Bit-Slice Based High-Precision Analog Optical Computing Method and System

Abstract

A high-precision analog optical computing method and system based on bit slicing, the steps include: decomposing a high-precision signal into low-precision signal segments, decomposing a high-precision weight into multiple low-precision weight segments, performing multiplication operations of the low-precision signal and the weight in the optical domain respectively, and finally combining these low-precision calculation results to restore a high-precision output. The present invention processes by splitting high-precision data into multiple low-precision bit slices, utilizes the high-speed parallel characteristics of photonic computing, simultaneously processes multiple input bit slices and weight bit slices, improves the computing efficiency of large-scale computing tasks, and reduces the hardware requirements for optical computing. In addition, by adjusting the values of M and N, it can flexibly adapt to computing tasks of different scales and precisions, has strong adaptability and flexibility, achieves a balance between computing precision and hardware complexity, and also fully utilizes the parallel processing advantage of photons, providing a solution for large-scale data processing and efficient optical domain computing.

Description

Technical Field

[0001] The present invention relates to the field of optical computing, and particularly to a high-precision analog optical computing method and system based on bit slicing. Background Art

[0002] Optical computing is a new technology that uses light as an information carrier and realizes computing through the physical properties of light. Its core principle relies on the characteristics of light wave propagation, interference, diffraction, and refraction. Different from traditional electronic computers that rely on electron flow for logical operations in semiconductors, optical computing completes similar operations through the behavior of photons. The basic units of optical computing include optical devices such as optical crystals, waveguides, modulators, and lasers. These devices manipulate the phase, amplitude, polarization, and wavelength of light to achieve data transmission and logical operations. At the same time, with its unique parallelization advantage, optical computing can achieve simultaneous computing of multi-channel radars through time multiplexing, space multiplexing, wavelength multiplexing, mode multiplexing, etc.

[0003] In recent years, good research foundations have been formed in the field of optical computing technology at home and abroad, and the high-throughput and low-latency advantages of optical processing means have been preliminarily verified. Among them, the University of Weibin used an optical frequency comb as a light source and realized ultra-high-speed optical convolutional computing using the principle of optical dispersion, with a clock frequency reaching 62.9 GHz (X. Xu, M. Tan, B. Corcoran, et al. 11TOPS photonic convolutional accelerator for optical neural networks. Nature589, 44–51 (2021)). Cornell University used a spatial light modulator to construct a large-scale neural network and proved that only less than the energy of one photon is required to complete one multiplication (T. Wang, SY. Ma, L.G. Wright, et al. An opticalneural network using less than 1 photon per multiplication. Nat Commun 13,123 (2022)).

[0004] However, although preliminary verification has been achieved in individual performances of optical computing, further improvement of its actual application efficiency still faces challenges. Among them, factors such as optical link insertion loss, environmental interference, and active noise are the main reasons for the decline in optical computing accuracy. These problems not only affect the stability and reliability of optical computing but also limit its application in more fields.

[0005] Currently, research on improving optical computing accuracy mainly focuses on weight accuracy regulation. To address the challenges of high sensitivity and limited weight accuracy in microring resonators, a research team at Princeton University proposed using feedforward and feedback methods to achieve microring resonance locking and continuous weight configuration control, achieving a 9-bit microring regulation accuracy. They successfully developed a high-precision control scheme for microring resonators that can be extended to large optical computing networks (W. Zhang, A. Tait, C. Huang, et al. Broadband physical layer cognitive radio with an integrated photonic processor for blind source separation. Nat Commun 14, 1107 (2023)). The research team at Peking University also overcame the calibration of multi-wavelength parallel optical computing systems and ultra-high-precision weight loading. The calibration method they developed for multi-wavelength computing systems can accurately regulate various optoelectronic devices, achieving a 9-bit weight regulation accuracy (B. Bai, Q. Yang, H. Shu et al. Microcomb-based integrated photonic processing unit. Nat Commun 14, 66 (2023)).

[0006] However, although these methods of improving weight accuracy have enhanced the accuracy of optical computing to some extent, they have not fundamentally solved the problem of improving the final computing accuracy. In particular, the optimization effect on the final computing results of analog optical computing is not obvious. When analog optical computing processes complex tasks, its accuracy and stability are often restricted by various factors, such as nonlinear effects, optical losses, and error accumulation. Therefore, a comprehensive solution for improving the final computing accuracy is needed to achieve high-precision analog optical computing. Summary of the Invention

[0007] The object of the present invention is to propose a high-precision analog optical computing method and system based on bit slicing in view of the above-mentioned deficiencies of the existing technologies. This scheme adopts a multi-computation approach. First, the input signal and weight signal are decomposed by bit slicing according to their accuracy requirements to generate multiple signal combinations with low-bit accuracy; subsequently, analog optical computing technology is used to independently compute these low-bit accuracy signals and weights; finally, multiple low-precision computing results are accumulated and combined according to weights to generate a high-precision matrix multiplication or convolution computing result.

