Systems and methods involving construction of a system interconnect in which different channels have different widths in numbers of bits. Example processes to construct such a heterogeneous channel NoC interconnect are disclosed herein, wherein the channel width may be determined based upon the provided specification of bandwidth and latency between various components of the system.

Heterogeneous channel capacities in an interconnect

An embodiment of the invention includes a packet processing pipeline. The packet processing pipeline includes match and action stages. Each match and action stage in incurs a match delay when match processing occurs and each match and action stage incurs an action delay when action processing occurs. A transport delay occurs between successive match and action stages when data is transferred from a first match and action stage to a second match and action stage.

Openflow match and action pipeline structure

Systems and methods described herein are directed to streaming bridge design implementations that help interconnect and transfer transaction packets between multiple source and destination host interfaces through a Network on Chip (NoC) interconnect, which includes a plurality of NoC router layers and virtual channels (VCs) connecting the router layers. Implementations are configured to support a variety of different traffic profiles, each having a different set of traffic flows. Streaming bridge design implementation can divide streaming bridge into a streaming TX bridge and a streaming RX bridge, wherein TX bridge is operatively coupled with host TX interfaces and RX bridge is operatively coupled with host RX interfaces, and where TX bridge forwards transaction packets from host TX interfaces to different router layers/VCs of NoC, and RX bridge, on the other hand, receives packets from NoC router layers/VCs and transmits the packets to host RX interfaces based on Quality of Service.

STREAMING BRIDGE DESIGN WITH HOST INTERFACES AND NETWORK ON CHIP (NoC) LAYERS

Example implementations are directed to more efficiently delivering a multicast message to multiple destination components from a source component. Multicast environment is achieved with transmission of a single message from a source component, which gets replicated in the NoC during routing towards the destination components indicated in the message. Example implementations further relate to an efficient way of implementing multicast in any given NoC topology, wherein one or more multicast trees in the given NoC topology are formed and one of these trees are used for routing a multicast message to its intended destination components mentioned therein.

Supporting multicast in NoC interconnect

Example implementations described herein are directed to a configurable building block, such as a router, for implementation of a Network on Chip (NoC). The router is parameterized by a software layer, which can include the number of virtual channels for a port, the number of ports, the membership information of the virtual channels, clock domain, and so forth. The router may further be configured to implement arbitration techniques and flit processing techniques based on the parameters specified by the software layer.

Configurable router for a network on chip (NoC)

An arithmetic computation circuit is implemented to significantly increase throughput and thereby permit processing of relatively large binary numbers in a relatively small period of time. In one embodiment, the arithmetic computation circuit adapted to add a first binary operand of N bits and a second binary operand of M bits, where N is greater than or equal to M. The circuit includes an adder and a multiplexer circuit. The adder is adapted to combine representative sets of least-significant bits of the first and second binary operands together to produce a least-significant bits partial sum and a carryout. The multiplexer circuit is adapted to output a most-significant bits partial sum by passing one of: a representative set of most-significant bits of the first binary operand, and an offset of the representative set of most-significant bits of the first binary operand; the selection of these inputs is responsive to selection data that is a function of the most-significant bit of the representative set of least-significant bits of the first binary operand.

High-speed computation in arithmetic logic circuit

The precharge of a domino logic stage is controlled based on the precharge delay of a prior domino logic stage. The precharge of the logic stage does not occur until the output of the prior logic stage corresponds to the precharge logic state. Because the precharge logic state output of a preceding stage is an inactive state of a subsequent logic stage, the logic function of the subsequent logic stage is in a non-conducting state when the output of the prior logic stage is in the precharge logic state. By providing the precharge to a subsequent stage only after the output of the prior stage is in the precharge state, there can be no DC current flow during the precharge of the subsequent stage, and the need for an evaluation transistor to block the DC current flow during precharge is eliminated. The elimination of the evaluation transistor eliminates the delay introduced by the evaluation transistor in a precharge logic stage, reduces the circuit area for the logic stage, reduces the load on the clock circuit, and reduces the power consumption of each logic stage.

Deepak Singh

Papers

High-speed computation in arithmetic logic circuit

Domino logic with self-timed precharge