JPH05119275A - Coupling network between optical processors - Google Patents

Abstract

PURPOSE:To enable microprocessors on different boards to easily make a communication by making arrays of light emitting elements on a plane light reception and a transmission device to correspond to the microprocessors and wiring respective light receiving and emitting elements so that they correspond to the microprocessors one to one. CONSTITUTION:All the light emitting and receiving elements are connected to (m) microprocessors on the boards. Optical spatial propagation is used for the communication among the boards. The signal light emitted from the reverse surface of the board 1-0 is received on the top surface of the board 1-1 or multiplexed with the signal light emitted by the board 1-1 and projected on the board 1-2 from the reverse surface of the board 1-1. This is repeated to sent the signal to a board tube and the signal light emitted from the tail board 1-(n-1) is reflected by mirrors 3-2, 3-4, and 3-1 and made incident on the top surface of the board 1-0. Consequently, the communication among the processors is made.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a network for connecting a plurality of microprocessors in a massively parallel computer and an optical device enabling the construction thereof.

[0002]

2. Description of the Related Art Recent advances in computer processing capacity have been remarkable due to research on semiconductor technology and architecture. As one of the methods for improving the processing capacity that has been actively researched, there is a method of performing processing in parallel using a plurality of microprocessors. In the conventional implementation of a computer with multiple microprocessors, it is difficult to share jobs among processors, effectively use peripheral resources, transfer processing results, and compensate for data validity. Parallel operation was the mainstream,
Recent research has revealed that this problem can be overcome and a massively parallel computer can be realized in which thousands to tens of thousands of microprocessors are operated in parallel.

However, when many microprocessors operate in parallel, a bus, which is a communication path between the microprocessors, becomes a problem. In a massively parallel computer, it is physically difficult to arrange all the microprocessors on one board, and it is disadvantageous in terms of wiring. Therefore, the microprocessors are distributed and mounted on a plurality of boards. Therefore, it is important to provide a bus between boards that enables smooth data exchange between microprocessors on different boards. Recent high-performance microprocessors have high operating speeds, and the input and output data are high-speed and large-capacity data.
In a high-performance massively parallel computer, the bus also needs to be capable of high-speed and large-capacity communication.

Conventionally, a backplane (motherboard) and a cable have been used for mounting a system composed of a plurality of substrates. However, if a lot of processors are mounted on the board like a massively parallel computer and it is necessary to provide a lot of communication paths with the processors on another board, the conventional method can take out from one side of the board to the backplane or cable. Since the number of signal lines is limited, a sparse coupling topology is used, or the number of wires is reduced by reducing the bit width of the communication path, so that sufficient communication performance may not be obtained.

Further, in the case of electrical wiring, in wiring with a distant board in a large-scale system such as a massively parallel computer, timing restrictions are severe as compared with the inside of the board.
It is difficult to increase the frequency because it is easily affected by noise. Therefore, it is necessary to increase the number of wirings to secure the communication performance, but it is difficult to provide too many electric wirings to the backplane as described above. When connecting using a cable, a large number of line drivers are required, and electrical wiring has drawbacks from the viewpoint of high-density mounting and energy consumption. Further, as shown in FIG. 6, the data input / output of the microprocessors on the boards where the buses are adjacent to each other are wired by conducting wires. However, in this case, the wiring between boards becomes large, high-speed data communication is difficult, and there is a big drawback that many microprocessors must be relayed to perform arbitrary communication between microprocessors.

On the other hand, there are great advantages in using light for communication, such as a wide band, high speed with little signal rounding and delay due to stray capacitance, no need for a ground line, and resistance to noise.

Therefore, an optical LAN has been put into practical use in recent years. There is also an example in which inter-board communication is performed using a fiber bundle on a digital exchange computer, and a digital optical link using several optical fiber bundles has been put to practical use. However, it is difficult to dramatically exceed the transfer rate of a thousand medium-speed electric signals that run on an electric backplane with a small number of optical signals like these. In addition, even if a large number of electric signals can be replaced by a small number of optical signals due to the ultra-high speed, there is a hardware cost for multiplexing and distributing the signals.

There are also examples in which a processor-to-processor coupling network of a parallel computer is constructed by using a light beam propagating in free space. However, they are replaced by holograms, a receiver is manually moved, or a liquid crystal switch is used for the configuration. Since it is switched, there is a drawback that it takes time to connect the line.

In addition to this, the concept of using a backplane in which an optical waveguide is formed has been reported. When using the backplane, there is a drawback that the size of the maximum configuration is suppressed by the size of the backplane.

Further, it has been reported that the light emitting element array chip and the light receiving element array chip are opposed to each other and the inter-processor coupling network is constructed by using a light beam propagating in free space. However, in this case, even if direct communication can be performed between the two substrates, it is difficult to make communication paths with more substrates by light.