[0008] The technical solution of the present invention is as follows:

[0009] On the one hand, the present invention provides a high-precision analog optical computing method based on bit slicing, which is characterized by including the following steps:

[0010] S1. Decompose the input signal and the weight signal respectively according to the predetermined precision requirements to generate a plurality of signal combinations with low-bit precision, where each signal combination with low-bit precision contains a part of the bit positions of the input signal or the weight signal;

[0011] S2. Use analog optical computing technology to independently calculate each signal combination with low-bit precision and the corresponding weight signal with low-bit precision generated in step S1 to obtain a plurality of calculation results with low precision;

[0012] S3. Accumulate and combine the plurality of calculation results with low precision obtained in step S2 according to the predetermined weights, so as to generate a calculation result of high-precision matrix multiplication or convolution.

[0013] Further, the bit slicing decomposition in step S1 is to decompose the input signal and the weight signal respectively into a plurality of binary bit segments, and the number of bits of each binary bit segment is determined according to the required calculation precision.

[0014] Further, the high-precision input data stream with a word length of K bit is split into M low-precision input bit slices with a word length of K / M bit through the input bit slicing reconstruction module, where K is a positive integer multiple of M, and M is a positive integer greater than or equal to 2;

[0015] The high-precision weight with a word length of L bit is split into N low-precision weight bit slices with a word length of L / N bit through the weight bit slicing reconstruction module, and M copies are output, where L is a positive integer multiple of N, and N is a positive integer greater than or equal to 2.

[0016] Further, the input bit slicing reconstruction module includes an electrical input port, M optical output ports and M electro-optical converters, and the electro-optical converters convert the electrical signals of the M low-precision input bit slices into optical signals and then output them.

[0017] Further, the weight bit slicing reconstruction module includes an electrical input port and M×N electrical output ports, and the M×N electrical output ports output M copies of the low-precision weight bit slices.

[0018] Further, it further includes:

[0019] A photon computing core module, which has M optical input ports, M×N electrical input ports, M×N electrical output ports, M×N weight regulators and M×N photodetectors, and is used to perform dot product calculations on the M low-precision input bit slices and the M×N low-precision weight bit slices, and convert the calculation results into electrical signals through the photodetectors and output them.

[0020] The output bit-slice combination module has M×N electrical input ports and one electrical output port, and is used to multiply M×N low-precision calculation results by corresponding weights and then sum them to obtain a high-precision calculation result, which is output through one electrical output port.

[0021] In a second aspect, the present invention further provides a high-precision analog optical computing system based on bit-slicing, which is characterized by including:

[0022] The input bit-slice reconstruction module has one electrical input port, M optical output ports, and M electro-optical converters, where M is a positive integer greater than or equal to 2; the high-precision input data with a word length of K bit flows in through the electrical input port, and is split into M low-precision input bit-slices with a word length of K / M bit through a reconstruction method. After being converted into optical signals by the electro-optical converters, they are output to the photon computing core module, where K is a positive integer multiple of M;

[0023] The weight bit-slice reconstruction module has one electrical input port and M×N electrical output ports, where N is a positive integer greater than or equal to 2; the high-precision weight with a word length of L bit flows in through the electrical input port, and is split into N low-precision weight bit-slices with a word length of L / N bit through a reconstruction method, and then copied M times and output to the photon computing core module, where L is a positive integer multiple of N;

[0024] The photon computing core module has M optical input ports connected to the M optical output ports of the input bit-slice reconstruction module, M×N electrical input ports connected to the M×N electrical output ports of the weight bit-slice reconstruction module, M×N electrical output ports, M×N weight regulators, and M×N photodetectors; M×N weight bit-slices are loaded onto the high-parallel photon computing core through the M×N weight regulators, and are calculated with M input bit-slices to obtain M×N output bit-slice results, which are respectively converted into electrical signals by the M×N photodetectors and transmitted to the output bit-slice combination module as output bit-slices;

[0025] The output bit-slice combination module has M×N electrical input ports connected to the M×N electrical output ports of the photon computing core module and one electrical output port; each input bit-slice and each weight bit-slice will calculate an output bit-slice result, numbered (m, n), that is, the result of the calculation of the m-th input bit-slice and the n-th weight bit-slice, where 1≤m≤M and 1≤n≤N. The output bit-slice numbered (m, n) is multiplied by the corresponding weight coefficient 2 (M -m)K / M+(N-n)L / N , and then the M×N output bit-slices are summed to obtain a high-precision calculation result, which is output through one electrical output port.

