Scaling multi-core neurosynaptic networks across chip boundaries

US9924490B2 · US · B2

Patent metadata
FieldValue
Publication numberUS-9924490-B2
Application numberUS-201314050072-A
CountryUS
Kind codeB2
Filing dateOct 9, 2013
Priority dateOct 9, 2013
Publication dateMar 20, 2018
Grant dateMar 20, 2018

How to read this patent

A practical reading order for non-experts. Skip the full description unless you need deep technical detail.

  1. Title

    What the patent document calls the invention.

  2. Abstract

    A short plain-language summary of the technical disclosure.

  3. Assignees and inventors

    Who owns or filed the patent and who is credited as inventor.

  4. Key dates

    Filing, priority, publication, and grant dates set the timeline.

  5. First independent claim

    The legal scope of protection — read this for what is actually claimed.

  6. CPC / IPC classifications

    Technology tags used to group this patent with similar filings.

  7. Citations and related patents

    Prior art links and similar publications in this corpus.

Abstract

Official abstract text for this publication.

Embodiments of the invention provide a system for scaling multi-core neurosynaptic networks. The system comprises multiple network circuits. Each network circuit comprises a plurality of neurosynaptic core circuits. Each core circuit comprises multiple electronic neurons interconnected with multiple electronic axons via a plurality of electronic synapse devices. An interconnect fabric couples the network circuits. Each network circuit has at least one network interface. Each network interface for each network circuit enables data exchange between the network circuit and another network circuit by tagging each data packet from the network circuit with corresponding routing information.

First claim

Opening claim text (preview).

