Retention drift correction in non-volatile memory arrays
AssigneeApplied Materials, Inc.
InventorChristophe J. Chevallier
Life of this patent
Estimating the term…
Abstract
Methods and architectures for refreshing memory elements in a memory array may initialize a reference array that stores each of the possible values stored in the memory element. The values in the memory array and the reference array will drift in parallel over time. To perform a refresh, the drifted values may be read from the reference array and mapped to the original values that were stored when the reference array was initialized. Next, each value may be read from the memory array and matched with a corresponding value from the reference array. The known original value stored in the reference array can then be used to refresh the corresponding memory element in the memory array.
Claims20
A method of refreshing memory elements in memory arrays, the method comprising:
- causing a value to be read from a memory element in a memory array;
- determining a plurality of reference values from a reference array;
- identifying a reference value in the plurality of reference values that corresponds to the value read from the memory element;
- determining an original value that was stored in at least one reference element in the reference array from which the reference value was determined;
- and
- causing the original value to be written to the memory element in the memory array.
A memory system comprising:
- a memory array comprising memory elements that are configured to store N different values, wherein N is an integer value greater or equal to 3;
- a reference array comprising N sets of reference elements, wherein each of the N sets of reference elements stores one of the N different values;
- and
- a controller configured to perform operations comprising:
- reading and writing values for the memory array and the reference array, wherein the memory elements are used as weights for a neural network;
- and
- maintaining a data structure that maps the N different values originally stored in the N sets of reference elements to N values that are later read from the N sets of reference elements.
Dependencies are inferred from the claim text.