# US 10,482,372 B2 · Interconnection scheme for reconfigurable neuromorphic hardware

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 10,482,372 B2
- Document ID: US10482372B2
- Status: Granted
- Kind: B2
- Application number: 14/757,397
- Filed: 2015-12-23
- Published: 2017-06-29
- Granted: 2019-11-19
- Earliest claimed: 2015-12-23
- Assignees: Intel Corporation
- Applicants: Intel Corporation
- Inventors: Gregory K. Chen; Jae-Sun Seo
- Web page: https://patentgrep.com/patent/US10482372B2

## CPC

- G06N 3/065: Analogue means
- G06F 13/4282: {on a serial bus, e.g. I2C bus, SPI bus (on daisy chain buses G06F13/4247)}
- G06N 3/02: Neural networks
- G06N 3/0499: Feedforward networks
- G06N 3/063: using electronic means
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

Systems and methods for an interconnection scheme for reconfigurable neuromorphic hardware are disclosed. A neuromorphic processor may include a plurality of corelets, each corelet may include a plurality of synapse arrays and a neuron array. Each synapse array may include a plurality of synapses and a synapse array router coupled to synapse outputs in a synapse array. Each synapse may include a synapse input, synapse output; and a synapse memory. A neuron array may include a plurality of neurons, each neuron may include a neuron input and a neuron output. Each synapse array router may include a first logic to route one or more of the synapse outputs to one or more of the neuron inputs.

## Claims

### Claim 1

A neuromorphic processor, comprising:
a plurality of corelets, each corelet to include:
a plurality of synapse arrays, each synapse array to include:
a plurality of synapses, each synapse to include:
a synapse input;
a synapse output; and
a synapse memory to store a synapse destination of the corresponding synapse and a synapse weight for the corresponding synapse;
a synapse array router coupled to the synapse outputs in the synapse array; and
a neuron array to include a plurality of neurons, each neuron to include a neuron input and a neuron output;
wherein at least one of the synapse array routers includes circuitry to:
read at least one synapse memory;
identify a corresponding synapse destination;
identify a corresponding synapse weight; and
transmit a packet including the synapse weight and the corresponding synapse destination to a particular neuron group router, based on a routing algorithm.

### Claim 2 (depends on 1)

The neuromorphic processor of claim 1, wherein a first synapse array router further includes circuitry to route a synapse output to a neuron input of a neuron in a different corelet.

### Claim 3 (depends on 1)

The neuromorphic processor of claim 1, wherein the neuron group router includes circuitry to:
receive the packet;
identify the corresponding synapse destination;
identify the corresponding synapse weight; and
transmit the corresponding synapse weight to at least one of the neuron inputs.

### Claim 4 (depends on 1)

The neuromorphic processor of claim 1, wherein in one of the corelets the number of synapses to be routed to one of the neuron inputs is greater than the number of synapses in the corelet.

### Claim 5 (depends on 1)

The neuromorphic processor of claim 1, further comprising a processor including logic to write initial synapse destinations and initial synapse weights into the synapse memory of each of the plurality of synapses in a first corelet, wherein at least one of the synapse destinations is to a neuron array in a second corelet.

### Claim 6

A reconfigurable neuromorphic processor logic unit, comprising:
a plurality of corelets, each corelet to include:
a plurality of synapse arrays, each synapse array to include:
a plurality of synapses, each synapse to include:
a synapse input;
a synapse output; and
a synapse memory to store a synapse destination of the corresponding synapse and a synapse weight for the corresponding synapse;
a synapse array router, the synapse array router coupled to the synapse outputs in the synapse array;
a neuron array, including a plurality of neurons, each of the plurality of neurons including a neuron input and a neuron output;
wherein each synapse array router includes circuitry to:
read at least one synapse memory;
identify a corresponding synapse destination;
identify a corresponding synapse weight; and
transmit a packet including the synapse weight and the corresponding synapse destination to a particular neuron group router, based on a routing algorithm.

### Claim 7 (depends on 6)

The reconfigurable neuromorphic processor logic unit of claim 6, wherein a first synapse array router includes circuitry to route a synapse output to a neuron input of a neuron in a different corelet.

### Claim 8 (depends on 6)

The reconfigurable neuromorphic processor logic unit of claim 6, wherein the neuron group router includes circuitry to:
receive the packet;
identify the corresponding synapse destination;
identify the corresponding synapse weight; and
transmit the corresponding synapse weight to at least one of the neuron inputs.

### Claim 9 (depends on 6)

The reconfigurable neuromorphic processor logic unit of claim 6, wherein in one of the corelets the number of synapses to be routed to one of the neuron inputs is greater than the number of synapses in the corelet.

### Claim 10 (depends on 6)

The reconfigurable neuromorphic processor logic unit of claim 6, further comprising a processor including logic to write initial synapse destinations and initial synapse weights into the synapse memory of each of the plurality of synapses in a first corelet, wherein at least one of the synapse destinations is to a neuron array in a second corelet.