Efficient reset and evaluation operation of multiplying bit-cells for in-memory computing
Various embodiments comprise systems, methods, architectures, mechanisms, apparatus, and improvements thereof for in-memory computing using charge-domain circuit operation to provide energy efficient, high speed, capacitor-based in-memory computing. Various embodiments contemplate controlling input signal presentation within in-memory computing structures/macros in accordance with predefined or dynamic switch selection criteria to reduce energy consumption associated with charging and/or discharging summing capacitors during reset and evaluation operating modes of multiplying bit-cells (M-BCs).
1. An apparatus for in-memory computing using charge-domain circuit operation, comprising:
a first plurality of transistors configured as memory bit cells;
a second plurality of transistors configured to perform in-memory computing of N-bit input vector elements using the memory bit cells to provide bit cell output signals;
a plurality of capacitors configured to store a result of in-memory computing from the memory bit cells, each of said plurality of capacitors having a respective output terminal coupled to a compute line (CL) to impart thereto a respective charge, said CL indicative of a computational result; and
a plurality of switches configured during a reset mode of operation to selectively couple capacitor input terminals to respective first voltages, and during an evaluation mode of operation to selectively couple capacitor input terminals to respective second voltages;
wherein one of the first and second voltages is a fixed voltage, and the other of the first and second voltages is determined using the N-bit input vector element.
2. The apparatus of claim 1 , wherein the sign of the N-bit input vector element determines whether the first voltage is a fixed voltage or a voltage determined using the N-bit input vector element.
3. The apparatus of claim 1 , wherein the fixed voltage comprises zero.
4. The apparatus of claim 1 , wherein the fixed voltage comprises a predefined portion of a maximum voltage.
5. The apparatus of claim 1 , wherein a voltage determined using the N-bit input vector element comprises an absolute value of the voltage represented by the N-bit input vector element.
6. The apparatus of claim 1 , wherein the apparatus comprises a column of M multiplying bit cells (M-BCs), where M is an integer greater than one.
7. The apparatus of claim 6 , wherein N columns of M M-BCs form an N×M array of M-BCs, where N is an integer greater than one.
8. The apparatus of claim 1 , wherein the capacitors comprise metal-fringing capacitors.
9. The apparatus of claim 1 , wherein the fixed voltage is selected as a voltage level determined in accordance with expected N-bit input vector elements.
10. The apparatus of claim 9 , wherein the determined voltage level corresponds to a mean or median of an expected distribution of N-bit input vector elements.
11. The apparatus of claim 9 , wherein the determined voltage level corresponds to a reverse polarity mean or median of an expected distribution of N-bit input vector elements.
12. The apparatus of claim 1 , wherein the first plurality of transistors is configured to store a plurality of matrix values.
13. The apparatus of claim 12 , wherein the second plurality of transistors is configured to receive an input signal, perform analog charge-domain computations using the input signal and the plurality of matrix values, and generate the result of in-memory computing.
14. The apparatus of claim 1 , wherein the second plurality of transistors is configured to receive an input signal, perform analog charge-domain computations using the input signal and a plurality of matrix values, and generate the result of in-memory computing.
15. The apparatus of claim 2 , wherein the first plurality of transistors is configured to store a plurality of matrix values.
16. The apparatus of claim 15 , wherein the second plurality of transistors is configured to receive an input signal, perform analog charge-domain computations using the input signal and the plurality of matrix values, and generate the result of in-memory computing.
17. The apparatus of claim 1 , wherein the apparatus is configured for use in a neural network.
18. The apparatus of claim 1 , wherein the first plurality of transistors are configured as one of single memory bit cells and multiple memory bit cells.
19. The apparatus of claim 1 , wherein the apparatus is configured to perform matrix-vector multiplication operations.
20. The apparatus of claim 1 , wherein the second plurality of transistors is further configured to perform logic operations.
21. The apparatus of claim 1 , wherein the apparatus is configured to form a one-dimensional array of M multiplying bit-cells (M-BCs), where M in an integer greater than one.
22. The apparatus of claim 1 , wherein the apparatus is configured to form a two-dimensional array of N×M multiplying bit-cells (M-BCs), where M and N are integers greater than one.
23. A method of adapting operation of a plurality of multiplying bit-cells (M-BCs), each of the M-BCs comprising transistors configured as memory bit cells, transistors configured to perform in-memory computing of N-bit input vector elements using the memory bit cells to provide a bit cell output signal to a respective capacitor, each respective capacitor having an output terminal coupled to a compute line (CL) to impart thereto a respective portion of a CL charge indicative of a computational result of the plurality of M-BCs, the method comprising:
each bit cell, during a reset mode of operation, selectively coupling an input terminal of its capacitor to a respective first voltage;
each bit cell, during an evaluation mode of operation, selectively coupling the input terminal of its capacitor to a respective second voltage;
wherein for each bit cell one of the first and second voltages is a fixed voltage, and the other of the first and second voltages is determined using the respective N-bit input vector element.
24. The method of claim 23 , wherein for each bit cell a sign of the N-bit input vector element determines whether the first voltage is a fixed voltage or a voltage determined using the N-bit input vector element.
25. The method of claim 23 , wherein for at least one of the bit cells the fixed voltage comprises zero.
26. The method of claim 23 , wherein for at least one of the bit cells the fixed voltage comprises a predefined portion of a maximum voltage.
27. The method of claim 23 , wherein for at least one of the bit cells a voltage determined using the N-bit input vector element comprises an absolute value of the voltage represented by the N-bit input vector element.
28. The method of claim 23 , wherein for at least one of the bit cells the fixed voltage is selected as a voltage level determined in accordance with expected N-bit input vector elements.
29. The method of claim 28 , wherein the determined voltage level corresponds to a mean or median of an expected distribution of N-bit input vector elements.
30. The method of claim 28 , wherein the determined voltage level corresponds to a reverse polarity mean or median of an expected distribution of N-bit input vector elements.