Arm Limited

Netspeed Systems

Other

Anup Gangwar

Nitin Kumar Agarwal

Honnahuggi Harinath Venkata Naga Ambica Prasad

Ravishankar Sreedharan

Sailesh Kumar

Vishnu Mohan Pusuluri

Narayana Sri Harsha Gade

Poonacha Kongetira

Rimu Kaushal

Zheng Xu

Santhosh Patchamatla

A computer-based method and system for synthesizing a Network-on-Chip (NoC) is provided. One method includes determining physical data, device data, bridge data, traffic data and domain data based on an input specification for the NoC; assigning a domain to each bridge port; partitioning each traffic flow into one of a plurality of bins based on the bridge port domain assignments and the domain crossing constraints; creating a virtual node at each bridge port endpoint; generating a candidate topology for each bin based on the physical data, the device data, the bridge data, the traffic data, the domain data and the virtual nodes, each candidate topology including bridge ports, a tree of routers, routes and connections; and generating a final topology by merging the candidate topologies.

Network-on-chip topology generation

The present disclosure provides computer-based methods and a system for synthesizing a NoC that advantageously generate balanced NoC topologies without end-to-end fairness or local credit-based arbitration, and improve NoC performance when destination device bridge ports support only one incoming physical link per channel. More particularly, a clock domain is assigned to certain routers that satisfies the minimum frequency for the router while reducing clock domain transitions to neighboring routers, and the traffic flows received by these routers are balanced based on the traffic flow packet rates.

Network-on-Chip topology generation

The present disclosure provides a computer-based method and system for synthesizing a NoC. Traffic data is determined or received, and a baseline topology is generated or received. For each router in the baseline topology, a number of edge virtual channel (EVC) combinations is determined, the transmittablility of the traffic classes are determined, and, when the traffic classes are not transmittable, the router is identified. A traffic class affinity graph (TCAG) is generated for each identified router. Traffic class combinations are generated for the identified routers based on the TCAGs and EVC combinations. The traffic classes of the identified routers are merged based on the traffic class combinations. A final EVC combination for each identified router is determined based on the merged traffic classes. A final topology is generated based, at least in part, on the merged traffic classes and the final EVC combinations for the identified routers.

The present disclosure provides a computer-based method and system for synthesizing a Network-on-Chip (NoC). Physical data, device data, bridge data and traffic data are determined based on an input specification for the NoC. A virtual channel (VC) is assigned to each traffic flow to create a plurality of VC assignments. A topology is generated, based on the physical data, the device data, the bridge data, the traffic data and the VC assignments, which includes bridge ports, routers and connections. Final locations for relocatable NoC elements (e.g., routers, etc.) are determined based on NoC element energy values for the relocatable NoC elements, and protocol-level pipelines may be inserted into the connections based on a timing parameter.

Network-on-chip element placement

The present disclosure provides a computer-based method and system for synthesizing a NoC. Physical data, device data, bridge data and traffic data are determined based on an input specification for the NoC. A virtual channel (VC) is assigned to each traffic flow. A head of line (HoL) conflict graph (HCG) is constructed based on the traffic data and the VC assignments. The HGC is modified based on bridge data and the traffic data to generate a modified HCG. A plurality of traffic graphs (TGs) are constructed based on the physical data, the bridge data, the traffic data and the modified HCG. A candidate topology is generated for each TG, which includes the bridge ports, routers and connections. The candidate topologies are merged to create a merged candidate topology, and the routers within the merged candidate topology are merged to generate a final topology for the NoC.

The present disclosure advantageously provides a system, a computer-readable medium and a method for synthesizing a Network-on-Chip (NoC). A plurality of route feature vectors are determined based on a network configuration for the NoC. The network configuration includes bridge ports, routers, connections and routes. A link size is determined for each router by providing route feature vectors to a supervised learning-based (SLB) model. The SLB model generates a plurality of route label vectors based on the route feature vectors. Each route label vector is associated with a route feature vector, and includes the link size and a route position for each router. A resizer is added between a bridge and a router with different link sizes or between adjacent routers with different link sizes. Pipeline and retiming components are added based on timing. An output specification is then generated for the NoC.

Network-on-chip link size generation

The present disclosure provides a computer-based method and system for synthesizing a NoC. Physical data, device data, bridge data and traffic data are determined based on an input specification for the NoC. A virtual channel (VC) is assigned to each traffic flow. A head of line (HoL) conflict graph (HCG) is constructed based on the traffic data and the VC assignments. A color is assigned to each HCG node to minimize HoL conflicts. A traffic graph (TG) is constructed for each color based on the physical data, the bridge data, the traffic data and the HCG, and a candidate topology is generated for each color based on the respective TG. The candidate topology for each color is merged to create a merged candidate topology, and the routers within the merged candidate topology are merged to generate a final topology for the NoC.

