Input and digital output mechanisms for analog neural memory in a deep learning artificial neural network
Numerous embodiments for reading or verifying a value stored in a selected memory cell in a vector-by-matrix multiplication (VMM) array in an artificial neural network are disclosed. In one embodiment, an input comprises a set of input bits that result in a series of input signals applied to a terminal of the selected memory cell, further resulting in a series of output signals that are digitized, shifted based on the bit location of the corresponding input bit in the set of input bits, and added to yield an output indicating a value stored in the selected memory cell.
1. A method comprising:
providing a plurality of input bits to an input circuit sequentially to an array of non-volatile memory cells;
for each of the plurality of input bits:
generating, by the input circuit, an input signal in response to the input bit;
applying the input signal to a terminal of a selected non-volatile memory cell in the array;
receiving, by an output circuit from the array, an output generated in response to the input bit;
converting the output into a digital output; and
shifting the digital output based on the bit location of the input bit within the plurality of input bits to generate a shifted result; and
adding the shifted results for all of the plurality of input bits to yield a digital output.
2. The method of claim 1 , wherein the selected non-volatile memory cell is a split-gate flash memory cell.
3. The method of claim 2 , wherein the terminal is a control gate terminal.
4. The method of claim 1 , wherein the selected non-volatile memory cell is part of a neural memory.
5. The method of claim 1 , wherein the input signal comprises a pulse.
6. The method of claim 1 , wherein the input signal comprises an analog bias signal.
7. The method of claim 1 , wherein the input signal comprises an analog bias signal and a pulse.
8. The method of claim 7 , wherein one of the plurality of input bits is coded as the analog bias signal.
9. A method comprising:
providing a plurality of input bits to an input circuit sequentially to an array of non-volatile memory cells; and
for each of the plurality of input bits:
generating, by the input circuit, an input signal in response to the input bit, wherein the input signal comprises an analog bias level component and a pulsewidth component;
applying the input signal to a terminal of a selected non-volatile memory cell in the array;
receiving, by an output circuit, an output generated in response to the input bit; and
converting the output into a digital output.
10. The method of claim 9 , further comprising:
shifting the digital output based on the bit location of the input bit within the plurality of input bits to generate a shifted result; and
adding the shifted results for all of the plurality of input bits to yield a digital output.
11. The method of claim 9 , wherein the output is generated using a current summer.
12. The method of claim 9 , wherein the analog bias level component is a binary weighted value.
13. The method of claim 9 , wherein the output value is generated based only on the pulsewidth modulation input.
14. The method of claim 9 , wherein the selected non-volatile memory cell is part of a neural memory.
15. A method comprising:
providing a plurality of input bits to an input circuit sequentially to an array of non-volatile memory cells; and
for each of the plurality of input bits:
generating, by the input circuit, an input signal in response to the multi-bits of the input bits, wherein the input signal is an analog bias level;
applying the input signal to a terminal of a selected non-volatile memory cell in the array; and
receiving, by an output circuit, an output generated in response to the input bit.
16. The method of claim 15 , further comprising:
converting the output into a digital output.
17. The method of claim 16 , further comprising:
adding the results for all of the plurality of input bits to yield a digital output.
18. The method of claim 16 , further comprising:
shifting the digital output based on the bit location of the input bit within the plurality of input bits to generate a shifted result; and
adding the shifted results for all of the plurality of input bits to yield a digital output.
19. The method of claim 17 , wherein the output is generated using a current summer.
20. The method of claim 15 , wherein the analog bias level is binary indexed value.
21. The method of claim 15 , wherein the analog bias level is such that resulting cell output value is output value is generated based only on the pulsewidth modulation input.
22. The method of claim 15 , wherein the selected non-volatile memory cell is part of a neural memory.