JPH05224767A - Parallel arithmetic unit - Google Patents

Abstract

PURPOSE:To facilitate optimum optical signal replacement to an applied data processing by providing plural two-dimensional optical signal replacement networks for two-dimensional arrangement between an arithmetic processing part and a data storage part based on rules set in advance. CONSTITUTION:Data processed at an arithmetic processing part 2 are inputted to an optical crossbar exchange part 33 as twodimensional optical signals 34 and in respect to the applied data processing, the data are inputted to an optimum replacement network 32 among three two-dimensional signal replacement networks 32. Then, the positions of respective optical signals in the two-dimensional optical signal arrangement are two-dimensionally rearranged based on the rules set in advance. These rearranged data are inputted through the optical crossbar exchange part 33 to a data storage part 3 again. On the other hand, data outputted from the data storage part 3 are inputted through the optical crossbar exchange part 33 to the two-dimensional optical signal replacement network 32 optimum for the applied data processing as two-dimensional optical signals 34. Then, the positions of respective optical signals are two-dimensionally rearranged based on the set rules and inputted through the optical crossbar exchange part 33 to the arithmetic processing part 2 again.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an optical computer in a broad sense, and more specifically, it is possible to simultaneously process an extremely large amount of data in parallel by utilizing the parallelism and incoherence of light. The present invention relates to a parallel arithmetic device.

[0002]

2. Description of the Related Art In recent years, a parallel arithmetic unit for optically performing parallel processing has been vigorously studied in place of a sequential serial type arithmetic unit using conventional electronic technology, and various devices have been proposed. 2 shows an example of such a parallel arithmetic unit. The parallel arithmetic device 1 is composed of an arithmetic processing unit 2, a data storage unit 3, and a two-dimensional optical signal replacement network 4, and the arithmetic processing unit 2 and the data storage unit 3 are connected via a two-dimensional optical signal replacement network 4. ing. The arithmetic processing unit 2 includes a plurality of arithmetic elements 5 having a function of transmitting and receiving optical signals.
Are arranged in a plane (two-dimensional) vertically and horizontally. The data storage unit 3 includes a plurality of storage elements 6, 6, ... Having a function of transmitting and receiving an optical signal, arranged vertically and horizontally on a plane (two-dimensional). The two-dimensional optical signal replacement network 4 sets the position of each optical signal 7 (8) output from the arithmetic processing unit 2 (data storage unit 3) and arranged vertically and horizontally on a plane (two-dimensional) to an arbitrary position on the plane. Rearranged data storage unit 3
The data is output to the (arithmetic processing unit 2).

As shown in FIG. 3, the arithmetic element 5 receives an electric circuit 11 for performing an arithmetic operation, a light emitting element 13 for converting an electric signal from the electric circuit 11 into an optical signal 12 and outputting the optical signal 12, and an optical circuit 14 for receiving the optical signal 14. It is composed of a light receiving element 15 for converting into an electric signal to be inputted to 11. As shown in FIG. 4, the storage element 6 includes an electric circuit 21 and an electric circuit 2 for storing data.
It is composed of a light emitting element 23 which converts an electric signal from 1 into an optical signal 22 and outputs it, and a light receiving element 25 which receives the optical signal 24 and converts it into an electric signal to be inputted to the electric circuit 21. A laser diode (LD), a light emitting diode (LED) or the like is preferably used as the light emitting elements 13 and 23, and the light receiving elements 15 and 25 are, for example, an avalanche photodiode (APD) and a pin. A photodiode (PIN-PD), a phototransistor or the like is preferably used. The two-dimensional optical signal replacement network 4 is, for example, an optical perfect shuffle network (CWStirk et al., Applied Optic
s, vol.27, No.2 pp.202-203 (1988)), optical crossover network (J.Jahns et al., Applied Optics, vol.27, No.15, p)
p.3155-3160 (1988)), cross network (SH.Lin et al., Appl
ied Optics, vol.27, No.9, pp.1734-1741 (1988)) and the like are preferably used.

In this parallel arithmetic unit 1, each arithmetic element 5, 5, ... Of the arithmetic processing section 2 is connected to an arbitrary storage element 6, 6, ... Of the data storage section 3 via the two-dimensional optical signal substitution network 4. Data can be written to and read from at the same time. Therefore, the respective arithmetic elements 5, 5, ... Can simultaneously send and receive data in parallel via the other arbitrary arithmetic element 5 and the arbitrary storage element 6.

