element terms for those noise cluster elements that tend to decrease noise, whereby noise is maximized for the canonical form. A corresponding computer program product and computer system are also disclosed.

Abstract: A first plurality of relational tables is obtained from a relational database. Each table of the first plurality of relational tables stores connectivity information for a graph that comprises a plurality of nodes and a plurality of edges connecting the nodes, and each of the nodes is assigned an initial identifier. The nodes are clustered into a plurality of clusters. Each cluster contains a subset of the nodes, and all nodes in each subset are close to each other according to a metric. Each node is assigned a new identifier. The new identifier comprises a concatenation of an identifier associated with the cluster to which the node belongs and an identifier associated with the node. A second plurality of relational tables is constructed and stores connectivity information for the graph. The node is identified in the second plurality of relational tables by the new identifier.

Abstract: Processing a neural network data flow graph having a set of nodes and a set of edges. An insertion point is determined for a memory reduction or memory restoration operation. The determination is based on computing tensor timing slacks (TTS) for a set of input tensors; compiling a candidate list (SI) of input tensors, from the set of input tensors, using input tensors having corresponding TTS values larger than a threshold value (thTTS); filtering the SI to retain input tensors whose size meets a threshold value (thS); and determining an insertion point for the operation using the SI based on the filtering. A new data flow graph is generated or an existing one is modified using this process.

Abstract: Systems and methods for improving timing closure of new and existing IC chips by breaking at least one parameter of interest into two or more partial parameters. More specifically, a method is provided for that includes propagating at least one timing analysis run for a semiconductor product. The method further includes identifying at least one parameter of interest used in the at least one timing analysis run. The method further includes splitting the at least one parameter into two parts comprising a controlled part and an uncontrolled part. The method further includes correlating or anti-correlating the controlled part with another parameter used in the at least one timing analysis run. The method further includes projecting timing using the correlation or anti-correlation between the controlled part and the another parameter and using the uncontrolled part of the at least one parameter.

Abstract: Techniques that facilitate model support in deep learning are provided. In one example, a system includes a graphics processing unit and a central processing unit memory. The graphics processing unit processes data to train a deep neural network. The central processing unit memory stores a portion of the data to train the deep neural network. The graphics processing unit provides, during a forward pass process of the deep neural network that traverses through a set of layers for the deep neural network from a first layer of the set of layers to a last layer of the set of layers that provides a set of outputs for the deep neural network, input data for a layer from the set of layers for the deep neural network to the central processing unit memory.

Abstract: Examples of techniques for statistical static timing analysis of an integrated circuit are disclosed. In one example according to aspects of the present disclosure, a computer-implemented method is provided. The method comprises performing an initial statistical static timing analysis of the integrated circuit to create a parameterized model of the integrated circuit for a plurality of paths using a plurality of timing corners to calculate a timing value for each of the plurality of paths, each of the plurality of timing corners representing a set of timing performance parameters. The method further comprises determining at least one worst timing corner from the parameterized model for each of the plurality of paths based on the initial statistical static timing analysis and calculated timing value for each of the plurality of paths. The method also comprises performing a subsequent analysis of the integrated circuit using the at least one worst timing corner.

Abstract: Solutions for integrating manufacturing feedback into an integrated circuit design are disclosed. In one embodiment, a computer-implemented method is disclosed including: defining an acceptable yield requirement for a first integrated circuit product; obtaining manufacturing data about the first integrated circuit product; performing a regression analysis on data representing paths in the first integrated circuit product to define a plurality of parameter settings based upon the acceptable yield requirement and the manufacturing data; determining a projection corner associated with the parameter settings for satisfying the acceptable yield requirement; and modifying a design of a second integrated circuit product based upon the projection corner and the plurality of parameter settings.

Abstract: A computer-implemented method includes identifying a noise cluster, representing the noise cluster according to a variational model, projecting the variational model onto one or more corners to yield a projected noise cluster, and determining a computed noise for the projected noise cluster. Optionally, the noise cluster includes one or more noise cluster elements, and each of the noise cluster elements are expressed as one or more circuit element terms, according to a canonical form. Optionally, at least one of the corners is a bounding corner. For the bounding corner, the projected noise cluster is generated by maximizing the circuit element terms for those noise cluster elements that tend to increase noise, and by minimizing the circuit element terms for those noise cluster elements that tend to decrease noise, whereby noise is maximized for the canonical form. A corresponding computer program product and computer system are also disclosed.

Abstract: Systems and methods for improving timing closure of new and existing IC chips by breaking at least one parameter of interest into two or more partial parameters. More specifically, a method is provided for that includes propagating at least one timing analysis run for a semiconductor product. The method further includes identifying at least one parameter of interest used in the at least one timing analysis run. The method further includes splitting the at least one parameter into two parts comprising a controlled part and an uncontrolled part. The method further includes correlating or anti-correlating the controlled part with another parameter used in the at least one timing analysis run. The method further includes projecting timing using the correlation or anti-correlation between the controlled part and the another parameter and using the uncontrolled part of the at least one parameter.