What is claimed is: 1. A system for scaling multi-core neurosynaptic networks, comprising: multiple neurosynaptic network circuits, wherein each network circuit comprises a plurality of neurosynaptic core circuits; an interconnect fabric coupling said multiple network circuits; and for each network circuit, at least one corresponding network interface, wherein each network interface for each network circuit enables data exchange between said network circuit and another network circuit of said multiple network circuits by: tagging each outgoing data packet from said network circuit with a corresponding tag comprising: source information indicative of a location of a source core circuit of said network circuit that generated said data packet; and routing information including: a first distance indicative of a first number of core circuits of said another network circuit that said outgoing data packet traverses horizontally within said another network circuit en route to a target core circuit of said another network circuit, wherein said outgoing data packet propagates along one of a first horizontal direction or a second horizontal direction opposite of the first horizontal direction if the first distance is one of a positive value or a negative value, respectively; and a second distance indicative of a second number of core circuits of said another network circuit that said outgoing data packet traverses vertically within said another network circuit en route to said target core circuit of said another network circuit, wherein said outgoing data packet propagates along one of a first vertical direction or a second vertical direction opposite of the first vertical direction if the second distance is one of a positive value or a negative value, respectively; wherein each core circuit comprises multiple electronic neurons, multiple electronic axons, and a plurality of electronic synapse devices for interconnecting said multiple neurons with said multiple axons. 2. The system of claim 1 , wherein each network interface for each network circuit further enables data exchange between said network circuit and another network circuit of said multiple network circuits by: receiving incoming data packets for said network circuit; de-serializing the incoming data packets; and routing each incoming data packet to a target core circuit of said network circuit based on routing information included in a corresponding tag that said incoming data packet is tagged with, wherein said routing information comprises a horizontal distance field indicating a first number of core circuits of said network circuit that said incoming data packet traverses in an eastbound direction or a westbound direction to reach said target core circuit, and a vertical distance field indicating a second number of core circuits that said incoming data packet traverses in an eastbound direction or a westbound direction to reach said target core circuit. 3. The system of claim 2 , wherein each network interface for each network circuit further enables data exchange between said network circuit and another network circuit of said multiple network circuits by: receiving outgoing data packets from said network circuit; tagging each outgoing data packet with a corresponding tag including routing information, wherein said routing information comprises a horizontal distance field indicating a first number of core circuits of another network circuit of said multiple network circuits that said outgoing data packet traverses in an eastbound direction or a westbound direction to reach a target core circuit of said another network circuit, and a vertical distance field indicating a second number of core circuits that said outgoing data packet traverses in an eastbound direction or a westbound direction to reach said target core circuit; merging the outgoing data packets; and serializing the outgoing data packets. 4. The system of claim 3 , wherein: each network interface for each network circuit comprises: at least one funnel device for tagging and merging outgoing data packets from said network circuit; and at least one device for serializing outgoing data packets from said network circuit and de-serializing incoming data packets for said network circuit. 5. The system of claim 3 , wherein data packets are exchanged between said multiple network circuits based on address event representation. 6. The system of claim 3 , wherein: each data packet represents a neural firing event. 7. The system of claim 4 , wherein: core circuits of each network circuit are arranged as a tillable array including multiple rows and multiple columns. 8. The system of claim 7 , wherein: each funnel device for each network circuit is configured as a binary tree. 9. The system of claim 7 , wherein: each funnel device for each network circuit is configured as a binary chain. 10. The system of claim 7 , wherein: each funnel device for each network circuit is configured as both a binary tree and a binary chain. 11. A method for scaling multi-core neurosynaptic networks, comprising: tagging data packets from multiple network circuits with routing information, wherein each network circuit has at least one corresponding network interface, and each network interface for each network circuit enables data exchange between said network circuit andanother network circuit of said multiple network circuits by tagging each outgoing data packet from said network circuit with a corresponding tag comprising: source information indicative of a location of a source core circuit of said network circuit that generated said data packet; and routing information including: a first distance indicative of a first number of core circuits of said another network circuit that said outgoing data packet traverses horizontally within said another network circuit en route to a target core circuit of said another network circuit, wherein said outgoing data packet propagates along one of a first horizontal direction or a second horizontal direction opposite of the first horizontal direction if the first distance is one of a positive value or a negative value, respectively; and a second distance indicative of a second number of core circuits of said another network circuit that said outgoing data packet traverses vertically within said another network circuit en route to said target core circuit of said another network circuit, wherein said outgoing data packet propagates along one of a first vertical direction or a second vertical direction opposite of the first vertical direction if the second distance is one of a positive value or a negative value, respectively; and exchanging data packets between said multiple network circuits via an interconnect fabric coupling said multiple network circuits; wherein each network circuit comprises a plurality of neurosynaptic core circuits; and wherein each core circuit comprises multiple electronic neurons, multiple electronic axons, and a plurality of electronic synapse devices for interconnecting said multiple neurons with said multiple axons. 12. The method of claim 11 , further comprising: for each network interface for each network circuit: receiving incoming data packets for said network circuit; de-serializing the incoming data packets; and routing each incoming data packet to a target core circuit of said network circuit based on routing information included in a corresponding tag the said incoming data packet is tagged with, wherein said routing information comprises a horizontal distance field indicating a first number of core circuits of said network circuit that said incoming data packet tra

Assignees

Inventors

Classifications

  • using electronic means · CPC title

  • G06N3/049Primary

    Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs · CPC title

  • Locating users or terminals {or network equipment} for network management purposes, e.g. mobility management · CPC title

  • H04W68/00Primary

    User notification, e.g. alerting and paging, for incoming communication, change of service or the like · CPC title

Patent family

Related publications grouped by family.

External sources

Frequently asked questions

Answers are generated from the same data shown on this page.

What does patent US9924490B2 cover?
Embodiments of the invention provide a system for scaling multi-core neurosynaptic networks. The system comprises multiple network circuits. Each network circuit comprises a plurality of neurosynaptic core circuits. Each core circuit comprises multiple electronic neurons interconnected with multiple electronic axons via a plurality of electronic synapse devices. An interconnect fabric couples t…
Who is the assignee on this patent?
IBM
What technology area does this patent fall under?
Primary CPC classification G06N3/049. Mapped technology areas include Physics.
When was this patent published?
Publication date Tue Mar 20 2018 00:00:00 GMT+0000 (Coordinated Universal Time) (B2). Legal status and post-grant events are not shown on this page.
What related patents are in patentsdb?
We list 8 related publications on this page (citations in our corpus or others sharing the same primary CPC).