METHODS AND SYSTEMS FOR REPLACEABLE SYNAPTIC WEIGHT STORAGE IN NEURO-PROCESSORS

• US 20120011087A1
• Filed: 07/07/2010
• Published: 01/12/2012
• Est. Priority Date: 07/07/2010
• Status: Active Grant

• Current Assignee: Qualcomm, Inc.
• Original Assignee: Qualcomm, Inc.
• Inventors: Aparin, Vladimir
• Granted Patent: US 8,676,734 B2
• Time in Patent Office: Days
• US Class Current: 706/25
• CPC Class Codes: G06N 3/063 using electronic means

Abstract

Certain embodiments of the present disclosure support techniques for storing synaptic weights separately from a neuro-processor chip into a replaceable storage. The replaceable synaptic memory gives a unique functionality to the neuro-processor and improves its flexibility for supporting a large variety of applications. In addition, the replaceable synaptic storage can provide more choices for the type of memory used, and might decrease the area and implementation cost of the overall neuro-processor chip.

Claims

1. An electrical circuit, comprising:

a neuro-processor chip with a plurality of neuron circuits and at least one synapse, wherein the at least one synapse connects a pair of neuron circuits; and

a removable memory connected to the neuro-processor chip storing weights of the at least one synapse, wherein the weights define, at least in part, a function of the neuro-processor chip.

2. The electrical circuit of claim 1, wherein:

the removable memory is connected to the neuro-processor chip via an interface circuit, and

the interface circuit carries the weights from the neuro-processor chip to the removable memory and from the removable memory to the neuro-processor chip.

3. The electrical circuit of claim 1, wherein the neuro-processor chip comprises a local memory for storing at least a portion of the weights.

4. The electrical circuit of claim 1, wherein the weights are trained for the pair of neuron circuits before being stored in the removable memory.

5. The electrical circuit of claim 4, wherein:

values of the trained weights are replicated and stored in another removable memory connected to another neuro-processor chip, and

the other neuro-processor chip executes the function of the neuro-processor chip based at least in part on the values of the weights.

6. The electrical circuit of claim 1, wherein:

the removable memory is replaced with another removable memory that stores different values of the weights than the removable memory, and

the values of the weights define, at least in part, another function of the neuro-processor chip.

7. The electrical circuit of claim 1, wherein the removable memory comprises a non-volatile memory device.

8. A method for implementing a neural system, comprising:

connecting a removable memory to a neuro-processor chip; and

storing synapse weights on the removable memory, wherein a synapse connects two of a plurality of neuron circuits of a neuro-processor chip, and wherein the weights define, at least in part, a function of the neuro-processor chip.

9. The method of claim 8, further comprising:

connecting the removable memory to the neuro-processor chip using an interface circuit, and

transferring the weights from the neuro-processor chip to the removable memory, and from the removable memory to the neuro-processor chip, via the interface circuit.

10. The method of claim 8, further comprising:

storing at least a portion of the synapse weights on a local memory within the neuro-processor chip.

11. The method of claim 8, further comprising:

training the weights for the two neuron circuits; and

storing the trained weights in the removable memory.

12. The method of claim 11, further comprising:

replicating values of the trained weights to another removable memory connected to another neuro-processor chip, wherein

the other neuro-processor chip executes the function of the neuro-processor chip based at least in part on the values of the weights.

13. The method of claim 8, further comprising:

replacing the removable memory with another removable memory that stores different values of the weights than the removable memory, wherein

the values of the weights define, at least in part, another function of the neuro-processor chip.

14. The method of claim 8, wherein the removable memory comprises a non-volatile memory device.

15. An apparatus for implementing a neural system, comprising:

means for connecting a removable memory to a neuro-processor chip; and

means for storing synapse weights on the removable memory, wherein a synapse connects two of a plurality of neuron circuits of a neuro-processor chip, and wherein the weights define, at least in part, a function of the neuro-processor chip.

16. The apparatus of claim 15, further comprising:

means for connecting the removable memory to the neuro-processor chip using an interface circuit, and

means for transferring the weights from the neuro-processor chip to the removable memory, and from the removable memory to the neuro-processor chip, via the interface circuit.

17. The apparatus of claim 15, further comprising:

means for storing at least a portion of the synapse weights on a local memory within the neuro-processor chip.

18. The apparatus of claim 15, further comprising:

means for training the weights for the two neuron circuits; and

means for storing the trained weights in the removable memory.

19. The apparatus of claim 18, further comprising:

means for replicating values of the trained weights to another removable memory connected to another neuro-processor chip, wherein

the other neuro-processor chip executes the function of the neuro-processor chip based at least in part on the values of the weights.

20. The apparatus of claim 15, further comprising:

means for replacing the removable memory with another removable memory that stores different values of the weights than the removable memory, wherein

the values of the weights define, at least in part, another function of the neuro-processor chip.

21. The apparatus of claim 15, wherein the removable memory comprises a non-volatile memory device.