Abstract: A system for semiconductor chip fabrication. A host system hosts a capacitance adjust tool for performing calculating a ground capacitance adjust for a wire segment going through routing tiles according to the following operations; providing the routing tiles having a plurality of wires wherein the wire segment being a victim wire and neighboring wires being aggressor wires; computing ground capacitance adjusts for a victim wire averaged across all aggressor slew values and across possible spacing values between the victim wire and the neighboring aggressor wires to take into account a potential coupling effect by the neighboring aggressor wires, to guide placement of the wire segment in the routing tiles to avoid coupling noise; and outputting the placement of the wire segments to a tool for manufacturing the semiconductor chip.

Abstract: A computer-implemented method includes identifying a noise cluster, representing the noise cluster according to a variational model, projecting the variational model onto one or more corners to yield a projected noise cluster, and determining a computed noise for the projected noise cluster. Optionally, the noise cluster includes one or more noise cluster elements, and each of the noise cluster elements are expressed as one or more circuit element terms, according to a canonical form. Optionally, at least one of the corners is a bounding corner. For the bounding corner, the projected noise cluster is generated by maximizing the circuit element terms for those noise cluster elements that tend to increase noise, and by minimizing the circuit element terms for those noise cluster elements that tend to decrease noise, whereby noise is maximized for the canonical form. A corresponding computer program product and computer system are also disclosed.

Abstract: Systems and methods for improving timing closure of new and existing IC chips by breaking at least one parameter of interest into two or more partial parameters. More specifically, a method is provided for that includes propagating at least one timing analysis run for a semiconductor product. The method further includes identifying at least one parameter of interest used in the at least one timing analysis run. The method further includes splitting the at least one parameter into two parts comprising a controlled part and an uncontrolled part. The method further includes correlating or anti-correlating the controlled part with another parameter used in the at least one timing analysis run. The method further includes projecting timing using the correlation or anti-correlation between the controlled part and the another parameter and using the uncontrolled part of the at least one parameter.

Abstract: The computer identifies an integrated circuit design; identifies a timing model associated with the identified integrated circuit design; defines one or more static single sided variables; defines one or more regions of one or more of the defined one or more static single sided variables that are treated linearly; defines one or more multi-sided variables based on the defined one or more regions of the one or more of the defined one or more static single sided variables; identifies one or more timing paths within the identified integrated circuit design; performs a statistical static timing analysis on the identified timing model for the identified one or more timing paths within the identified integrated circuit design utilizing the defined one or more multi-sided variables; provides one or more timing quantities that project within a multi-parameter space based on the performed statistical static timing analysis.

Abstract: Creating an integrated circuit with non-linear variations, the computer identifies an integrated circuit design; identifies a timing model associated with the identified integrated circuit design; defines one or more static single sided variables; defines one or more regions of one or more of the defined one or more static single sided variables that are treated linearly; defines one or more multi-sided variables based on the defined one or more regions of the one or more of the defined one or more static single sided variables; identifies one or more timing paths within the identified integrated circuit design; performs a statistical static timing analysis on the identified timing model for the identified one or more timing paths within the identified integrated circuit design utilizing the defined one or more multi-sided variables; provides one or more timing quantities that project within a multi-parameter space based on the performed statistical static timing analysis.

Abstract: A system for semiconductor chip fabrication. A host system hosts a capacitance adjust tool for performing calculating a ground capacitance adjust for a wire segment going through routing tiles according to the following operations; providing the routing tiles having a plurality of wires wherein the wire segment being a victim wire and neighboring wires being aggressor wires; computing ground capacitance adjusts for a victim wire averaged across all aggressor slew values and across possible spacing values between the victim wire and the neighboring aggressor wires to take into account a potential coupling effect by the neighboring aggressor wires, to guide placement of the wire segment in the routing tiles to avoid coupling noise; and outputting the placement of the wire segments to a tool for manufacturing the semiconductor chip.

Abstract: A computer implemented method for calculating a ground capacitance adjust for a wire segment going through a given routing tile. The method includes providing the routing tile having a plurality of wires wherein the wire segment being a victim wire and neighboring wires being aggressor wires; computing ground capacitance adjusts for a victim wire averaged across all aggressor slew values and across possible spacing values between the victim wire and the neighboring aggressor wires to take into account a potential coupling effect by the neighboring aggressor wires, assuming a distribution of signal slews of wires belonging to the routing tile and assuming the victim wire's neighbors have signal slews from the distribution of slews for this tile for possible spacing values responsible for the coupling effect, to guide placement of the wire segment in the routing tile to avoid coupling noise.