Abstract: A disclosed semiconductor device includes (1) a silicon stack comprising a front-side Back-End-of-Line (BEOL) stack and a back side BEOL stack, the front-side BEOL stack comprising a plurality of signal routes and the back-side BEOL stack comprising a plurality of power delivery routes, and (2) a plurality of auxiliary power paths formed within the front-side BEOL stack and electrically coupled to the plurality of power delivery routes of the back-side BEOL stack via a plurality of programmable switches, the plurality of power delivery routes, the plurality of programmable switches, and the plurality of auxiliary power paths forming a programmable power delivery network (PDN). Various other apparatuses, systems, and methods of operation are also disclosed.

Abstract: A method for increasing capacitance density within an integrated passive device can include forming a first trench capacitor within a substrate, forming a second trench capacitor within an insulating layer overlying the substrate, and connecting the first and second trench capacitors through connection vias that extend through the insulating layer to form an integrated passive device (IPD) capacitor. A high capacitance density device can include a stacked and co-integrated architecture of two or more tiers of trench capacitors.

Abstract: A method for increasing capacitance density within an integrated passive device can include forming a first trench capacitor within a first insulating layer overlying a substrate, forming a second trench capacitor within a second insulating layer overlying the first insulating layer, and connecting the first and second trench capacitors through connection vias that extend through the second insulating layer to form an integrated passive device (IPD) capacitor. A high capacitance density device can include a stacked and co-integrated architecture of two or more such layers.

Abstract: An apparatus includes one or more processors that are configured to determine a pixel-by-pixel bounds for a perturbed image, generate an adversarial example using an adversarial example generation technique, and modify the adversarial example to generate the perturbed image based on the pixel-by-pixel bounds. When an initial perturbed image does not reside within the pixel-by-pixel bounds, the one or more processors adjust the initial perturbed image to generate the perturbed image by a Weber-Fechner based adversarial perturbation to reside within the pixel-by-pixel bounds. The one or more processors provide the perturbed image to a computing device in an image-based Completely Automated Public Turing Test to tell Computers and Humans Apart (CAPTCHA).

Abstract: In order to efficiently process graphics data, operations are performed including allocating a first set of resource slots for a first execution instance of a pipeline shader program; correlating the first set of resource slots with graphics pipeline passes; and on a second execution instance of the pipeline shader program, assigning resource slots, from the first set of resource slots, to the graphics pipeline passes, based on the correlating.

Abstract: Systems, apparatuses, and methods for dynamically coalescing multi-bank memory commands to improve command throughput are disclosed. A system includes a processor coupled to a memory via a memory controller. The memory also includes processing-in-memory (PIM) elements which are able to perform computations within the memory. The processor generates memory requests targeting the memory which are sent to the memory controller. The memory controller stores commands received from the processor in a queue, and the memory controller determines whether opportunities exist for coalescing multiple commands together into a single multi-bank command. After coalescing multiple commands into a single combined multi-bank command, the memory controller conveys, across the memory bus to multiple separate banks, the single multi-bank command and a multi-bank code specifying which banks are targeted. The memory banks process the command in parallel, and the PIM elements process the data next to each respective bank.

Abstract: Package lids with carveouts configured for processor connection and alignment are described. Lid carveouts are configured to align and mechanically secure a cooling device to the package lid by receiving protrusions of the cooling device. Because the lid carveouts ensure precise alignment and orientation of a cooling device relative to a package lid, the lid design enables targeted cooling of discrete portions of the lid. Lid carveouts are further configured to expose one or more connectors disposed on a surface that supports package internal components. When contacted by corresponding connectors of a cooling device, the lid carveouts enable direct connections between the package and the attached cooling device. By creating a direct connection between package components and an attached cooling device, the lid carveouts enable a high-speed connection for proactive and on-demand cooling actuation.

Abstract: A graphics processing unit (GPU) includes a plurality of programmable processing cores configured to process graphics primitives and corresponding data and a plurality of fixed-function hardware units. The plurality of processing cores and the plurality of fixed-function hardware units are configured to implement a configurable number of virtual pipelines to concurrently process different command flows. Each virtual pipeline includes a configurable number of fragments and an operational state of each virtual pipeline is specified by a different context. The configurable number of virtual pipelines can be modified from a first number to a second number that is different than the first number. An emulation of a fixed-function hardware unit can be instantiated on one or more of the graphics processing cores in response to detection of a bottleneck in a fixed-function hardware unit. One or more of the virtual pipelines can then be reconfigured to utilize the emulation instead of the fixed-function hardware unit.

