Hierarchical routing for two-way information flow and structural plasticity in neural networks

1. A network of core modules, comprising:

multiple core modules, wherein each core module comprises a core circuit that includes:

a plurality of electronic neurons;

a plurality of incoming connections for receiving incoming signals targeting the plurality of neurons, wherein each incoming connection has a predetermined address; and

a plurality of outgoing connections for transmitting outgoing signals generated by the plurality of neurons, wherein each outgoing connection targets an incoming connection in a core module among the multiple core modules; and

a routing system comprising multiple routers that selectively route signals among the multiple core modules based on a reconfigurable hierarchical organization of the core modules.

2. The network in claim 1 , wherein each core circuit of each core module further comprises:

a plurality of axons, wherein the plurality of axons include the plurality of incoming connections;

a plurality of dendrites; and

an electronic synapse array comprising multiple digital synapses interconnecting the plurality of axons with the plurality of dendrites, wherein a synapse interconnects an axon of a presynaptic neuron with a dendrite of a postsynaptic neuron;

wherein each postsynaptic neuron connects to an outgoing connection; and

wherein each neuron integrates input spikes and generates a spike event in response to the integrated input spikes exceeding a firing threshold.

3. The network of claim 1 , wherein:

the hierarchical organization of the core modules comprises a reconfigurable tree structure including a plurality of linked nodes with a core module at each node such that a core module is a root node, one or more core modules are intermediate nodes, and one or more core modules are leaf nodes, wherein:

the root node has no predecessor node and has one or more successor nodes comprising one or more of: an intermediate node and a leaf node;

at least one intermediate node has a predecessor node comprising the root node;

at least one intermediate node has a successor node comprising a leaf node; and

each leaf node has no successor nodes.

4. The network of claim 3 , wherein:

for each node, the routing system comprises:

a corresponding routing table at said node, wherein said corresponding routing table includes routing information for routing a signal originating from a source node to a destination node in the tree structure; and

a corresponding router at said node for routing a signal originating from a source node to a destination node in the tree structure based on routing information in said corresponding routing table.

5. The network of claim 4 , wherein:

for each node, a corresponding router for said node directs an incoming signal to another node based on routing information in a corresponding routing table for said node, such that:

a signal can only originate at a leaf node;

each originating leaf node selectively routes an outgoing signal to one or more of its predecessor nodes or to itself;

each receiving intermediate node selectively routes an incoming signal to one or more of: its predecessor nodes and its successor nodes;

the root node selectively routes an incoming signal to one or more of its successor nodes; and

each receiving leaf node consumes the incoming signal.

6. The network of claim 4 , wherein:

for each node, routing information in a corresponding routing table for said node is reconfigurable such that a corresponding router for said node directs an incoming signal to one or more different nodes based on reconfigured routing information in said corresponding routing table, thereby achieving structural plasticity.

7. The network of claim 6 , wherein:

for each node, routing information in a corresponding routing table for said node comprises a forwarding address for an incoming signal, wherein the forwarding address is reconfigurable.

8. The network of claim 7 , wherein:

the network approximates a fully connected network such that each neuron can target and reach any axon in the network.

9. The network of claim 7 , wherein:

the network is reversible, allowing two-way information flow such that if an originating leaf node sends a signal to a consuming leaf node via the network, then the consuming leaf node can send a signal to the originating leaf node by a retracing path.

10. A method for signal delivery in a network including multiple core modules, wherein each core module comprises a plurality of electronic neurons, a plurality of incoming connections with predetermined addresses, and a plurality of outgoing connections, the method comprising:

in each core module:

receiving incoming signals targeting the plurality of neurons of the core module via the plurality of incoming connections of the core module; and

transmitting outgoing signals generated by the plurality of neurons of the core module via the plurality of outgoing connections of the core module, wherein each outgoing connection targeting an incoming connection in a core module among the multiple core modules; and

selectively routing, via multiple routers, signals among the multiple core modules based on a reconfigurable hierarchical organization of the core modules.