[0005]

However, in the parallel computing device 1 described above, in order to perform the optimum two-dimensional optical signal replacement for given data processing, the two-dimensional optical signal replacement network 4 has a two-dimensional structure. It is necessary to have the function of arbitrarily rearranging the positions of the optical signals 7 (8) arranged in
Therefore, it is necessary to provide the two-dimensional optical signal replacement network 4 with a control function for designating a rearrangement pattern, resulting in a problem that the optical system becomes very complicated. Also, 2
Each optical signal 7 arranged two-dimensionally on the four-dimensional optical signal replacement network 4
If (8) does not have the function of arbitrarily rearranging the positions on the two-dimensional array, there is a drawback that optimum two-dimensional optical signal replacement cannot be performed for given data processing.

The present invention has been made in view of the above circumstances, and uses optimum two-dimensional optical signal replacement for given data processing by using a two-dimensional optical signal replacement network with a complicated optical system. It is to provide a parallel arithmetic device that can be realized without the need.

[0007]

In order to solve the above problems, the present invention adopts the following parallel arithmetic unit. That is, an arithmetic processing unit having a plurality of arithmetic elements having a function of transmitting and receiving an optical signal arranged two-dimensionally, and a plurality of storage elements having a function of transmitting and receiving an optical signal arranged two-dimensionally. Data storage unit, a plurality of two-dimensional optical signal permutation networks that rearrange the positions of the two-dimensionally arranged optical signals in two dimensions based on a preset rule, the arithmetic processing unit, and the data storage unit And the two-dimensional optical signal output from each of the two-dimensional optical signal replacement networks is crossbar exchanged as the two-dimensional optical signal, and is output as a two-dimensional optical signal to each of the arithmetic processing unit, the data storage unit and the two-dimensional optical signal replacement network. And an optical crossbar exchanging unit that operates.

[0008]

In the parallel arithmetic device of the present invention, the data processed by the arithmetic processing unit is input to the optical crossbar exchanging unit as a two-dimensional optical signal. The optical crossbar switching unit selects an optimum one from a plurality of two-dimensional optical signal replacement networks for given data processing, and inputs data to this optimum two-dimensional optical signal replacement network. The two-dimensional optical signal replacement network rearranges the positions of the two-dimensionally arranged optical signals in two dimensions based on a preset rule. The rearranged data is input to the data storage unit again via the optical crossbar switching unit.

The data output from the data storage section is input to the optical crossbar switching section as a two-dimensional optical signal.
The optical crossbar switching unit selects an optimum one from a plurality of two-dimensional optical signal permutation networks for given data processing,
Data is input to this optimum two-dimensional optical signal replacement network. The two-dimensional optical signal replacement network rearranges the positions of the two-dimensionally arranged optical signals in two dimensions based on a preset rule. The rearranged data is input again to the arithmetic processing unit via the optical crossbar exchanging unit.

[0010]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A parallel arithmetic unit according to the present invention will be described below with reference to FIG. In the parallel arithmetic unit 31 of FIG. 1, the same components as those of the parallel arithmetic unit 1 of FIGS.

This parallel arithmetic unit 31 includes an arithmetic processing unit 2,
The data storage unit 3, the three two-dimensional optical signal replacement networks 32, and the optical crossbar switching unit 33 are included. The arithmetic processing unit 2, the data storage unit 3, and the output sides of the three two-dimensional optical signal replacement networks 32 are optical crossbar switching units. It is connected to the input side of the section 33, and the input side of the arithmetic processing section 2, the data storage section 3, and the three two-dimensional optical signal replacement network 32 is connected to the output side of the optical crossbar switching section 33.

The two-dimensional optical signal replacement network 32 rearranges the positions of the respective optical signals of the two-dimensional optical signal 34 output from the optical crossbar exchanging unit 33 into two dimensions based on a preset rule, and the optical crossbar It is output to the switching unit 22, and has a configuration in which a control element for route selection is removed from the optical perfect shuffle network, the optical crossover network, the cross network, etc. shown in the conventional example.

The optical crossbar exchanging section 33 has three two-dimensional optical signal replacement networks 32, 32, for the given data processing.
The optimum one is selected from among the arithmetic processing unit 2, the data storage unit 3, and the two-dimensional optical signal 34 from each of the two-dimensional optical signal replacement networks 32, and the two-dimensional optical signal is crossbar-exchanged as it is, and the arithmetic processing unit 2, a data storage unit 3, and an optical space switch that outputs a two-dimensional optical signal to each of the two-dimensional optical signal replacement networks 32.

Next, the operation of the parallel arithmetic unit 31 will be described. The data processed by the arithmetic processing unit 2 is input to the optical crossbar exchanging unit 33 as a two-dimensional optical signal 34. The optical crossbar exchanging unit 33 selects the optimum one from the three two-dimensional optical signal replacement networks 32 for the given data processing, and inputs the data to the optimum two-dimensional optical signal replacement network 32. The two-dimensional optical signal replacement network 32 rearranges the positions of the two-dimensionally arranged optical signals in two dimensions based on a preset rule. The rearranged data is input to the data storage unit 3 again via the optical crossbar exchange unit 33.

