Optical computing apparatus and system, and computing method
An optical computing apparatus and system and a computing method are provided. The optical computing apparatus includes a linear operation module, a first delay module, and a coupler. The linear operation module can modulate, based on received electrical signals, optical signals input to the linear operation module; the first delay module may adjust a delay of optical signals output by the linear operation module; and after the first delay module adjusts the delay of the optical signals output by the linear operation module, the coupler may combine a plurality of groups of optical signals successively output by the linear operation module, to output one group of optical signals used to indicate a computing result that is obtained after a multiply-add operation is performed on one group of data and weights.
1 . An optical computing apparatus, comprising:
a linear operation processor, configured to:
separately receive a first group of optical signals and a third group of optical signals, wherein the first group of optical signals and the third group of optical signals indicate a first group of data;
receive first electrical signals that indicate a first part of first weights, and modulate the first group of optical signals based on the first electrical signals to output a second group of optical signals, wherein the second group of optical signals indicates a computing result of the first group of data and the first part of the weights, and the first weights are weights of a first neural network layer; and
receive second electrical signals that indicate a second part of the first weights, and modulate the third group of optical signals based on the second electrical signals to output a fourth group of optical signals, wherein the fourth group of optical signals indicates a computing result of the first group of data and the second part of the weights;
a first delayer, configured to adjust a delay of the second group of optical signals; and
a coupler, configured to combine the delayed second group of optical signals and the fourth group of optical signals into a fifth group of optical signals, wherein the fifth group of optical signals indicates a first computing result that is obtained after multiply-add operations are performed on the first group of data and the first part of the weights and on the first group of data and the second part of the weights.
2 . The optical computing apparatus according to claim 1 , wherein
the linear operation processor is further configured to:
receive a sixth group of optical signals, wherein the sixth group of optical signals indicates a second computing result that is obtained after a multiply-add operation is performed on the first group of data and the first weights, and the second computing result comprises the first computing result; and
receive third electrical signals that indicate second weights, and modulate the sixth group of optical signals based on the third electrical signals to output a seventh group of optical signals, wherein the second weights are weights of a second neural network layer; and
the optical computing apparatus further comprises:
a filter, configured to: receive the seventh group of optical signals, filter out an invalid signal from the seventh group of optical signals, and output an eighth group of optical signals, wherein the eighth group of optical signals indicates a computing result that is obtained after a pooling operation is performed on the second computing result.
3 . The optical computing apparatus according to claim 2 , wherein
the linear operation processor is further configured to:
receive a ninth group of optical signals, wherein the ninth group of optical signals indicates a second group of data; and
receive fourth electrical signals that indicate third weights, and modulate the ninth group of optical signals based on the fourth electrical signals to output a tenth group of optical signals, wherein the tenth group of optical signals indicates a third computing result that is obtained after a multiply-add operation is performed on the second group of data and the third weights, and the third weights are weights of a third neural network layer; and
the filter is further configured to: receive the tenth group of optical signals, filter out an invalid signal from the tenth group of optical signals, and output an eleventh group of optical signals, wherein the eleventh group of optical signals indicates a computing result that is obtained after a convolution operation is performed on the third computing result.
4 . The optical computing apparatus according to claim 3 , further comprising:
a nonlinear operation processor, configured to perform a nonlinear operation on the sixth group of optical signals, the fifth group of optical signals, the eighth group of optical signals, or the eleventh group of optical signals, wherein the nonlinear operation is an activation function operation.
5 . The optical computing apparatus according to claim 1 , further comprising:
a dual-mode beam splitter, configured to: receive a twelfth group of optical signals that indicates the first group of data, and split the twelfth group of optical signals into the first group of optical signals and the third group of optical signals.
6 . The optical computing apparatus according to claim 5 , further comprising a second delayer, wherein
the dual-mode beam splitter is further configured to transmit the delayed second group of optical signals to the second delayer; and
the second delayer is further configured to adjust a delay of the second group of optical signals, so that the second group of optical signals and the fourth group of optical signals simultaneously arrive at the coupler.
7 . The optical computing apparatus according to claim 1 , wherein the linear operation processor comprises N optoelectronic modulators, and
each of the N optoelectronic modulators is configured to: receive one optical signal in the first group of optical signals and one of the first electrical signals that indicate the first part of the weights, and modulate the received optical signal based on the received electrical signal.
8 . The optical computing apparatus according to claim 7 , wherein the linear operation processor further comprises N phase modulators and at least (N−1) phase locking modules;
one of the N phase modulators is connected to one of the N optoelectronic modulators, and is configured to adjust a phase of one optical signal that is in the first or second group of optical signals and that is modulated by the phase modulator; and
one of the phase locking modules is connected to two of the N phase modulators, and is configured to lock phases of two optical signals that are in the first or second group of optical signals and that are modulated by a two phase modulators.
9 . The optical computing apparatus according to claim 3 , wherein the linear operation processor further comprises:
an amplifier, configured to amplify powers of the fifth group of optical signals, the sixth group of optical signals, the eighth group of optical signals, or the eleventh group of optical signals.
