65 results about "Interconnection architecture" patented technology

A data center path switch architecture permits path switching of the signal path of incoming signals to one or more output paths in real time without the need for manual intervention, and without delays associated with current data center network switches. In this architecture, a switching core capable of switching signals directly from the ingress of the switching core to alternate destination ports in real time, either under software or hardware control.

Data center interconnections, which encompass WSCs as well as traditional data centers, have become both a bottleneck and a cost/power issue for cloud computing providers, cloud service providers and the users of the cloud generally. Fiber optic technologies already play critical roles in data center operations and will increasingly in the future. The goal is to move data as fast as possible with the lowest latency with the lowest cost and the smallest space consumption on the server blade and throughout the network. Accordingly, it would be beneficial for new fiber optic interconnection architectures to address the traditional hierarchal time-division multiplexed (TDM) routing and interconnection and provide reduced latency, increased flexibility, lower cost, lower power consumption, and provide interconnections exploiting N×M×D Gbps photonic interconnects wherein N channels are provided each carrying M wavelength division signals at D Gbps.

Various embodiments of the present invention are directed to photonic interconnection architectures that provide high-speed interconnections of microscale or nanoscale devices. In one embodiment of the present invention, a photonic interconnection for interconnecting and synchronizing operation of components within a computing device comprises an optical transmission path for transmitting a number of independent frequency channels within an optical signal provided by an optical signal source. The photonic crystal may include one or more filters located near the waveguide for extracting specific frequency channels transmitted by the waveguide, the frequency channel including one or more frequency channels carrying a clock signal. One or more photodetectors positioned near the one or more filters convert the extracted frequency channels into electrical signals for use by one or more components of the computing device, the electrical signals including one or more clock signals for synchronizing operation of the one or more components.

The invention belongs to the technical field of hardware design of image processing algorithms, and specifically discloses hardware interconnection architecture of a reconfigurable convolutional neural network. The interconnection architecture comprises a data and parameter off-chip caching module, a basic calculation unit array module and an arithmetic logic unit calculation module, wherein the data and parameter off-chip caching module is used for caching pixel data in input to-be-processed pictures and parameters input during convolutional neural network calculation; the basic calculation unit array module is used for realizing core calculation of the convolutional neural network; and the arithmetic logic unit calculation module is used for processing calculation results of the basic calculation unit array module and accumulating a down-sampling layers, activation functions and partial sums. The basic calculation unit array module is interconnected according to a two-dimensional array manner; in a row direction, input data is shared and parallel calculation is realized by using different pieces of parameter data; and in a column direction, a calculation result is transferred rowby row to serve as input of the next row to participate in the operation. The hardware interconnection architecture is capable of reducing the bandwidth demand while enhancing the data reusing ability through structure interconnection.

A multi-celled chip. The chip includes a plurality of hexagonal cells arranged in an array. A plurality of interconnects including Y's connect the cells in clusters of three cells each, so that each of the cells is interconnected.

The present invention relates to a modular interconnection architecture for an expandable multiprocessor machine. It comprises a first interconnection level (MI) comprising connection agents (NCSi) that connect the multiprocessor modules and handle the transactions between the multiprocessor modules, and a second interconnection level (SI) comprising external connection nodes (NCEj) that connect the nodes (Nj) to one another and handle the transactions between the nodes (Nj). Each external connection node (NCEj) comprises two connection agents identical to the connection agent (NCSi), connected head-to-tail, one of the two agents (NCS′j) receives and filters the transactions sent by the node (Nj) to which it is connected. The other agent (NCS″j) receives and filters the transactions sent by the other nodes (Nj) to which it is connected. Its applications specifically include the construction of an entire range of machines: UMA, QUASI-UMA, NUMA, cluster, etc.