[0026] Furthermore, the electro-optical converter in the input bit-slice reconstruction module is a coherent modulator, which works in the zero-bias point mode, linearly modulates the input electrical signal onto the optical amplitude quantity, and the value of the signal is represented by the amplitude value of the light.

[0027] Furthermore, the reconstruction methods in the input bit-slice reconstruction module and the weight bit-slice reconstruction module are implemented by a single-chip microcomputer or a digital computer.

[0028] Furthermore, the photon computing core module adopts an on-chip coherent optical computing architecture, an on-chip incoherent optical computing architecture, a spatial diffraction optical computing architecture; or adopts different ways in terms of time, space, mode, and wavelength to achieve operations on different signals and weights.

[0029] Furthermore, the photon computing core module can use, including but not limited to, an on-chip coherent optical computing architecture, an on-chip incoherent optical computing architecture, a spatial diffraction optical computing architecture; it can also use, including but not limited to, the same optical computing architecture at different time points, or can also be an optical computing architecture different in space.

[0030] Furthermore, the weight regulator in the photon computing core module can use, including but not limited to, photon devices such as microrings, phase shifters, phase change materials, spatial light modulators, etc. that have an adjustment effect on the optical field amplitude.

[0031] Compared with the prior art, the technical advantages of the present invention are as follows:

[0032] (1) Compared with the traditional optical computing architecture, the present invention adopts a bit-slice optical computing method, enabling a low-precision optical computing device to achieve high-precision optical computing tasks, greatly expanding the application scope of optical computing, providing a new solution for processing in complex scenarios, and breaking through the limitations faced by the traditional optical computing architecture when processing high-precision data.

[0033] (2) The present invention can make full use of the existing low-precision computing system resources for high-precision computing, reducing the requirements for the integration and packaging of the optical computing system, making the present invention easier to integrate and upgrade with the existing system, and providing convenience for the popularization and application of optical computing technology.

[0034] (3) The present invention realizes high-precision analog optical computing through the reconstruction of input and weights, the efficient dot product calculation in the photon computing core, and the shift addition combination of the output results, not only improving the accuracy and efficiency of the calculation, but also making full use of the parallel processing advantages of photons, providing a promising solution for large-scale data processing.

[0035] (4) By splitting high-precision data into multiple low-precision bit slices for processing, and utilizing the high-speed parallel characteristics of photonic computing, multiple input bit slices and weight bit slices can be processed simultaneously, significantly improving the computing efficiency of large-scale computing tasks. In addition, by adjusting the values of M and N, it is possible to flexibly adapt to computing tasks of different scales and precisions, with strong adaptability and flexibility. Description of the Drawings

[0036] Figure 1 It is a schematic diagram of an embodiment of the high-precision analog optical computing system based on bit slices of the present invention.

[0037] Figure 2 It is a schematic diagram of the high-precision analog optical computing data flow of the present invention.

[0038] Figure 3 It is a schematic diagram of the principle of high-precision dot product calculation of the present invention.

[0039] Figure 4 It is a schematic diagram of the principle of reconstructing high-precision input and weights into bit slices of the present invention. Detailed Embodiment

[0040] The technical solutions of the present invention will be described in detail below in conjunction with the drawings and embodiments. Detailed implementation manners and structures are given, but the protection scope of the present invention is not limited to the following embodiments.

[0041] Please refer to Figure 1 , Figure 1 It is a schematic diagram of an embodiment of the high-precision analog optical computing system based on bit slices of the present invention. As shown in the figure, the high-precision analog optical computing system based on bit slices of this embodiment includes an input bit slice reconstruction module 100, a weight bit slice reconstruction module 200, a photonic computing core module 300, and an output bit slice combination module 400.

[0042] The input bit slice reconstruction module 100 has one electrical input port, M optical output ports, and M electro-optical converters. In this embodiment, M = 3.

[0043] The 3 optical output ports of this module are connected to the optical input ports of the photonic computing core module 300. A high-precision input data stream with a word length of K = 6 bit enters through the electrical input port, and is split into 3 low-precision input bit slices with a word length of 2 bit by using a reconstruction algorithm. These low-precision input bit slices are numbered 1, 2, and 3 in order from the high bit to the low bit. Then, the 3 electrical signals are converted into optical signals through the electro-optical converters and output to the photonic computing core module 300.

[0044] The weight bit slice reconstruction module 200 has one electrical input port and M×N electrical output ports. In this embodiment, N = 3.