Network on-chip topology generation

Virtual channel assignment in a network is achieved by constructing a Traffic Conflict Graph (TCG) dependent upon a network interconnect topology. The TCG has vertices corresponding to traffic entries in a network specification and edges that connect pairs of vertices. An edge weight, dependent upon interconnect topology and traffic flow characteristics, is assigned to each edge. The vertices are colored using minimum or soft coloring and the virtual channels are mapped to the traffic entries, according to the resulting colors, to provide a virtual channel assignment. The TCG may be constructed by generating a vertex in the TCG to represent each traffic entry, assigning a traffic flow characteristic of a traffic entry to a corresponding vertex and generating an edge between first and second vertices when a number of 'qualified' common edges, across all routes for corresponding traffic entries, is greater than zero.

Virtual channel assignment for topology constrained network-on-chip design

A computer-method is provided for designing a router network to connect components of an integrated circuit, where the router network comprises a plurality of connected data routing elements. The method comprises generating an undirected graph to represent a mesh of candidate router elements, where the candidate data routing elements are positioned dependent on at least one characteristic of the integrated circuit. The undirected graph comprises a node to represent each candidate data routing element and an edge to represent each connection between the candidate data routing elements. A candidate router network generation process is then performed a plurality of times, where each performance of the candidate router network generation process comprises: generating a tree structure from a plurality of possible tree structures to interconnect the nodes of the undirected graph; for each node in the tree structure, computing a compactness index indicative of a path cost associated with connections to at least a subset of other nodes in the tree structure; for each component to be connected to the router network, identifying a set of available nodes within the tree structure and connecting the component to a selected node within that set taking account of the compactness index of the nodes within the set, to thereby establish a candidate router network; and computing a cost value for the candidate router network taking into account each pair of components that are to communicate with each other via the candidate router network. One of the candidate router network is then selected as the router network based on the computed cost value for each candidate router network.

Integrated circuit design

A computer-implemented method of integrated circuit design comprises: using a computer, detecting data communication paths and associated data traffic requirements between a set of data handling nodes in an integrated circuit layout, the data handling nodes acting as routing nodes or either source nodes or sink nodes for a given data communication path, each source node and each sink node having a respective provisional data width, each data communication path having at least one routing node between the source node and the sink node; using the computer, assigning a provisional data width to each routing node so that, for each of the detected data communication paths, the one or more routing nodes in that data communication path have a provisional data width sufficient to handle the data traffic requirement associated with that communication path; using the computer, performing one or more iterations of modifying the integrated circuit topology by:

Systems and methods of the present disclosure relate to automatically and/or dynamically generating one or more power management sequences for SoC and NoC architectures from a given input specification having one or a combination of NoC design specification, traffic specification, traffic profile, power profile information, initiator-consumer relationship, interdependency between components, retention information, external factors, among other allied configurations/information to enable efficient switching of one or more hardware elements from one power profile to another.

Automatic generation of power management sequence in a SoC or NoC

A computer-implemented method of generating an integrated circuit design comprises: using a computer, detecting communication paths between data handling nodes, the data handling nodes comprising data source nodes, data sink nodes and data routing nodes operating according to respective power domains, clock domains and data traffic parameters, in a network of the data handling nodes; using the computer, for a given communication path in a direction of data flow from a data source node to a data sink node, for each given data routing node in the given communication path to which data is communicated in the direction of data flow by a set of one or more other data handling nodes, to perform the following steps: (i) detecting a power domain and data traffic parameters of each data handling node of the set of one or more other data handling nodes communicating data to said each given data routing node; (ii) assigning a power domain to said each given data routing node in dependence upon the detected power domains and the detected data traffic parameters of the set of one or more other data handling nodes; and (iii) assigning a clock domain to said each given data routing node, from a set of one or more candidate clock domains applicable to the assigned power domain, so that said each given data routing node, operating in the assigned clock domain, provides data routing according to the detected data traffic parameters of at least those of the set of one or more other data handling nodes operating according to the assigned power domain of said each given data routing node.

Aspects of the present disclosure relate to a method and system for hybrid and/or distributed implementation of generation and/or execution of power profile management instructions. An embodiment of the present disclosure provides a hardware element of a SoC/NoC that can be configured to generate and/or execute power profile management instructions using a hybrid combination of software and hardware, wherein the hardware element can be run in parallel with other hardware elements of the SoC/NoC to generate and execute power profile management instructions for different segments or regions of the SoC/NoC for efficient and safe working thereof.

Hardware and software enabled implementation of power profile management instructions in system on chip

Aspects of the present disclosure relate to methods, systems, and computer readable mediums for generating transition state specifications that include information regarding low power behavior of a System on Chip (SoC) and/or a Network on Chip (NoC). Such transition state specifications can enable verification of switching behavior when elements/components of a SoC/NoC or a subset thereof switch from one power profile to another, or when the elements/components switch in stable states of power based on inputs such as voltages, clocks, power domains, and traffic.

Growing community of inventors

Austin, TX, United States of America

Anup Gangwar