There is also an example of constructing a broadcast bus by reflecting a light beam having a spread that propagates in free space with a cylindrical mirror. However, since the reflection angle is limited, there is not much space between substrates unless the mirror is placed far away. However, there is a drawback in that the alignment becomes difficult as the substrate is separated from the mirror because it cannot be connected. In addition, this system has a bus configuration in which the logical sum of the optical signals output from the processor is taken, and its performance is much lower than that of crossbar coupling.

[0012]

As described above, the number of wirings between the boards by electricity is limited, and there are drawbacks such as low reliability due to noise and difficulty in speeding up in terms of frequency. An object of the present invention is to increase the inter-substrate transfer frequency as compared with the electric type, suppress the reliability deterioration due to noise, and dramatically increase the number of inter-substrate wirings. further,
The number of connectable boards is increased by using a connection network including HyperCup and a small amount of hardware, and then a connection network with a short distance between processors is realized. A high-density connection network for massively parallel computers with a dense topology close to a crossbar network is provided, and high-speed communication and flexible communication between processors are also possible.
In addition, broadband ISDN that requires more than 10 gigabits per second
Development of super parallel type InGaAs / InAlAs APD (Avalanche multiplication photodetector) and GaAlAs / GaAs superlattice laminated structure active layer forming surface active laser for ultra high speed image information optical communication in (integrated digital communication network) Optical computer image information Solving the problem of optical communication.

[0013]

According to a first aspect of the present invention, in an optical inter-processor coupling network, light receiving / transmission is provided which has an array of a plurality of marix-shaped surface-type optical elements at one place on a board. An optical device having a modulation function is an optical device that can output received information from the back surface of the device, and the optical device is mounted on a substrate area through which light can pass in a direction perpendicular to the substrate surface. Of the input light beam from the surface type optical element at a predetermined position of the optical device of
At least one of the intensities is controlled by the input information to the first optical device, outputs a light beam that is modulated by applying a bias, and receives the light received at a predetermined position of the facing second optical device. A light beam bundle formed by repeating a mode in which an optical input signal is received as it is or is relayed / modulated while being received by a modulation element, and at least the intensity is increased and a light beam is output toward a third optical device. An optical ring bus is formed by guiding the first optical device back to the first optical device, and using this optical ring bus,
The processor located at the two-dimensional coordinates (X, Y) is X
The optical information communication from the processor having an arbitrary position coordinate in which either one of the coordinates and the Y coordinate matches can be received, and exchange is performed with at least one of the X coordinate and the Y coordinate. It is characterized in that the other remaining coordinate makes it possible to receive optical information from the processor having position coordinates of arbitrary values.

According to the second aspect of the invention, in the optical interprocessor coupling network, the port for receiving the information received by the light receiving element in the Xth row (digit) and the information received by the light receiving element in the Yth column are received. An optical device having a port is used, an optical device for reading in the X direction and an optical device for routing in the Y direction are provided, and the coordinates of the processor on the same substrate are X,
The optical devices for the X-direction routing and the optical devices for the Y-direction routing are of the same type by arranging so that they are different in Y.

In the third invention, in the optical inter-processor coupling network, when the information is transmitted from the processor located at the two-dimensional coordinates (a, b) to the processor located at the coordinates (c, d), the coordinates (c, routing to relay the processor located in d) and the coordinate (a, d) when transmitting information from the processor located in two-dimensional coordinates (a, b) to the processor located in coordinate (c, d). ) Can be routed to relay the processor located in.

According to a fourth aspect of the present invention, in an optical inter-processor coupling network, two optical fiber support portions, each of which has end faces of a plurality of optical waveguides arranged circumferentially around a rotation axis, are made to face each other with their rotation axes aligned with each other. An optical switch that switches the coupling mode of optical fiber groups by coupling optical fiber groups and rotating them around a rotation axis, and the optical switch is provided in the coupling section between the processor and the optical device. And

[0017]

[Function] An array of a plurality of surface-type optical elements arranged in a marix shape is provided at one place on the board (an optical device having a light receiving / transmitting / modulating function can output the received information from the back surface of the device). The optical device is mounted on a substrate area through which light can pass in a direction perpendicular to the substrate surface, and the phase and wavelength of the input light beam from the planar optical element at a predetermined position of the first optical device. , At least one or more of the intensity is modulated and output by applying a bias that is controlled by the input information to the first optical device, and the light beam is received at a predetermined position of the second optical device. A light beam bundle formed by repeating a mode in which an optical input signal is received as it is or is relayed / modulated while being received by a modulator and at least the intensity is increased to output a light beam to a third optical device. An optical ring bus is formed by guiding the optical device back to the first optical device, and this optical ring bus can be used to source any device in the devices on the board. When the device is the source, the second optical device can output the received information from the back side of the device, so the information from the first device is transmitted to the optical device on the third substrate facing the second substrate. To be done.