[0045] The nine electrical output ports of this module are connected to the electrical input ports of the photon computing core module (300).

[0046] High-precision weights with a word length of L = 6 bits are input through the electrical input ports. Using a reconstruction algorithm, they are split into three low-precision weight bit slices with a word length of 2 bits, which are 1, 2, and 3 from the high bit to the low bit in sequence. Then, these low-precision weight bit slices are copied three times and output to the photon computing core module 300.

[0047] The photon computing core module 300 has M optical input ports, M×N electrical input ports, M×N electrical output ports, M×N weight controllers, and M×N photodetectors; that is, in this embodiment, there are 3 optical input ports, 9 electrical input ports, 9 electrical output ports, 9 weight controllers, and 9 photodetectors. The nine electrical input ports of this module are connected to the electrical output ports of the weight bit slice reconstruction module 200, the three optical input ports of this module are connected to the optical output ports of the input bit slice reconstruction module 100, and the nine electrical output ports of this module are connected to the electrical input ports of the output bit slice combination module 400.

[0048] Three input bit slices are input through the optical input ports, and nine weight bit slices are input through the electrical input ports. They are loaded onto the high-parallel photon computing core through nine weight controllers, and nine output bit slice results are obtained after passing through the nine weight controllers for the three input bit slices. Each input bit slice and each weight bit slice will calculate an output bit slice result, numbered (m, n), that is, the result of the calculation of the m-th input bit slice and the n-th weight bit slice, where 1 ≤ m ≤ 3 and 1 ≤ n ≤ 3. Finally, they are converted into electrical signals through the photodetectors and output to the output bit slice combination module 400.

[0049] The output bit slice combination module 400 has nine electrical input ports and one electrical output port. The nine electrical input ports of this module are connected to the electrical output ports of the photon computing core module (300). The output bit slice numbered (m, n) is multiplied by the corresponding weight coefficient multiplied by 2 (M-m)K / M+(N-n)L / N and then summed to obtain a high-precision calculation result. In this embodiment, the (m, n)-th output bit slice is multiplied by 2 2(6-m-n) and the nine output bit slices are weighted and summed to obtain a high-precision calculation result, which is output through one electrical output port, that is, the final high-precision calculation result is completed.

[0050] Please refer to Figure 2 , Figure 2 which is the schematic diagram of the high-precision analog optical computing data flow of the present invention.

[0051] The input high-precision image data IN and the convolution kernel WEIGHT are respectively reconstructed into high bits and low bits, that is, IN 1 and IN 2 , WEIGHT 1 and WEIGHT 2 . Through the high-parallel photon computing core, the input bit slices and the convolution kernel bit slices are combined pairwise to obtain four output bit slices OUT 1,1 , OUT 1,2 , OUT 2,1 , OUT 2,2 . Finally, the high-precision output result OUT High is obtained through shift combination, and its result is significantly better than the low-precision calculation result OUT LOW .

[0052] Please refer to Figure 3 and Figure 4 , Figure 3 , which is the schematic diagram of the principle of the high-precision dot product calculation of the present invention Figure 4 illustrating the principle of reconstructing high-precision inputs and weights into bit slices

[0053] 6-bit input and weight data can be divided into 3 groups of 2-bit input and weight bit slices according to the number of bits, that is, 3 groups of 2-bit data streams. Through the pairwise combination dot product of the bit slices, 4-bit output bit slices can be obtained, and then through the shift method, that is, multiplying by the coefficient during reconstruction, and finally adding up to obtain a 12-bit high-precision output result. It can be seen that its calculation result is the same as that directly calculated with 6-bit precision

[0054] In the present invention, the photon computing core module utilizes the parallelism and high-speed transmission characteristics of photons to perform dot product calculations. That is, each weight bit slice and input bit slice are encoded as optical signals, and then interact with each other through optical elements in the photon computing core to achieve dot product operations. Due to the parallel processing ability of light, the photon computing core can process multiple data slices simultaneously, thus significantly improving the calculation speed. After obtaining the dot product results of all low-precision bit slices, these results are combined through shift and addition operations to restore the results of the original high-precision calculation, which is completed through the output bit slice combination module, thereby obtaining the final high-precision output result. The present invention realizes high-precision analog optical computing through the reconstruction of inputs and weights, the efficient dot product calculation in the photon computing core, and the shift addition combination of the output results, not only improving the calculation accuracy and efficiency, but also making full use of the parallel processing advantages of photons, providing a promising solution for large-scale data processing

Claims

1. A high-precision analog optical computing method based on bit slicing, characterized in that: The following steps are involved: S1. Decomposing the input signal and the weight signal by bit slicing according to the predetermined precision requirements, and generating a plurality of signal combinations with low bit precision, wherein each signal combination with low bit precision contains a part of the bits of the input signal or the weight signal; S2. Using analog optical computing technology, each low-bit precision signal combination generated in step S1 and the corresponding low-bit precision weight signal are independently calculated to obtain multiple low-precision calculation results; S3. Accumulate and combine the multiple low-precision calculation results obtained in step S2 according to predetermined weights to generate a high-precision matrix multiplication or convolution calculation result.

