Output array neuron conversion and calibration for analog neural memory in deep learning artificial neural network
Configurable input blocks and output blocks and physical layouts are disclosed for analog neural memory systems that utilize non-volatile memory cells. An input block can be configured to support different numbers of arrays arranged in a horizontal direction, and an output block can be configured to support different numbers of arrays arranged in a vertical direction. Adjustable components are disclosed for use in the configurable input blocks and output blocks. Systems and methods are utilized for compensating for leakage and offset in the input blocks and output blocks the in analog neural memory systems.
1. A method, comprising:
measuring leakage current from one or more decoding circuits and one or more column write circuits for use with a non-volatile memory array and storing a measured leakage current value in a register;
retrieving the leakage current value from the register; and
converting a neuron output from the non-volatile memory array into a digital value and subtracting the leakage current value from the digital value.
2. A method comprising:
measuring leakage current from one or more decoding circuits and one or more write circuits for use with an array of analog neural memory cells using an analog-to-digital converter to generate a digital leakage current value;
storing the digital leakage current value as a first value in a counter; and
generating an output from a neuron output from the array using the counter and the first value, wherein the output is equal to the neuron output minus the first value.
3. The method in claim 2 , wherein the generating comprises converting a current from the array of analog neural memory cells into the output by counting down on the counter from the stored first value until the counter reaches zero and then counting up to generate the output.
4. The method in claim 2 , wherein the generating comprises measuring a current from the array of analog neural memory cells using the counter and then subtracting the stored first value from the measured current to generate the output.
5. The method of claim 2 , wherein the analog-to-digital converter comprises an integrating analog-to-digital converter.
6. The method of claim 2 , wherein the analog-to-digital converter comprises a ramp analog-to-digital converter.
7. The method of claim 2 , wherein the analog-to-digital converter comprises an algorithmic analog-to-digital converter.
8. The method of claim 2 , wherein the analog-to-digital converter comprises a sigma delta analog-to-digital converter.
9. The method of claim 2 , wherein the analog-to-digital converter comprises a successive approximation register analog-to-digital converter.
10. The method of claim 2 , further comprising:
converting the output into a voltage.
11. The method of claim 2 , further comprising:
converting the output into one or more pulses where the width of the one or more pulses is proportional to the value of the output.
12. The method of claim 2 , wherein the analog neural memory cells are split-gate flash memory cells.
13. The method of claim 2 , wherein the analog neural memory cells are stacked-gate flash memory cells.