Optical solutions to address and overcome the issues of superseding/replacing electrical interconnection networks have generally exploited some form of optical space switching. Such optical space switching architectures required multiple switching elements, leading to increased power consumption and footprint issues. Accordingly, it would be beneficial for new optical, e.g. fiber optic or integrated optical, interconnection architectures to address the traditional hierarchal time-division multiplexed (TDM) space based routing and interconnection to provide reduced latency, increased flexibility, lower cost, and lower power consumption. Accordingly, it would be beneficial to exploit networks operating in multiple domains by overlaying mode division multiplexing to provide increased throughput in bus, point-to-point networks, and multi-cast networks, for example, discretely or in combination with wavelength division multiplexing.

Data center interconnections, which encompass WSCs as well as traditional data centers, have become both a bottleneck and a cost/power issue for cloud computing providers, cloud service providers and the users of the cloud generally. Fiber optic technologies already play critical roles in data center operations and will increasingly in the future. The goal is to move data as fast as possible with the lowest latency with the lowest cost and the smallest space consumption on the server blade and throughout the network. Accordingly, it would be beneficial for new fiber optic interconnection architectures to address the traditional hierarchal time-division multiplexed (TDM) routing and interconnection and provide reduced latency, increased flexibility, lower cost, lower power consumption, and provide interconnections exploiting scalable optical modular optically switched interconnection network as well as temporospatial switching fabrics allowing switching speeds below the slowest switching element within the switching fabric.

This paper presents a methodology for designing System-On-Chip interconnection architectures providing a high level of protection from crosstalk effects. An event driven simulator enriched with fault injection capabilities is exploited to evaluate the dependability level of the system being designed. The simulation environment supports several bus coding protocols and thus designers can easily evaluate different design alternatives. To enhance the dependability level of the interconnection architecture, we propose a distributed bus guardian scheme, where dedicated hardware modules monitor the integrity of the information transmitted over the bus and provide error correction mechanisms.

The invention pertains to the field of data communication, and relates to an RapidIO network recursive enumeration method. The invention provides an RapidIO network recursive enumeration networking method based on depth-first path weighting, the shortest communication path between a master machine and a slave machine in the RapidIO network is caculated, and an optimal path dynamic enumeration method which has advantages of simple structure, good adaptability and good reliability and can be used to support the change of external connection device. IDT Tsi578 is used as the RapidIO networking switchboard to realize the optimal RapidIO networking scheme based on the Tsi578 switchboard device. The method of the invention adopts the following technical scheme that in a system in which the RapidIO bus is used as the interconnection architecture, the RapidIO master machine mode is taken as the reference, and the route hop is taken as the weighted value to calculate the distances between each slave machine node in the RapidIO network and RapidIO master machine mode, for each slave machine mode that newly found, the shortest path between each slave machine mode and the RapidIO master machine mode is selected dynamically, and the route information table of each slave machine mode ID and the corresponding switchboard device Tsi578 is set according to the shortest path.

The present system and methods are directed to the interconnection of clusters of emulation processors comprising emulation processors in a software-driven hardware design verification system. The processors each output one NBO output signal. The clusters are interconnected by partitioning a common NBO bus into a number of smaller NBO busses, each carrying unique NBO signals but together carrying every NBO. Each of the smaller NBO busses are passed into a series of multiplexers, each dedicated to a particular processor. The multiplexers select a signal for output back to the emulation clusters. The multiplexers that handle these smaller NBO busses are narrower than was previously required, thus reducing the amount of power, interconnect, and area required by the multiplexer array and dedicated interconnect.

The invention relates to a network architecture planning information generation method and device, a computer readable storage medium and computer equipment. The method comprises the steps: obtaininginterconnection information between a to-be-constructed network architecture object and an interconnection architecture object; searching a hardware module model of the hardware module matched with the interconnection port attribute, and determining the number of the hardware modules of the hardware module according to the number of the interconnection opposite ends and the number of the interconnection ports; obtaining hardware material information for constructing the network architecture object according to the hardware module model and the hardware module number; generating connection relation information of interconnection ports between the network architecture object and the interconnection architecture object according to the number of the interconnection opposite ends and the number of the interconnection ports; and generating network architecture planning information according to the hardware material information and the connection relation information. According to the schemeprovided by the invention, the network architecture planning information meeting the construction requirement of the business server can be dynamically generated, and the waste of equipment resourcesis effectively avoided.

