Current compensation block and method for programming analog neural memory in deep learning artificial neural network
View Patent ↗Numerous embodiments are disclosed for a high voltage generation algorithm and system for generating high voltages necessary for a particular programming operation in analog neural memory used in a deep learning artificial neural network. Different calibration algorithms and systems are also disclosed. Optionally, compensation measures can be utilized that compensate for changes in voltage or current as the number of cells being programmed changes.
1. A system for compensating for changes in current applied by a high voltage generator to a memory array in an artificial neural network during a programming operation, the system comprising:
a vector by matrix multiplication array of non-volatile memory cells, wherein the cells are arranged in columns and rows, wherein each column is coupled to a bit line; and
a high voltage generator for selectively applying a high voltage to the array during a programming operation;
wherein the columns are arranged into sets of columns and each set of columns comprises a column coupled to a bit line that is a dummy bit line that is turned on when one or more other bit lines in the set are not turned on during a programming operation to reduce variance in current drawn during programming operations.
2. The system of claim 1 , wherein each set in the sets of columns consists of 32 bit lines.
3. The system of claim 1 , wherein the non-volatile memory cells are split-gate flash memory cells.
4. The system of claim 2 , wherein the non-volatile memory cells are split-gate flash memory cells.
5. A method for compensating for changes in current applied by a high voltage generator to a vector by matrix multiplication array of non-volatile memory cells in an artificial neural network during a programming operation, wherein the cells in the array are arranged in columns and rows and each column is coupled to a bit line and the columns are arranged into sets of columns and each set of columns comprises a column coupled to a bit line that is a dummy bit line, the method comprising:
selectively applying a high voltage to the vector by matrix multiplication array during a programming operation;
turning on a dummy bit line in a set of columns when one or more other bit lines in the set are not turned on during the programming operation to reduce variance in current drawn during programming operations.
6. The method of claim 5 , wherein each set in the sets of columns consists of 32 bit lines.
7. The method of claim 5 , wherein the non-volatile memory cells are split-gate flash memory cells.
8. The method of claim 6 , wherein the non-volatile memory cells are split-gate flash memory cells.