Analog multiply-accumulate unit for multibit in-memory cell computing
Systems, apparatuses and methods include technology that receives, with a first plurality of multipliers of a multiply-accumulator (MAC), first digital signals from a memory array, wherein the first plurality of multipliers includes a plurality of capacitors. The technology further executes, with the first plurality of multipliers, multibit computation operations with the plurality of capacitors based on the first digital signals, and generates, with the first plurality of multipliers, a first analog signal based on the multibit computation operations.
1 . A computing system comprising:
a processor;
a memory array to store and output digital signals corresponding to bits of a multibit value; and
a multiply-accumulator (MAC), wherein the MAC includes a plurality of multipliers, and the plurality of multipliers include a plurality of capacitors including a group of capacitors and a further group of capacitors, a plurality of switches, and a plurality of branches that include the plurality of switches and the group of capacitors;
wherein the plurality of multipliers are configured to:
receive, by the plurality of switches, the digital signals from the memory array,
execute multibit computation operations with the plurality of capacitors and the plurality of switches forming a C-2C ladder based on the digital signals by controlling the plurality of switches using the digital signals to electrically connect or disconnect an input analog signal to the group of capacitors, and
generate an output analog signal based on the multibit computation operations.
2 . The computing system of claim 1 , wherein thesecond further group of capacitors connect the plurality of branches, further wherein a capacitance of the further group of capacitors is greater than a capacitance of the group of capacitors.
3 . The computing system of claim 1 , wherein the plurality of capacitors include a plurality of pairs of capacitors that each correspond to a different bit of the multibit value.
4 . The computing system of claim 1 , wherein the MAC further comprises:
a plurality of further multipliers that include a plurality of further capacitors that is to generate a further output analog signal based on further digital signals; and
an adder to add the output analog signal and the further output analog signal.
5 . The computing system of claim 1 , wherein the multibit value is associated with a weight of a neural network.
6 . The computing system of claim 1 , wherein the input analog signal is associated with an input activation of a neural network, and the output analog signal is associated with an output activation of the neural network.
7 . The computing system of claim 1 , wherein the output analog signal is associated with a product of the input analog signal and the multibit value.
8 . The computing system of claim 1 , wherein executing the multibit computation operations further comprises controlling the plurality of switches using the digital signals to electrically connect or disconnect a ground node to the group of capacitors.
9 . A semiconductor apparatus comprising:
one or more substrates; and
logic coupled to the one or more substrates, wherein the logic is implemented at least partly in one or more of configurable or fixed-functionality hardware, the logic comprising:
a memory array to store and output digital signals corresponding to bits of a multibit value; and
a multiply-accumulator (MAC) connected to the memory array, wherein the MAC includes a plurality of multipliers, and the plurality of multipliers include a plurality of capacitors including a group of capacitors and a further group of capacitors, a plurality of switches, and a plurality of branches that include the plurality of switches and the group of capacitors;
wherein the plurality of multipliers are configured to:
receive, by the plurality of switches, the digital signals from the memory array;
execute multibit computation operations with the plurality of capacitors and the plurality of switches forming a C-2C ladder based on the digital signals by controlling the plurality of switches using the digital signals to electrically connect or disconnect an input analog signal to the group of capacitors; and
generate an output analog signal based on the multibit computation operations.
10 . The semiconductor apparatus of claim 9 , wherein the further group of capacitors connect the plurality of branches, further wherein a capacitance of the further group of capacitors is greater than a capacitance of the group of capacitors.
11 . The semiconductor apparatus of claim 9 , wherein the plurality of capacitors include a plurality of pairs of capacitors that each correspond to a different bit of the multibit value.
12 . The semiconductor apparatus of claim 9 , wherein the MAC further comprises:
a plurality of further multipliers that include a plurality of further capacitors that are to generate a further output analog signal based on further digital signals; and
an adder to add the output analog signal and the further output analog signal.
13 . The semiconductor apparatus of claim 9 , wherein the multibit value is associated with a weight of a neural network.
14 . The semiconductor apparatus of claim 9 , wherein the logic coupled to the one or more substrates includes transistor channel regions that are positioned within the one or more substrates.
15 . The semiconductor apparatus of claim 9 , wherein the input analog signal is associated with an input activation of a neural network, and the output analog signal is associated with an output activation of the neural network.
16 . The semiconductor apparatus of claim 9 , wherein the output analog signal is associated with a product of the input analog signal and the multibit value.
17 . The semiconductor apparatus of claim 9 , wherein executing the multibit computation operations further comprises controlling the plurality of switches using the digital signals to electrically connect or disconnect a ground node to the group of capacitors.
18 . A method comprising:
receiving, with a plurality of switches of a plurality of multipliers of a multiply-accumulator (MAC), digital signals corresponding to bits of a multibit value from a memory array, wherein the plurality of multipliers include a plurality of capacitors including a group of capacitors and a further group of capacitors, the plurality of switches, and a plurality of branches that include the plurality of switches and the group of capacitors;
executing, with the plurality of capacitors and the plurality of switches forming a C-2C ladder, multibit computation operations based on the digital signals by controlling the plurality of switches using the digital signals to electrically connect or disconnect an input analog signal to the group of capacitors; and
generating, with the plurality of multipliers, an output analog signal based on the multibit computation operations.
19 . The method of claim 18 , wherein the further group of capacitors connect the plurality of branches, further wherein a capacitance of the further group of capacitors is greater than a capacitance of the group of capacitors.
20 . The method of claim 18 , wherein the plurality of capacitors include a plurality of pairs of capacitors that each correspond to a different bit of the multibit value.
21 . The method of claim 18 , further comprising:
generating, with a plurality of further multipliers of the MAC, a further output analog signal based on further digital signals, wherein the plurality of further multipliers include a plurality of further capacitors; and
adding the output analog signal and the further output analog signal.
22 . The method of claim 18 , wherein the multibit value is associated with a weights of a neural network.
23 . The method of claim 18 , wherein the input analog signal is associated with an input activation of a neural network, and the output analog signal is associated with an output activation of the neural network.
24 . The method of claim 18 , wherein the output analog signal is associated with a product of the input analog signal and the multibit value.
25 . The method of claim 18 , wherein executing the multibit computation operations further comprises controlling the plurality of switches using the digital signals to electrically connect or disconnect a ground node to the group of capacitors.