The invention provides a server, which comprises a main circuit board, a plurality of CPU (central processing unit) circuit boards and a plurality of bridging plates. At least one CPU is arranged on each CPU circuit board, connectors used for being electrically connected are disposed on two sides of each CPU circuit board, and each CPU circuit board is electrically connected with one of the main circuit boards and the bridging plates by the connectors. More CPUs can be arranged on the main circuit board by the CPU circuit boards to improve computation capacity of the server, interconnection architecture among the CPU circuit boards, the CPU bridging plates and the main circuit board is realized, accordingly, information transmission among the CPUs and the main circuit board is realized, and stability of the server is enhanced.

The invention relates to the technical field of artificial neural networks, in particular to an asynchronous communication interconnection architecture and a brain-like chip with the architecture, and the asynchronous communication interconnection architecture comprises an in-chip asynchronous communication interconnection architecture, an inter-chip asynchronous communication interconnection architecture, a neuron calculation unit and an on-chip routing unit; the neuron calculation unit and the on-chip routing unit are respectively provided with an independent clock management module in an independent clock domain, the neuron calculation unit and the on-chip routing unit are connected in the same brain chip, and the on-chip routing unit and the adjacent on-chip routing unit are connected through the on-chip asynchronous communication interconnection architecture. The adjacent brain-like chips are connected with each other through the asynchronous communication interconnection architecture between the chips. According to the method, a large number of neuron computing units can be efficiently integrated in the brain-like chip, and meanwhile, the brain-like chip is efficiently cascaded and expanded, so that neuron computing resources of a large scale are obtained.

The invention discloses a buffer-free optical interconnection architecture and method for a data center. The architecture is divided into an inlet wire clamp, a three-stage exchange module and an outlet wire clamp, wherein the exchange module comprises a first-stage AWGR, a third-stage AWGR and a second-stage TWC; the first-stage AWGR input port is provided with an inlet wire clamp, the third-stage AWGR output port is provided with an outlet wire clamp, and the first-stage AWGR output port and the third-stage AWGR input port are connected through a second-stage TWC. Signals of the server are subjected to wavelength division multiplexing and then forwarded to the switching module through wavelength modulation of the entrance line card, optical signals with specific wavelengths are forwardedto the exit line card through TWC wavelength conversion and AWGR cyclic routing, and the exit line card processes the optical signals and then sends the optical signals to the server. High-capacity,high-reliability and low-power-consumption all-optical interconnection is achieved, a data center is effectively flattened, the defects that a traditional space switch is low in port bandwidth and limited in port number are overcome, and the all-optical interconnection system has the advantages of being high in reliability, low in complexity, low in delay and the like.

The systems and methods are configured to efficiently and effectively include processing capabilities in memory. In one embodiment, a processing in memory (PIM) chip a memory array, logic components, and an interconnection network. The memory array is configured to store information. In one exemplary implementation the memory array includes storage cells and array periphery components. The logic components can be configured to process information stored in the memory array. The interconnection network is configured to communicatively couple the logic components. The interconnection network can include interconnect wires, and a portion of the interconnect wires are located in a metal layer area that is located above the memory array.

The invention provides an IPTV network interconnection architecture and an interconnection method thereof, which are used for the interconnection of IPTV networks with different service patterns. The IPTV network interconnection architecture comprises a terminal user functional entity, an application functional entity, a content transmission functional entity, and a network functional entity or a transmission functional entity. The network functional entity or the network functional entity further comprises boundary functional entities which are connected with a core transmission functional entity in the network functional entity or the network functional entity, and when the IPTV networks with different service patterns are interconnected, the transmission layer interconnection is realized through the interconnection of the boundary functional entities.