Abstract: Systems, apparatuses, and methods for moving data from a memory in a computing system to an I/O device. A system includes a processor, memory, I/O controller, and power management circuitry. An application stores data in the memory that is to be ultimately conveyed to an I/O device. The I/O controller is configured to convey the data to an I/O device according to a service interval. The I/O controller is configured to fetch a first data item from the memory stored by the application, and prefetch one or more additional data items from memory. The first data and prefetched data are stored in a locally accessible buffer of the I/O controller. The I/O controller is then configured to convey each of the first data and one or more data items from the buffer to the I/O device at regular intervals of time during a given period of time, prior to initiating a fetch of additional data from the memory.

Abstract: Sparse matrix operations using processing-in-memory is described. In accordance with the described techniques, a processing-in-memory component of a memory module receives a request for a vector element stored at a first location in memory of the memory module. The processing-in-memory component identifies an index value for a non-zero element in a sparse matrix using a representation of the sparse matrix stored at a second location in the memory. The processing-in-memory component then outputs a result that includes the vector element by retrieving the vector element from the first location in memory using the index value.

Abstract: A method for driver calibration in die-to-die interfaces can include calibrating a delay lock loop of a delay line unit cell, by at least one processor, based on drive and load conditions of one or more driver unit cells of a physical layer of a die-to-die interconnect. The method can additionally include generating a clock signal, by the at least one processor, based on the delay lock loop. The method can further include communicating data, by the at least one processor, over the die-to-die interconnect based on the clock signal. Various other methods and systems are also disclosed.

Abstract: Techniques are disclosed for communicating between a machine learning accelerator and one or more processing cores. The techniques include obtaining data at the machine learning accelerator via an input/output die; processing the data at the machine learning accelerator to generate machine learning processing results; and exporting the machine learning processing results via the input/output die, wherein the input/output die is coupled to one or more processor chiplets via one or more processor ports, and wherein the input/output die is coupled to the machine learning accelerator via an accelerator port.

Abstract: Techniques are provided for executing wavefronts. The techniques include at a first time for issuing instructions for execution, performing first identifying, including identifying that sufficient processing resources exist to execute a first set of instructions together within a processing lane; in response to the first identifying, executing the first set of instructions together; at a second time for issuing instructions for execution, performing second identifying, including identifying that no instructions are available for which sufficient processing resources exist for execution together within the processing lane; and in response to the second identifying, executing an instruction independently of any other instruction.

Abstract: An accelerator device includes a first processing unit to access a structure of a graph dataset, and a second processing unit coupled with the first processing unit to perform computations based on data values in the graph dataset.

Abstract: A method and an apparatus for decoding an image are disclosed. A region of the image is selected and the decoding is selected region and associated metadata is performed. Pixels for a generated for a decoded image based on the decoded selected region and metadata.

Abstract: Techniques for performing memory operations are disclosed herein. The techniques include obtaining statistics for operation of a device, the statistics including either or both of performance statistics and memory access statistics; generating a plurality of visualizations of the statistics in one of an overlay mode or a scene annotation mode; and displaying the plurality of visualizations.

Abstract: A message passing interface processing system is described. In accordance with message passing logic, a node selects an affinity domain for communication of data associated with a message passing interface and selects a first rank of a first process of the message passing interface assigned to a first partition of the affinity domain as a first partition leader rank and an affinity domain leader rank. The node selects a second rank of a second process of the message passing interface assigned to a second partition of the affinity domain as second partition leader rank, receives the data at the first partition leader rank, and communicates the data from the first partition leader rank to the second partition leader rank.

Abstract: In accordance with the described techniques, a command processor processes a fiber graph that includes fibers each having one or more tasks and indicates dependencies between the fibers and between tasks within the fibers. As part of this, the command processor dispatches a task from a fiber for execution by a processing element array based on the fiber being enqueued in a ready queue and the dependencies of the task being resolved. While the task is dispatched and unexecuted by the processing element array, the command processor enqueues the fiber in a sleep queue. Further, the command processor enqueues the fiber in a check queue based on the one or more tasks of the fiber having been executed by the processing element array. Based on the fiber being in the check queue, the command processor enqueues a dependent fiber in the ready queue that depends from the fiber.

Abstract: A compilation technique is provided. The technique includes including a first instruction into a first executable for a first auxiliary processor, wherein the first instruction specifies execution by the first auxiliary processor; and including a second instruction into the first executable, wherein the second instruction targets resources that have affinity with the first auxiliary processor.

Abstract: A system and method for ranking computing resources in a distributed computing marketplace is disclosed. Ranking may be based on the performance factors that the system predicts will have the greatest impact on the particular application the user plans to run. A performance database stores historical performance data for applications that have been executed on multiple different computer systems. The database is checked to see if the application, or one similar, has already been run on any of the computing systems participating in the distributed computing marketplace. If so, the existing performance data is used to predict which performance factors will have the greatest impact on the application. Those factors are then used to rank the available computing systems options for the user.