11. The method of claim 10 , wherein:

each core module further comprises:

a plurality of axons, wherein the plurality of axons include the plurality of incoming connections;

a plurality of dendrites; and

an electronic synapse array comprising multiple digital synapses interconnecting the plurality of axons with the the plurality of dendrites, wherein a synapse interconnects an axon of a presynaptic neuron with a dendrite of a postsynaptic neuron; each postsynaptic neuron connects to an outgoing connection; and

the method further comprises:

each neuron integrating input spikes and generating a spike event in response to the integrated input spikes exceeding a firing threshold.

12. The method of claim 10 , wherein:

the hierarchical organization of the core modules comprises a reconfigurable tree structure including a plurality of linked nodes with a core module at each node;

the method further comprises designating the core modules in the tree structure as root, intermediate and leaf nodes, wherein a core module is a root node, one or more core modules are intermediate nodes, and one or more core modules are leaf nodes, such that:

the root node has no predecessor node and has one or more successor nodes comprising one or more of: an intermediate node and a leaf node;

at least one intermediate node has a predecessor node comprising the root node;

at least one intermediate node has a successor node comprising a leaf node; and

each leaf node has no successor nodes.

13. The method of claim 12 , wherein selective routing further comprises:

for each node:

maintaining a corresponding routing table for said node, wherein said corresponding routing table includes routing information for routing a signal originating from a source node to a destination node in the tree structure; and

providing a corresponding router at said node for routing a signal originating from a source node to a destination node in the tree structure based on routing information in said corresponding routing table.

14. The method of claim 13 , wherein:

for each node, a corresponding router for said node directs an incoming signal to another node based on routing information in a corresponding routing table for said node, such that:

a signal can only originate at a leaf node;

each originating leaf node selectively routing an outgoing signal to one or more of its predecessor nodes or to itself;

each receiving intermediate node selectively routing an incoming signal to one or more of its predecessor nodes and its successor nodes; and

the root node selectively routing an incoming signal to one or more of its successor nodes; and

each receiving leaf node consuming the incoming signal.

15. The method of claim 13 , wherein:

for each node, routing information in a corresponding routing table for said node is reconfigurable such that a corresponding router for said node directs an incoming signal to one or more different nodes based on reconfigured routing information in said corresponding routing table, thereby achieving structural plasticity.

16. The method of claim 15 , wherein:

for each node, routing information in a corresponding routing table for said node comprises a forwarding address for an incoming signal, wherein the forwarding address is reconfigurable.

17. The method of claim 16 , wherein:

the network approximates a fully connected network such that each neuron can target and reach any axon in the neural network.

18. The method of claim 16 , wherein:

the network is reversible, allowing two-way information flow such that if an originating leaf node sends a signal to a consuming leaf node via the network, then the consuming leaf node can send a signal to the originating leaf node by a retracing path.

19. A non-transitory computer-useable storage medium for signal delivery in a network including multiple core modules, wherein each core module comprises a plurality of electronic neurons, a plurality of incoming connections with predetermined addresses, and a plurality of outgoing connections, said medium having a computer-readable program, wherein the program upon being processed on a computer causes the computer to implement the steps of:

in each core module:

receiving incoming signals targeting the plurality of neurons of the core module via the plurality of incoming connections of the core module; and

transmitting outgoing signals generated by the plurality of neurons of the core module via the plurality of outgoing connections of the core module, wherein each outgoing connection targeting an incoming connection in a core module among the multiple core modules; and

selectively routing, via multiple routers, signals among the multiple core modules based on a reconfigurable hierarchical organization of the core modules.

20. The non-transitory computer-useable storage medium of claim 19 , wherein:

the hierarchical organization of the core modules comprises a reconfigurable tree structure including a plurality of linked nodes with a core module at each node; and

the program upon being processed on a computer causes the computer to further implement the steps of:

designating the core modules in the tree structure as root, intermediate and leaf nodes, wherein a core module is a root node, one or more core modules are intermediate nodes, and one or more core modules are leaf nodes, such that:

the root node has no predecessor node and has one or more successor nodes comprising one or more of: an intermediate node and a leaf node;

at least one intermediate node has a predecessor node comprising the root node;

at least one intermediate node has a successor node comprising a leaf node; and

each leaf node has no successor nodes; and

maintaining a routing table for each node for selectively routing a signal originating from a source node to a destination node in the tree structure.