Google Inc.

Xilinx, Inc.

Marvell Asia Pte., Ltd.

Falcon Computing

Peng Zhang

Cody Hao Yu

Xuechao Wei

Peichen Pan

Paul Schelte Albert Reitsma

Gaurav Bhaya

Runzi Chang

Robert Stets

Jonathan William Ray

Glenn Wilson

Eric Wood

Troy L Walker

Thomas J Murray, Iv

Farnaz Azmoodeh

Congxing Cai

Walter Alexander Mostowy

Rudiger Lippert

Walter A Mostowy

Jeffrey E Shaw

A semiconductor device includes a magnetic tunnel junction (MTJ) and first and second electrodes. The MTJ includes: (a) a first ferromagnetic (FM) layer, configured to have a magnetic spin in a first spin direction, and retain the first spin direction while MTJ subjected to electrical current in first and second directions, (b) a second FM layer, configured to have the magnetic spin selectively altered between the first and second spin direction, in response to altering the electrical current between the first and second directions, respectively, and (c) a stack of tunnel barrier (TB) layers, having: a first TB layer disposed over the first FM layer and having a first morphological structure, and a second TB layer, disposed between the first TB layer and the second FM layer and having a second, different, morphological structure. The first and second electrodes are electrically connected to the first and second FM layers, respectively.

Performance of magnetic channel junctions

Systems and methods for automated systolic array design from a high-level program are disclosed. One implementation of a systolic array design supporting a convolutional neural network includes a two-dimensional array of reconfigurable processing elements arranged in rows and columns. Each processing element has an associated SIMD vector and is connected through a local connection to at least one other processing element. An input feature map buffer having a double buffer is configured to store input feature maps, and an interconnect system is configured to pass data to neighboring processing elements in accordance with a processing element scheduler. A CNN computation is mapped onto the two-dimensional array of reconfigurable processing elements using an automated system configured to determine suitable reconfigurable processing element parameters.

Systems and methods for systolic array design from a high-level program

Systems and methods for designing an information processing system are described. In one embodiment, a design space is partitioned into a plurality of independent partitions based on a defined set of rules. A unique processing core is assigned to each partition. A plurality of starting points is generated for each partition, where each starting point is associated with a machine learning algorithm. The starting points for each partition may include a performance driven seed and an area-driven seed. A set of feasible designs associated with the information processing system are determined.

Accelerator automation framework for heterogeneous computing in datacenters

Systems and methods to determine the status of transmitted audio packets. The audio packets can include digital components that are presented by a client computing device. Based on presentation variables and determinations of presentation status, the system can generate a predictive model that can be used to dynamically determines presentation values of audio packets prior to their transmission to a client computing device.

Audio data packet status determination

Systems and methods of identifying ad elements of web pages via a computer network are provided. A script configured to identify an ad slot on a web page can be obtained by at least one data processing system having an ad slot evaluation module. The script can be executed to obtain advertisement parameters of the ad slot and web page parameters of the ad slot and generate a unique identifier of the ad slot. The generated unique identifier is associated with a URL identifier of the web page. The data processing system evaluates at least one database to determine one of a presence and an absence of the historical ad slot data. The data processing system provides content for display in the ad slot of the web page based at least in part on a result of the evaluation of the database.

Web page ad slot identification

A module may be included with a content item. The module may be configured such that the module performs a number of tasks at a first rate via a timer when the module is visible and at a second rate via a timer when the module is not visible. A script may be used with the module to determine the number of times the task has been executed by the module and to compare the number of times the task has been executed to a value. If the number of times that the task has been executed by the module is greater than the value then data may be output to indicate the that content item has been viewed. In some instances, the module may be used with a content item served in a cross-domain iframe.

Detecting visibility of a content item using tasks triggered by a timer

Methods, systems, and apparatus, including computer programs encoded on a computer-readable storage medium, and including a method for predicting visibility of web pages. The method comprises identifying web pages that include slots for presenting content. The method further comprises, for each slot, creating a fingerprint based on a URL associated with the page and structural information associated with the slot's location on the page, determining a likelihood of the slot being visible when the web page is initially presented, the likelihood in the form of a predictive model, and saving the fingerprint and predictive model in a data structure. The method further comprises receiving a request for content from a web page, creating fingerprints associated with the web page, and comparing the fingerprints to fingerprints in the data structure. When a matching fingerprint is located, prediction data is used that is associated with a matching fingerprint for the web page.

Predicting visibility of content items

Growing community of inventors

Los Angeles, CA, United States of America

Peng Zhang