The data output from the data storage section 3 is input to the optical crossbar switching section 33 as a two-dimensional optical signal 34. The optical crossbar switching unit 33 selects an optimum one from the three two-dimensional optical signal replacement networks 32 for the given data processing, and the optimum two-dimensional optical signal replacement network 32 is selected.
Enter the data in. This two-dimensional optical signal replacement network 32 is
The positions of the two-dimensionally arranged optical signals are two-dimensionally rearranged based on a preset rule. The rearranged data is input to the arithmetic processing unit 2 again via the optical crossbar exchanging unit 33.

As described above, according to the parallel arithmetic unit 31 of this embodiment, the arithmetic elements 5, 5, of the arithmetic processing unit 2 are arranged.
... is any storage element 6 of the data storage unit 3 via the two-dimensional optical signal replacement network 32 that is optimal for given data processing.
Data can be written to or read from 6, ... Therefore, each arithmetic element 5, 5, ...
... can exchange data in parallel via other arbitrary operation elements 5, 5, ... And arbitrary storage elements 6, 6 ,.

In addition, in this parallel arithmetic unit 31, three 2
Although the two-dimensional optical signal replacement network 32 is required, these two
Since the three-dimensional optical signal substitution network 32 is a simple optical system that does not require control, the number of arithmetic elements 5, 5, ... In the arithmetic processing unit 2 and the storage elements 6, 6 ,. The degree of parallelism of the device can be increased.

As described above, a parallel arithmetic unit capable of realizing optimum two-dimensional optical signal replacement for given data processing without using a two-dimensional optical signal replacement network with a complicated optical system is provided. be able to.

[0019]

As described above, according to the parallel arithmetic device of the present invention, an arithmetic processing unit in which a plurality of arithmetic elements having a function of transmitting and receiving an optical signal are two-dimensionally arranged, and an optical signal A data storage unit in which a plurality of storage elements having a function of transmitting and receiving light are arranged in a two-dimensional manner, and the positions of the optical signals arranged in a two-dimensional manner are rearranged in a two-dimensional manner based on a preset rule. A plurality of two-dimensional optical signal replacement networks that
The two-dimensional optical signal output from each of the arithmetic processing unit, the data storage unit and the two-dimensional optical signal replacement network is crossbar exchanged as the two-dimensional optical signal, and the arithmetic processing unit, the data storage unit and the two
Since each two-dimensional optical signal replacement network is provided with an optical crossbar exchanging section for outputting as a two-dimensional optical signal, each arithmetic element of the arithmetic processing section is optimal for a given data processing.
Data can be simultaneously written to and read from any storage element of the data storage unit via the three-dimensional optical signal replacement network. Therefore, each arithmetic element can exchange data in parallel via another arbitrary arithmetic element and an arbitrary storage element.

Further, although this parallel arithmetic device requires a plurality of two-dimensional optical signal permutation networks, since these two-dimensional optical signal permutation networks are simple optical systems that do not require control, the arithmetic operation of the arithmetic processing unit is performed. It is possible to increase the number of elements and storage elements of the data storage unit, that is, the degree of parallelism of the device.

As described above, a parallel arithmetic unit capable of realizing optimum two-dimensional optical signal replacement for given data processing without using a two-dimensional optical signal replacement network with a complicated optical system is provided. be able to.

[Brief description of drawings]

FIG. 1 is a schematic configuration diagram showing a parallel arithmetic device according to the present invention.

FIG. 2 is a schematic configuration diagram showing a conventional parallel arithmetic device.

FIG. 3 is a configuration diagram showing an arithmetic element of a conventional parallel arithmetic device.

FIG. 4 is a configuration diagram showing a storage element of a conventional parallel arithmetic device.

[Explanation of symbols]

 31 parallel arithmetic device 2 arithmetic processing unit 3 data storage unit 5 arithmetic element 6 storage element 32 two-dimensional optical signal replacement network 33 optical crossbar exchange unit 34 two-dimensional optical signal

Claims (1)

[Claims] 1. An arithmetic processing unit comprising a plurality of arithmetic elements having a function of transmitting and receiving an optical signal arranged two-dimensionally, and a plurality of storage elements having a function of transmitting and receiving an optical signal two-dimensionally. An arrayed data storage section, a plurality of two-dimensional optical signal permutation networks that rearrange the positions of the two-dimensionally arrayed optical signals in two dimensions based on a preset rule, and the arithmetic processing section, The two-dimensional optical signal output from each of the data storage unit and the two-dimensional optical signal replacement network is crossbar-exchanged as the two-dimensional optical signal, and the arithmetic processing unit, the data storage unit, and
A parallel arithmetic device, comprising: an optical crossbar switching unit that outputs a two-dimensional optical signal to each of the two-dimensional optical signal replacement networks.