2. The high-precision analog optical computing method based on bit slicing according to claim 1, characterized in that: The bit slice decomposition in step S1 is to decompose the input signal and the weight signal into a plurality of binary bit segments respectively, and the number of bits of each binary bit segment is determined according to the required calculation accuracy.

3. The high-precision analog optical computing method based on bit slicing according to claim 2, characterized in that: The high-precision input data stream with a word length of K bits is split into M low-precision input bit slices with a word length of K / M bits through an input bit slice reconstruction module, where K is a positive integer multiple of M and M is a positive integer ≥ 2; The high-precision weight with a word length of L bits is split into N low-precision weight bit slices with a word length of L / N bits through the weight bit slice reconstruction module, and M copies of the output are copied, where L is a positive integer multiple of N and N is a positive integer ≥2.

4. The high-precision analog optical computing method based on bit slicing according to claim 3, characterized in that: The input bit slice reconstruction module includes an electrical input port, M optical output ports and M electro-optical converters. The electro-optical converter converts the electrical signals of the M low-precision input bit slices into optical signals and then outputs them.

5. The high-precision analog optical computing method based on bit slicing according to claim 3, characterized in that: The weight bit slice reconstruction module includes an electrical input port and M×N electrical output ports, and the M×N electrical output ports output M copies of low-precision weight bit slices.

6. A high-precision analog optical computing system based on bit slicing, characterized in that: include: An input bit slice reconstruction module having an electrical input port, M optical output ports, and M electro-optical converters, where M is a positive integer ≥ 2; The high-precision input data stream with a word length of K bits is input through the electrical input port, and is split into M low-precision input bit slices with a word length of K / M bits by reconstruction. The M electrical signals are converted into optical signals by the electro-optical converter and then output to the photonic computing core module, where K is a positive integer multiple of M. The weight bit slice reconstruction module has an electrical input port and M×N electrical output ports, where N is a positive integer ≥ 2; a high-precision weight with a word length of L bits is input through the electrical input port, split into N low-precision weight bit slices with a word length of L / N bits by reconstruction, and then copied M times and output to the photon computing core module, where L is a positive integer multiple of N; A photon computing core module, comprising M optical input ports connected to the M optical output ports of the input bit slice reconstruction module, M×N electrical input ports connected to the M×N electrical output ports of the weight bit slice reconstruction module, M×N electrical output ports, M×N weight controllers and M×N photodetectors; the M×N weight bit slices are loaded onto the high-parallel photon computing core via the M×N weight controllers, and the M×N output bit slice results are obtained by calculating with the M input bit slices, and are converted into electrical signals via the M×N photodetectors respectively, and transmitted to the output bit slice combination module as output bit slices; The output bit slice combination module has M×N electrical input ports connected to the M×N electrical output ports of the photon computing core module and one electrical output port; each input bit slice and each weight bit slice calculate an output bit slice result, which is numbered (m, n), that is, the result of the calculation of the mth input bit slice and the nth weight bit slice, where 1≤m≤M, 1≤n≤N, and the output bit slice numbered (m, n) is multiplied by the corresponding weight coefficient 2 (M-m)K / M+(N-n)L / N , then the M×N output bit slices are summed to obtain a high-precision calculation result, which is output through an electrical output port.

7. The high-precision analog optical computing system based on bit slicing according to claim 6, characterized in that: The electro-optical converter in the input bit slice reconstruction module is a coherent modulator, which works in a zero bias point mode and linearly modulates the input electrical signal to an optical amplitude quantity, and the value of the signal is represented by the amplitude value of the light.

8. The high-precision analog optical computing system based on bit slicing according to claim 6, characterized in that: The reconstruction methods in the input bit slice reconstruction module and the weight bit slice reconstruction module are implemented by a single chip microcomputer or a digital computer.

9. The high-precision analog optical computing system based on bit slicing according to claim 6, characterized in that: The photonic computing core module adopts an on-chip coherent optical computing architecture, an on-chip incoherent optical computing architecture, or a spatial diffraction optical computing architecture; or adopts different time, space, mode, and wavelength methods to realize operations on different signals and weights.

Images