Dynamic modification of programming duration based on number of cells to be programmed in analog neural memory array in deep learning artificial neural network
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, the duration of a programming voltage can change as the number of cells to be programmed changes.
1. A method for generating a high voltage output in response to the number of cells to be programmed in a memory array in a single programming operation, the method comprising:
determining the number of cells in a selected plurality of cells to be programmed;
determining a first programming duration to be used in programming the selected plurality of cells, wherein the first programming duration varies depending on the number of cells in the selected plurality of cells to be programmed;
providing a first voltage during the first programming duration, wherein the first voltage is applied concurrently to all of for the selected plurality of cells during at least part of the first programming duration;
verifying which cells in the selected plurality of cells have been correctly programmed;
determining a second programming duration to be used in programming a subset of cells, the subset of cells comprising the cells in the plurality of cells that have not been correctly programmed, wherein the second programming duration varies depending on the number of cells in the subset of cells to be programmed; and
providing a second voltage during the second programming duration, wherein the second voltage is applied concurrently to all of the subset of cells during at least part of the second programming duration.
2. The method of claim 1 , further comprising:
verifying which cells in the subset of cells have been correctly programmed;
determining a third programming duration to be used in programming a second subset of cells, the second subset of cells comprising the cells in the subset of cells that have not been correctly programmed; and
applying a third voltage during the third programming duration, wherein the third voltage is applied concurrently to all of the second subset of cells during at least part of the third programming duration.
3. The method of claim 1 , wherein the second programming duration is shorter than the first programming duration if the number of cells in the subset of cells is less than the number of cells in the selected plurality of cells.
4. The method of claim 1 , wherein the memory array is a vector-by-matrix-multiplication array.
5. The method of claim 4 , wherein the vector-by-matrix-multiplication array provides a neuron read out on a bit line of the memory array.
6. The method of claim 4 , wherein the vector-by-matrix-multiplication array provides a neuron read out on a source line of the memory array.
7. The method of claim 4 , wherein the memory cells in the vector-by-matrix-multiplication array perform weight multiplication.
8. The method of claim 4 , wherein the memory cells in the vector-by-matrix-multiplication array perform synapses addition.
9. The method of claim 2 , wherein the memory array is a vector-by-matrix-multiplication array.
10. The method of claim 9 , wherein the vector-by-matrix-multiplication array provides a neuron read out on a bit line of the memory array.
11. The method of claim 9 , wherein the vector-by-matrix-multiplication array provides a neuron read out on a source line of the memory array.
12. The method of claim 9 , wherein the memory cells in the vector matrix multiplier perform weight multiplication.
13. The method of claim 9 , wherein the memory cells in the vector-by-matrix-multiplication array perform synapses addition.