Abstract: Embodiments of a microelectronic assembly includes: an interposer comprising a first portion in contact along an interface with a second portion; a first integrated circuit (IC) die embedded in a dielectric material in the first portion of the interposer; and a second IC die coupled to the first portion of the interposer opposite to the second portion, wherein: the second portion comprises a glass substrate with a channel within the glass substrate, a portion of the channel has an opening at the interface, a conductive pad in the first portion is exposed in the opening, and the conductive pad is coupled to a circuit in at least one of the first IC die or the second IC die.

Abstract: Embodiments of a package substrate includes: a conductive via in a first layer, the first layer comprising a positive-type photo-imageable dielectric; a conductive trace in a second layer, the second layer comprising a negative-type photo-imageable dielectric; and an insulative material between the first layer and the second layer, the insulative material configured to absorb electromagnetic radiation in a wavelength range between 10 nanometers and 800 nanometers. The conductive via is directly attached to the conductive trace through the insulative material, the positive-type photo-imageable dielectric is soluble in a photoresist developer upon exposure to the electromagnetic radiation, and the negative-type photo-imageable dielectric is insoluble in the photoresist developer upon exposure to the electromagnetic radiation.

Abstract: Embodiments are generally directed to memory prefetching in multiple GPU environment. An embodiment of an apparatus includes multiple processors including a host processor and multiple graphics processing units (GPUs) to process data, each of the GPUs including a prefetcher and a cache; and a memory for storage of data, the memory including a plurality of memory elements, wherein the prefetcher of each of the GPUs is to prefetch data from the memory to the cache of the GPU; and wherein the prefetcher of a GPU is prohibited from prefetching from a page that is not owned by the GPU or by the host processor.

Abstract: Disclosed herein are radio frequency (RF) front-end structures, as well as related methods and devices. In some embodiments, an RF front-end package may include an RF package substrate including an embedded passive circuit element. At least a portion of the embedded passive circuit element may be included in a metal layer of the RF package substrate. The RF package substrate may also include a ground plane in the metal layer.

Abstract: Embodiments described herein include software, firmware, and hardware logic that provides techniques to perform arithmetic on sparse data via a systolic processing unit. One embodiment provides for data aware sparsity via compressed bitstreams. One embodiment provides for block sparse dot product instructions. One embodiment provides for a depth-wise adapter for a systolic array.

Abstract: Disclosed herein are tools and methods for subtractively patterning metals. These tools and methods may permit the subtractive patterning of metal (e.g., copper, platinum, etc.) at pitches lower than those achievable by conventional etch tools and/or with aspect ratios greater than those achievable by conventional etch tools. The tools and methods disclosed herein may be cost-effective and appropriate for high-volume manufacturing, in contrast to conventional etch tools.

Abstract: An apparatus to facilitate increasing processing resources in processing cores of a graphics environment is disclosed. The apparatus includes a plurality of processing resources to execute one or more execution threads; a plurality of message arbiter-processing resource (MA-PR) routers, wherein a respective MA-PR router of the plurality of MA-PR routers corresponds to a pair of processing resources of the plurality of processing resources and is to arbitrate routing of a thread control message from a message arbiter between the pair of processing resources; a plurality of local shared cache (LSC) sequencers to provide an interface between at least one LSC of the processing core and the plurality of processing resources; and a plurality of instruction caches (ICs) to store instructions of the one or more execution threads, wherein a respective IC of the plurality of ICs interfaces with a portion of the plurality of processing resources.

Abstract: Disclosed herein are source/drain regions in integrated circuit (IC) structures, as well as related methods and components. For example, in some embodiments, an IC structure may include: an array of channel regions, including a first channel region and an adjacent second channel region; a first source/drain region proximate to the first channel region; a second source/drain region proximate to the second channel region; and an insulating material region at least partially between the first source/drain region and the second source/drain region.

Abstract: A non-linear activation function may be approximated by linear functions. The input range of the activation function may be divided into input segments. One or more input segments may be selected based on statistical analysis of input data elements in the input range. A parameter of a first linear function that approximates the activation function for at least part of a selected input segment may be stored in a first portion of a first look-up table (LUT). The first portion of the first LUT is dedicated to a first group of post processing engines (PPEs). A parameter of a second linear function that approximates the activation function for at least part of an unselected input segment may be stored in a shared pool of LUT entries, which includes a second portion of the first LUT and a portion of a second LUT and is shared by multiple groups of PPEs.

