Self-evaluating array of memory

• US 10,607,715 B2
• Filed: 06/13/2017
• Issued: 03/31/2020
• Est. Priority Date: 06/13/2017
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Buyuktosunoglu, Alper, Venkataramani, Swagath, Joshi, Rajiv, Swaminathan, Karthik V., Eldridge, Schuyler, Bose, Pradip
• Primary Examiner(s): Lamarre, Guy J
• Application Number:
  US15/621,281
• Publication Number:
  US 20180358110A1
• Time in Patent Office: 1,022 Days
• CPC Class Codes: G06N 3/04 Architecture, e.g. interconnection topology
• CPC Class Codes: G06N 3/063 using electronic means
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G11C 2029/5004 Voltage
• CPC Class Codes: G11C 29/50004 of threshold voltage

Abstract

A first voltage may be applied to a memory in a neural network. The memory may include one or more memory cells. A processor may determine that a first memory cell in the memory is faulty at the first voltage. The first voltage may be a low voltage. The processor may identify a first factor in the neural network. The first factor may have a low criticality in the neural network. The processor may determine to store the first factor in the first memory cell. The processor may store the first factor in the first memory cell.

Claims

1. A computer-implemented method comprising:

applying, by a memory controller, a first voltage to a memory in a neural network, wherein the memory includes one or more memory cells;

determining that a first memory cell in the memory is faulty at the first voltage, wherein the first voltage is a minimum operating voltage for the memory, and wherein the first memory cell is determined to be faulty when the first memory cell does not transition a logic value at the first voltage;

identifying a first factor in the neural network, wherein the first factor has a low criticality in the neural network;

determining to store the first factor in the first memory cell;

storing the first factor directly in the first memory cell; and

operating all of the one or more memory cells at the first voltage.

2. The method of claim 1, wherein determining to store the first factor in the first memory cell further comprises:

applying a second voltage to the memory in the neural network; and

determining that the first memory cell in the memory is faulty at the second voltage, wherein the second voltage is above the first voltage.

3. The method of claim 2, wherein applying the second voltage to the memory in the neural network includes applying the first and second voltage to the memory before a computation is performed by the neural network.

4. The method of claim 1, further comprising:

determining that a second memory cell in the memory is not faulty at the first voltage;

identifying a second factor in the neural network, wherein the second factor has a high criticality in the neural network;

determining to store the second factor in the second memory cell; and

storing the second factor directly in the second memory cell.

5. The method of claim 4, wherein determining to store the second factor in the second memory cell further comprises:

applying a second voltage to the memory in the neural network; and

determining that the second memory cell in the memory is not faulty at the second voltage, wherein the second voltage is above the first voltage.

6. The method of claim 5, further comprising:

determining that a third memory cell in the memory is not faulty at the first voltage;

determining that the third memory cell in the memory is faulty at the second voltage;

identifying a third factor in the neural network, wherein the third factor has a standard criticality in the neural network, wherein the standard criticality is below the high criticality and above the low criticality;

determining to store the third factor in the third memory cell; and

storing the third factor directly in the third memory cell.

7. The method of claim 5, further comprising:

determining that a third memory cell in the memory is faulty at the first voltage;

determining that the third memory cell in the memory is not faulty at the second voltage;

identifying a third factor in the neural network, wherein the third factor has a low criticality in the neural network;

determining to store the third factor in the third memory cell; and

storing the third factor directly in the third memory cell.

8. The method of claim 1, further comprising:

applying a second voltage to all of the one or more memory cells included in the memory, wherein each of the one or more memory cells includes a respective logic value;

determining which memory cells of the one or more memory cells are faulty by identifying which memory cells do not transition their respective logic value to a second logic value at the second voltage;

determining which memory cells of the one or more memory cells are not faulty by identifying which memory cells do transition their respective logic value to the second logic value at the second voltage;

analyzing one or more factors in the neural network;

categorizing each factor as having the low criticality, a standard criticality, or a high criticality,

wherein the low criticality is determined by analyzing historical outputs of the neural network and identifying the factor as being below a first threshold,

wherein the standard criticality is determined by analyzing the historical outputs of the neural network and identifying the factor as being between the first threshold and a second threshold, and wherein the second threshold is higher than the first threshold, and

wherein the high criticality is determined by analyzing the historical outputs of the neural network and identifying the factor as being above the second threshold; and

assigning each factor to a specific memory cell of the one or more memory cells based on the criticality of the factor and the fault of the specific memory cell, wherein low criticality factors are assigned to faulty memory cells, and wherein standard criticality and high criticality factors are assigned to non-faulty memory cells.