10 . A computing method, applied to an optical computing apparatus, wherein the optical computing apparatus comprises a linear operation processor, first delayer, and a coupler, and the method comprises:
receiving, by the linear operation processor, a first group of optical signals and first electrical signals, wherein the first group of optical signals indicates a first group of data, and the first electrical signals indicate a first part of first weights; and modulating the first group of optical signals based on the first electrical signals to output a second group of optical signals, wherein the second group of optical signals indicates a computing result of the first group of data and the first part of the weights, and the first weights are weights of a first neural network layer;
receiving, by the linear operation processor, a third group of optical signals and second electrical signals, wherein the third group of optical signals indicates the first group of data, and the second electrical signals indicate a second part of the first weights; and modulating the third group of optical signals based on the second electrical signals to output a fourth group of optical signals, wherein the fourth group of optical signals indicates a computing result of the first group of data and the second part of the weights;
adjusting, by the first delayer, a delay of the second group of optical signals; and
combining, by the coupler, the delayed second group of optical signals and the fourth group of optical signals into a fifth group of optical signals, wherein the fifth group of optical signals indicates a first computing result that is obtained after multiply-add operations are performed on the first group of data and the first part of the weights and on the first group of data and the second part of the weights.
11 . The computing method according to claim 10 , wherein the optical computing apparatus further comprises a filter, and the method further comprises:
receiving, by the linear operation processor, a sixth group of optical signals, and third electrical signals that indicate second weights, wherein the sixth group of optical signals indicates a second computing result that is obtained after a multiply-add operation is performed on the first group of data and the first weights, the second computing result comprises the first computing result, and the second weights are weights of a second neural network layer;
modulating, by the linear operation processor, the sixth group of optical signals based on the third electrical signals to output a seventh group of optical signals; and
receiving, by the filter, the seventh group of optical signals, filtering out an invalid signal from the seventh group of optical signals, and outputting an eighth group of optical signals, wherein the eighth group of optical signals indicates a computing result that is obtained after a pooling operation is performed on the second computing result.
12 . The computing method according to claim 11 , further comprising:
receiving, by the linear operation processor, a ninth group of optical signals, wherein the ninth group of optical signals indicates a second group of data;
receiving, by the linear operation processor, fourth electrical signals indicating third weights, and modulating the ninth group of optical signals based on the fourth electrical signals to output a tenth group of optical signals, wherein the tenth group of optical signals indicates a third computing result that is obtained after a multiply-add operation is performed on the second group of data and the third weights, and the third weights are weights of a third neural network layer; and
receiving, by the filter, the tenth group of optical signals, filtering out an invalid signal from the tenth group of optical signals, and outputting an eleventh group of optical signals, wherein the eleventh group of optical signals indicates a computing result that is obtained after a convolution operation is performed on the third computing result.
13 . An optical computing system, comprising a processor and an optical computing apparatus connected to the processor,
wherein the optical computing apparatus comprises a linear operation processor, configured to:
separately receive a first group of optical signals and a third group of optical signals, wherein the first group of optical signals and the third group of optical signals indicate a first group of data,
receive first electrical signals that indicate a first part of first weights, and modulate the first group of optical signals based on the first electrical signals to output a second group of optical signals, wherein the second group of optical signals indicates a computing result of the first group of data and the first part of the weights, and the first weights are weights of a first neural network layer; and
receive second electrical signals that indicate a second part of the first weights, and modulate the third group of optica signals based on the second electrical signals to output a fourth group of optical signals, wherein the fourth group of optical signals indicates a computing result of the first group of data and the second part of the weights,
a first delayer, configured to adjust a delay of the second group of optical signals, and
a coupler, configured to combine the delayed second group of optical signals and the fourth group of optical signals into a fifth group of optical signals, wherein the fifth group of optical signals indicates a first computing result that is obtained after multiply-add operations are performed on the first group of data and the first part of the weights and on the first group of data and the second part of the weights, and
wherein the processor of the optical computing system is configured to send to-be-computed data to the optical computing apparatus, and the data comprises the first group of data.
14 . The optical computing system according to claim 13 , wherein the linear operation processor is further configured to:
receive a sixth group of optical signals, wherein the sixth group of optical signals indicates a second computing result that is obtained after a multiply-add operation is performed on the first group of data and the first weights, and the second computing result comprises the first computing result; and
receive third electrical signals that indicate second weights, and modulate the sixth group of optical signals based on the third electrical signals to output a seventh group of optical signals, wherein the second weights are weights of a second neural network layer, and
the optical computing apparatus further comprises:
a filter, configured to: receive the seventh group of optical signals, filter out an invalid signal from the seventh group of optical signals, and output an eighth group of optical signals, wherein the eighth group of optical signals indicates a computing result that is obtained after a pooling operation is performed on the second computing result.
15 . The optical computing system according to claim 14 , wherein the linear operation processor is further configured to:
receive a ninth group of optical signals, wherein the ninth group of optical signals indicates a second group of data; and
receive fourth electrical signals that indicate third weights, and modulate the ninth group of optical signals based on the fourth electrical signals to output a tenth group of optical signals, wherein the tenth group of optical signals indicates a third computing result that is obtained after a multiply-add operation is performed on the second group of data and the third weights, and the third weights are weights of a third neural network layer, and
the filter is further configured to: receive the tenth group of optical signals, filter out an invalid signal from the tenth group of optical signals, and output an eleventh group of optical signals, wherein the eleventh group of optical signals indicates a computing result that is obtained after a convolution operation is performed on the third computing result.