Granular refresh rate control for memory devices based on bit position
View Patent ↗A system and method for refreshing memory cells of a memory device includes storing each bit of a B-bit word in a different sub-array of a memory device. Each of the bits is associated with a bit position, and the memory device includes a plurality of sub-arrays. The system and method also include determining a refresh interval for a plurality of the bit positions based upon a relative importance of the plurality of the bit positions to a performance of a machine learning or signal processing task involving the B-bit word. The refresh interval is based upon a fidelity metric and a resource metric. The system and method further include refreshing the plurality of sub-arrays based upon the refresh interval determined for the plurality of bit positions, and dynamically updating the refresh interval for the plurality of bit positions upon receiving a new fidelity metric or a new resource metric.
1. A method comprising:
storing each bit of a B-bit word in a different sub-array of a memory device, wherein each of the bits is associated with a bit position in the B-bit word and each of the bits is stored in a respective sub-array according to its respective bit position with the most significant bit in a first sub-array and the least significant bit in a second sub-array, and wherein the memory device comprises a plurality of sub-arrays;
determining a refresh interval for a plurality of the bit positions including a first refresh interval for the most significant bit and a second refresh interval for the least significant bit based upon a relative importance of the plurality of the bit positions to a performance of a machine learning or signal processing task involving the B-bit word, wherein the refresh interval is based upon a fidelity metric and a resource metric, the first refresh interval is shorter than the second refresh interval;
refreshing the plurality of sub-arrays based upon the refresh interval determined for the plurality of bit positions, wherein refreshing an individual sub-array of the plurality of sub-arrays includes reading data from memory cells of the individual sub-array and writing the data back to the memory cells; and
dynamically updating the refresh interval for the plurality of bit positions upon receiving a new fidelity metric or a new resource metric.
2. The method of claim 1 , wherein the refresh interval is optimized to minimize the fidelity metric while satisfying the resource metric.
3. The method of claim 1 , wherein the refresh interval is optimized to minimize the resource metric while satisfying the fidelity metric.
4. The method of claim 1 , wherein the refresh interval is optimized to satisfy a granularity.
5. The method of claim 1 , further comprising identifying the relative importance of the plurality of bit positions before determining the refresh interval.
6. The method of claim 1 , wherein a most significant bit of the B-bit word is more important than a least significant bit of the B-bit word.
7. The method of claim 1 , wherein the refresh interval of the most significant bit and the refresh interval of the least significant bit are multiples of a value of a refresh interval parameter.
8. The method of claim 1 , wherein the bit position that is of greater relative importance is assigned a shorter refresh interval than the bit position that is of lesser relative importance.
9. The method of claim 1 , wherein the memory device comprises a dynamic random access memory device.
10. The method of claim 1 , wherein the fidelity metric comprises at least one of mean square error or peak signal-to-noise ratio.
11. The method of claim 1 , wherein the resource metric comprises refresh power consumption.
12. The method of claim 1 , wherein dynamically updating of the refresh interval occurs in real-time or substantial real-time.
13. The method of claim 1 , wherein determining the refresh interval comprises inputting the fidelity metric or the resource metric into a look-up table, and receiving the refresh interval as an output from the look-up table.
14. A system comprising:
a memory controller configured to connect with each of a plurality of sub-arrays of a memory device, wherein the memory controller comprises programmed instructions to:
receive a plurality of B-bit words from a machine learning or signal processing application;
store one bit of each B-bit word in each of the plurality of sub-arrays such that a most significant bit of each B-bit word is stored in a first sub-array and a least significant bit of each B-bit word is stored in a second sub-array;
determine a first refresh interval for the most significant bit of the plurality of B-bit words and a second refresh interval for the least significant bit of the plurality of B-bit words, wherein the first refresh interval is less than the second refresh interval, and wherein the first refresh interval and the second refresh interval are determined to either minimize a fidelity metric or minimize a resource metric; and
refresh the first sub-array of the plurality of sub-arrays storing the most significant bit based on the first refresh interval and refresh the second sub-array of the plurality of sub-arrays storing the least significant bit based on the second refresh interval.
15. The system of claim 14 , wherein the memory device comprises a dynamic random access memory module comprising a plurality of dynamic random access memory chips, and wherein each of the plurality of dynamic random access memory chips comprises a plurality of banks.
16. The system of claim 15 , wherein each of the plurality of sub-arrays comprises one of the plurality of banks of one of the plurality of dynamic random access memory chips.
17. The system of claim 15 , wherein each of the plurality of sub-arrays comprises one of the plurality of dynamic random access memory chips.
18. The system of claim 14 , wherein the memory controller further comprises programmed instructions to dynamically update the first refresh interval and the second refresh interval upon receiving a new fidelity metric or a new resource metric.
19. A non-transitory computer-readable medium comprising computer-readable instructions stored thereon that when executed by a processor associated with a dynamic random access memory module causes the processor to:
receive a fidelity metric;
receive a resource metric;
receive a granularity;
receive a plurality of B-bit words from a machine learning or signal processing application;
store the plurality of B-bit words in a plurality of sub-arrays such that a most significant bit of each B-bit word is stored in a first sub-array and a least significant bit of each B-bit word is stored in a second sub-array;
determine a refresh interval for a plurality of bit positions of the B-bit word to satisfy the granularity and the resource metric while minimizing the fidelity metric, wherein the refresh interval for a more important bit position to a performance of a machine learning or signal processing task involving the B-bit word is less than the refresh interval for a lesser important bit position; and
dynamically update the refresh interval of the plurality of bit positions upon receiving a new fidelity metric, a new resource metric, or a new granularity.
20. The non-transitory computer-readable medium of claim 19 , wherein the more important bit position is the most significant bit, and wherein the lesser important bit position is the least significant bit.
21. The non-transitory computer-readable medium of claim 19 , wherein the refresh interval is a multiple of a value of a refresh interval parameter.