9. The method of claim 1, wherein applying the first voltage to the memory in the neural network includes identifying that the neural network is performing a computation, and wherein the computation includes the first factor.

10. The method of claim 9, wherein the computation is a first test pattern, the first test pattern including the first memory cell performing a read operation and a write operation, and wherein the read operation and the write operation are associated with the logic value.

11. A system comprising:

a memory; and

a processor in communication with the memory, the processor being configured to perform operations comprising;

applying, by a memory controller, a first voltage to a memory in a neural network, wherein the memory includes one or more memory cells;

determining that a first memory cell in the memory is faulty at the first voltage, wherein the first voltage is a minimum operating voltage for the memory, and wherein the first memory cell is determined to be faulty when the first memory cell does not transition a logic value at the first voltage;

identifying a first factor in the neural network, wherein the first factor has a low criticality in the neural network;

determining to store the first factor in the first memory cell;

storing the first factor directly in the first memory cell, and

operating all of the one or more memory cells at the first voltage.

12. The system of claim 11, wherein the operations further comprise:

determining that a second memory cell in the memory is not faulty at the first voltage;

identifying a second factor in the neural network, wherein the second factor has a high criticality in the neural network;

determining to store the second factor in the second memory cell; and

storing the second factor directly in the second memory cell.

13. The system of claim 12, wherein determining to store the second factor in the second memory cell further comprises:

applying a second voltage to the memory in the neural network; and

determining that the second memory cell in the memory is not faulty at the second voltage, wherein the second voltage is above the first voltage.

14. The system of claim 13, wherein the operations further comprise:

determining that a third memory cell in the memory is not faulty at the first voltage;

determining that the third memory cell in the memory is faulty at the second voltage;

identifying a third factor in the neural network, wherein the third factor has a standard criticality in the neural network, wherein the standard criticality is below the high criticality and above the low criticality;

determining to store the third factor in the third memory cell; and

storing the third factor directly in the third memory cell.

15. The system of claim 13, wherein the operations further comprise:

determining that a third memory cell in the memory is faulty at the first voltage;

determining that the third memory cell in the memory is not faulty at the second voltage;

identifying a third factor in the neural network, wherein the third factor has a low criticality in the neural network;

determining to store the third factor in the third memory cell; and

storing the third factor directly in the third memory cell.

16. The system of claim 11, further comprising:

applying a second voltage to all of the one or more memory cells included in the memory, wherein each of the one or more memory cells includes a respective logic value;

determining which memory cells of the one or more memory cells are faulty by identifying which memory cells do not transition their respective logic value to a second logic value at the second voltage;

determining which memory cells of the one or more memory cells are not faulty by identifying which memory cells do transition their respective logic value to the second logic value at the second voltage;

analyzing one or more factors in the neural network;

categorizing each factor as having the low criticality, a standard criticality, or a high criticality,

wherein the low criticality is determined by analyzing historical outputs of the neural network and identifying the factor as being below a first threshold,

wherein the standard criticality is determined by analyzing the historical outputs of the neural network and identifying the factor as being between the first threshold and a second threshold, and wherein the second threshold is higher than the first threshold, and

wherein the high criticality is determined by analyzing the historical outputs of the neural network and identifying the factor as being above the second threshold; and

assigning each factor to a specific memory cell of the one or more memory cells based on the criticality of the factor and the fault of the specific memory cell, wherein low criticality factors are assigned to faulty memory cells, and wherein standard criticality and high criticality factors are assigned to non-faulty memory cells.

17. The system of claim 11, wherein determining to store the first factor in the first memory cell further comprises:

applying a second voltage to the memory in the neural network; and

determining that the first memory cell in the memory is faulty at the second voltage, wherein the second voltage is above the first voltage.