Storage system and method for inference of read thresholds based on memory parameters and conditions
A storage system has an inference engine that can infer a read threshold based on a plurality of parameters of the memory. The read threshold can be used in reading a wordline in the memory during a regular read operation or as part of an error handling process. Using this machine-learning-based approach to infer a read threshold can provide significant improvement in read threshold accuracy, which can reduce bit error rate and improve latency, throughput, power consumption, and quality of service.
1. A storage system comprising:
a memory; and
a processor coupled with the memory and configured to:
train a machine-learning model of an inference engine using state-by-state cell-voltage-distribution measurements obtained at different conditions of the memory, wherein the trained machine-learning model non-linearly couples a plurality of parameters of the memory with a respective plurality of read thresholds;
use the inference engine to infer a read threshold based on current parameters of the memory; and
use the inferred read threshold in reading a wordline in the memory.
2. The storage system of claim 1 , wherein the inference engine is used in response to a read command received from a host.
3. The storage system of claim 1 , wherein the inference engine is used in response to invoking a read error handling process.
4. The storage system of claim 1 , wherein the current parameters of the memory comprise one or more of the following: time and temperature group information, temperature information, bit error rate information, program-erase count (PEC) information, physical page location information, a read threshold on a representative wordline, a data retention level of a block, and a state-by-state cell-voltage-distribution measurement.
5. The storage system of claim 1 , wherein the inference engine in configured to infer correction to the read threshold with respect to a time and temperature group read threshold.
6. The storage system of claim 1 , wherein the machine learning model is trained in an off-line process.
7. The storage system of claim 1 , wherein the processor is further configured to:
monitor a quality of the read threshold; and
re-train the machine learning model based on the monitored quality.
8. The storage system of claim 1 , wherein the inference engine is implemented as a pure hardware implementation in a controller.
9. The storage system of claim 1 , wherein the inference engine is implemented as firmware in a controller.
10. The storage system of claim 1 , wherein the processor is further configured to use the inference engine to infer a program/verify threshold.
11. The storage system of claim 1 , wherein the processor is further configured to use the inference engine to infer a soft-bit read threshold.
12. The storage system of claim 1 , wherein the processor is further configured to use the inference engine to infer a soft-bit delta value.
13. The storage system of claim 1 , wherein the memory comprises a three-dimensional memory.
14. In a storage system comprising a memory, a method comprising:
training a machine-learning model of an inference engine using state-by-state cell-voltage-distribution measurements obtained at different conditions of the memory, wherein the trained machine-learning model non-linearly couples a plurality of parameters of the memory with a respective plurality of read thresholds;
using the inference engine to infer a read threshold based on current parameters of the memory; and
using the inferred read threshold in reading a wordline in the memory.
15. The method of claim 14 , wherein the inference engine is used in response to a read command received from a host.
16. The method of claim 14 , wherein the inference engine is used as part of a read error handling (REH) process.
17. The method of claim 14 , wherein the current parameters of the memory comprise one or more of the following: time and temperature group information, temperature information, bit error rate information, program-erase count (PEC) information, physical page location information, a read threshold on a representative wordline, a data retention level of a block, and a state-by-state cell-voltage-distribution measurement.
18. The method of claim 14 , further comprising generating an inference of a program/verify threshold.
19. A storage system comprising:
a memory; and
means for:
training a machine-learning model of an inference engine using state-by-state cell-voltage-distribution measurements obtained at different conditions of the memory, wherein the trained machine-learning model non-linearly couples a plurality of parameters of the memory with a respective plurality of read thresholds;
using the inference engine to infer a read threshold based on current parameters of the memory; and
using the inferred read threshold in reading a wordline in the memory.
20. The method of claim 14 , further comprising generating an inference of one or both of a soft-bit read threshold and a soft-bit delta value.