Abstract: One embodiment provides for a computer-readable medium storing instructions that cause one or more processors to perform operations comprising determining a per-layer scale factor to apply to tensor data associated with layers of a neural network model and converting the tensor data to converted tensor data. The tensor data may be converted from a floating point datatype to a second datatype that is an 8-bit datatype. The instructions further cause the one or more processors to generate an output tensor based on the converted tensor data and the per-layer scale factor.

Abstract: A computing platform comprising a plurality of disaggregated data center resources and an infrastructure processing unit (IPU), communicatively coupled to the plurality of resources, to compose a platform of the plurality of disaggregated data center resources for allocation of micro service s cluster.

Abstract: A method and system of image processing with multi-skeleton tracking uses a temporal object key point loss metric.

Abstract: In an example, an apparatus comprises a plurality of execution units comprising at least a first type of execution unit and a second type of execution unit and logic, at least partially including hardware logic, to expose embedded cast operations in at least one of a load instruction or a store instruction; determine a target precision level for the cast operations; and load the cast operations at the target precision level. Other embodiments are also disclosed and claimed.

Abstract: A bumpless hybrid organic glass interposer. One or more high density pattern (HDP) routing layers are placed on a functional, thin, carrier, separate from the intended organic substrate patch or package. The HDP layer(s) is/are then attached to the substrate package. The interposers achieve electrical connections between the HDP layer and underlying routing layer of the substrate package by utilizing a self-align dry etch process through landing pads connected to the HDP routing.

Abstract: Described herein is a truncated modified Booth squarer that is commutative and accurate to 1 unit in the last place. In various embodiments, the truncated Booth squarer is a radix-4 Booth squarer or a radix-8 Booth squarer. The truncated Booth squarer can be included within integer, floating-point, or fixed-point units within a graphics processor or compute accelerator, including matrix accelerator units or tensor processors.

Abstract: Computer computation of correctly rounded floating point summation is described. An example of apparatus includes a first circuit to sort multiple floating point (FP) values based on an exponent of each FP value and store the sorted FP values in a buffer, and to provide the plurality of FP values for summation sequentially in a sorted order starting with a FP value having a smallest exponent; a second circuit to iteratively sum the FP values and store an accumulated value, generate and store a residual value representing fully resolved bits from the accumulator, and generate an intermediate output including the residual value; and a third circuit to perform final rounding of the output, the final rounded output being a correctly rounded summation of the maximum floating point values.

Abstract: An apparatus includes a central processing unit (CPU), including a plurality of processing cores, each having a cache memory, a fabric interconnect coupled to the plurality of processing cores and cryptographic circuitry, coupled to the fabric interconnect including mesh stop station to receive memory data and determine a destination of the memory data and encryption circuitry to encrypt/decrypt the memory data based on a destination of the memory data.

Abstract: A first die has a port to couple the first die to a second die over a die-to-die interconnect. The port includes circuitry to implement a physical layer of the die-to-die interconnect, send first protocol identification data over the physical layer to identify a first protocol in a plurality of protocols, send first data over the interconnect to the second die, wherein the first data comprise data of the first protocol, send second protocol identification data over the physical layer to identify a different second protocol in the plurality of protocols, and send second data over the interconnect to the second die, wherein the second data comprise flits of the second protocol.

Abstract: Examples described herein include a system comprising: a processing unit package comprising: at least one core and at least one offload processing device communicatively coupled inline between the at least one core and a network interface controller, the at least one offload processing device configurable to perform packet processing. In some examples, the at least one offload processing device is to allow mapping of packet processing pipeline stages of networking applications among software running on the at least one core and the at least one offload processing device to permit flexible entry, exit, and re-entry points among the at least one core and the at least one offload processing device.

Abstract: Techniques are disclosed for using neural network architectures to estimate predictive uncertainty measures, which quantify how much trust should be placed in the deep neural network (DNN) results. The techniques include measuring reliable uncertainty scores for a neural network, which are widely used in perception and decision-making tasks in automated driving. The uncertainty measurements are made with respect to both model uncertainty and data uncertainty, and may implement Bayesian neural networks or other types of neural networks.