# US 2021/0344618 A1 · Interconnection Network With Adaptable Router Lines For Chiplet-Based Manycore Architecture > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 2021/0344618 A1 - Document ID: US20210344618A1 - Status: Published application - Kind: A1 - Application number: 17/085,454 - Filed: 2020-10-30 - Published: 2021-11-04 - Earliest claimed: 2020-05-04 - Assignees: The George Washington University - Applicants: The George Washington University - Inventors: Hao ZHENG; Ke WANG; Ahmed LOURI - Web page: https://patentgrep.com/patent/US20210344618A1 ## CPC - H04L 49/109: Integrated on microchip, e.g. switch-on-chip - H04L 12/44: Star or tree networks - H04L 41/0813: characterised by the conditions triggering a change of settings - H04L 41/0816: the condition being an adaptation, e.g. in response to network events - H04L 41/12: Discovery or management of network topologies - H04L 45/306: {Route determination based on the nature of the carried application} - H04L 45/48: Routing tree calculation - H04L 45/60: Router architectures - H04L 49/15: Interconnection of switching modules - H04L 49/254: {Centralised controller, i.e. arbitration or scheduling} - H04L: TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION (arrangements common to telegraphic and telephonic communication H04M) ## Abstract An interconnection network for a processing unit having an array of cores. The interconnection network includes routers and adaptable links that selectively connect routers in the interconnection network. For example, each router may be electrically connected to one or more of the adaptable links via one or more multiplexers and a link controller may control the multiplexers to selectively connect routers via the adaptable links. In another example, adaptable links may be formed as part of an interposer and the link controller selectively connect routers via the adaptable links in the interposer using interposer switches. The adaptable links enable the interconnection network to be dynamically partitioned. Each of those partitions may be dynamically reconfigured to form a topology. ## Claims ### Claim 1 An on-chip interconnection network for a processing unit having an array of cores, the interconnection network comprising: a plurality of routers, each router being associated with one or more of the cores; adaptable links that are configured to selectively connect each router with one of a plurality of the routers; and a link controller configured control the adaptable links to: dynamically partition the interconnection network; and select a topology for each partition; dynamically connect the routers of each partition to form a topology. ### Claim 2 (depends on 1) The interconnection network of claim 1, wherein: each router is electrically connected to one or more of the adaptable links via one or more multiplexers; and the link controller controls the multiplexers to selectively connect routers via the adaptable links. ### Claim 3 (depends on 2) The interconnection network of claim 2, wherein: the routers arranged in rows and columns; and the adaptable links include: a first plurality of adaptable links that each selectively connect routers along one of the rows; and a second plurality of adaptable links that each selectively connect routers along one of the columns. ### Claim 4 (depends on 3) The interconnection network of claim 3, further comprising: link switches along each adaptable link, each link switch being between connections to two of the routers selectively connected to the adaptable link, wherein the link controller controls the multiplexers and the link switches to selectively connect routers via the adaptable links. ### Claim 5 (depends on 4) The interconnection network of claim 4, wherein each of the link switches are configured to open and close, each open link switch along each adaptable link segmenting the adaptable link into two segments, each segment providing functionality to selectively connect two of the routers. ### Claim 6 (depends on 3) The interconnection network of claim 3, further comprising: an interposer comprising: a third plurality of adaptable links that each selectively connect routers along one of the rows; and a fourth plurality of adaptable links that each selectively connect routers along one of the columns; and a plurality of interposer switches, each interposer switch being associated one of the routers and selectively connecting the associated router with one of the third plurality of adaptable links and one of the fourth plurality of adaptable links, wherein the link controller controls the multiplexers and the interposer switches to selectively connect routers via the adaptable links. ### Claim 7 (depends on 6) The interconnection network of claim 6, further comprising: link switches along each adaptable link, each link switch being between connections to two of the routers selectively connected to the adaptable link, wherein the link controller controls the multiplexers, the interposer switches, and the link switches to selectively connect routers via the adaptable links. ### Claim 8 (depends on 7) The interconnection network of claim 7, wherein each of the link switches are configured to open and close, each open link switch along each adaptable link segmenting the adaptable link into two segments, each segment providing functionality to selectively connect two of the routers. ### Claim 9 (depends on 1) The interconnection network of claim 1, wherein the processing unit selects the topology for each partition based on characteristics and demands of a software application to be executed by the partition. ### Claim 10 (depends on 9) The interconnection network of claim 9, wherein the characteristics and demands of the software application include traffic intensiveness, memory intensiveness, network latency, throughput, or power consumption. ### Claim 11 (depends on 10) The interconnection network of claim 10, wherein, in response to a determination that the software application is not traffic intensive, the link controller reconfigures the partition to perform network concentration. ### Claim 12 (depends on 11) The interconnection network of claim 11, wherein the link controller performs network concentration by the reconfiguring the partition to form a cmesh topology or other concentrated topology where multiple cores are connected to a single router. ### Claim 13 (depends on 10) The interconnection network of claim 10, wherein, in response to a determination that the software application is traffic intensive and a determination that the software application is not memory intensive, the link controller reconfigures the partition to place express links between non-adjacent routers and form a torus topology or other express link placement that electrically connects a pair of non-adjacent routers with the most intensive communications. ### Claim 14 (depends on 10) The interconnection network of claim 10, wherein, in response to a determination that the software application is traffic intensive and determination that the software application is memory intensive, the link controller reconfigures the partition to form a tree topology or other topology supporting a one-to-many communication pattern. ### Claim 15 (depends on 10) The interconnection network of claim 10, wherein, in response to a determination that certain topology can fulfill the demands of the software application in terms of network latency, network throughput, or power consumption, the link controller reconfigures the partition to form the most beneficial topology including any regular and irregular communication paths. ### Claim 16 A method of dynamically reconfiguring an interconnection network that includes an array of routers, the method comprising: analyzing a software application to be executed by the new partition to determine whether the software application is traffic intensive; analyzing the software application to determine whether the software application is memory intensive; partitioning the interconnection network to include a new partition; and in response to a determination that the software application is not traffic intensive, reconfiguring the new partition to perform network concentration; or in response to a determination that the software application is traffic intensive and a determination that the software application is not memory intensive, reconfiguring the new partition to place express links between non-adjacent routers; or in response to a determination that the software application is traffic intensive and a determination that the software application is memory intensive, reconfiguring the new partition to form a tree topology. ### Claim 17 (depends on 16) The method of claim 16, wherein the network concentration is performed by reconfiguring the new partition to form a cmesh topology or other concentrated topology where multiple cores are connected to a single router. ### Claim 18 (depends on 16) The method of claim 16, wherein the express links are placed between non-adjacent routers by reconfiguring the new partition to form a torus topology or other express link placement that electrically connects a pair of non-adjacent routers with the most intensive communications. ### Claim 19 (depends on 16) The method of claim 16, wherein the software application is analyzed using control logic, a heuristic algorithm, or a machine learning-based algorithm. ### Claim 20 (depends on 16) The method of claim 16, further comprising: analyzing the software application to determine latency sensitivity, bandwidth sensitivity, network latency, throughput, or power consumption; reconfiguring the new partition based on the determined latency sensitivity, bandwidth sensitivity, network latency, throughput, or power consumption of the software application.