Since the optical elements are arrayed, such light beams form a bundle and penetrate the substrate. A ring bus is formed by penetrating the bundle of light beams created in this way so as to return to the first optical device, so that as the light beams propagate one after another, the substrate of an arbitrary transmission source will eventually be transmitted. The light beam can reach.

Each device can capture the information from the first device by outputting a light beam toward the optical device on the facing substrate and receiving the light beam by the light receiving element. As described above, according to the first and second aspects of the present invention, it is possible to install a large number of optical high-speed and noise-resistant buses that enable communication between arbitrary substrates.

Here, according to the first invention, two-dimensional coordinates (X,
When the processor positioned in Y) can receive the optical information from the processor having the position coordinates where one of the coordinates matches, the communication to any processor is performed twice, that is, the transfer for adjusting the X coordinate and the transfer for adjusting the Y coordinate. Can be achieved by transferring.

In the hypercube, since the processors whose Hamming distances of either X or Y coordinates are 1 (1 bit is different in binary display) are connected, one of the coordinates is displayed in binary. The method of the first aspect of the present invention in which even if all bits are different includes a hypercube. This versatility is high.

Further, even if the range of possible values of the X coordinate and the Y coordinate is different, for example, 0≤X≤31, 0≤Y≤63, the above property is maintained, so the total number N of processors is not always square. Not limited to numbers.

When used in the first invention, the processor (X,
When the element (X, Y) on the optical device is caused to emit light when Y) transmits, the light emitting element in the Xth column is the processor (X, Y).
The optical information transmitted by *) is transmitted, and the light receiving elements in the Yth row receive the optical information transmitted by the processor (*, Y). Therefore, if the port for taking out the optical information received by the light receiving element in the Xth row and the port for taking out the optical information received by the light receiving element in the Yth column are connected to the processor (X, Y), the coordinates of one of them coincide. Since the optical information from the processor having the position coordinates can be received, the communication to an arbitrary processor can be realized by two transfers, that is, a transfer for adjusting the X coordinate and a transfer for adjusting the Y coordinate. However, since the number of receiving ports is twice as many as the number of transmitting ports, it is difficult to realize when the scale is large.

According to the first invention, since the optical device for X-direction routing and the optical device for Y-direction routing are provided, the number of ports per device becomes smaller than that in the case of applying the first invention. The number of devices increases, but the scale can be increased. Further, since the device for performing inter-processor communication within the board and the device for performing inter-processor communication between the boards can be separated, the intra-board communication can be performed independently of the inter-board communication.

According to the second aspect of the invention, since the processors on the same board are arranged so that the coordinates of X and Y are different from each other, all inter-processor communication is inter-board communication. Therefore, when the second invention is applied, only one processor is connected to the reception ports of each row and each column. When the first invention is applied and different optical devices are used for each direction, the same type of optical device can be used for X-direction communication and Y-direction communication.

According to the third invention, two-dimensional coordinates (a,
When transmitting information from the processor positioned in b) to the processor positioned in coordinates (c, d), first, a, which is the X coordinate of (a, b), is changed to c (c, b).
Is optically connected to (a, b), so information can be transferred directly. Next, the Y coordinate b of (c, b) is changed to d (c,
Since d) is optically connected to (c, b), information can be directly transmitted. Thus, the information transmitted by (a, b) is (c,
can be transferred to d).

If the second invention is used, two-dimensional coordinates (a,
When transmitting information from the processor located at b) to the processor located at coordinates (c, d), first the Y coordinate b of (a, b) is changed to d (a, d) is (a, b). ), It is possible to transfer information directly because it is optically connected. Next, since (c, d) in which a, which is the X coordinate of (a, d), is changed to c is optically connected to (a, d), information can be directly transferred. Thus, the information transmitted by (a, b) can be transferred to (c, d).

According to the fourth aspect of the invention, the n sets of optical waveguides are changed, and the two optical fiber support portions arranged on the circumference around the rotation axis have their rotation axes aligned with each other so that their end faces face each other. A pair of optical fiber groups can be combined, and the channel i of the optical fiber can be switched so as to be shifted to the channel modulo (i + k) by rotating about the rotation axis.

If the fourth invention is applied, since the optical switch of the fourth invention is provided in the coupling portion between the processor and the optical device, the channel i is changed to the channel modulo (i + k) by changing k for each substrate. It is possible to switch to shift, and even when the first invention is applied, the same type of substrate should be manufactured, which reduces the manufacturing cost of the substrate,
It is easy to change the position of the board.

In the connection to which the second invention is applied, the processor positioned at the two-dimensional coordinates (X, Y) exchanges one coordinate with the X coordinate and the Y coordinate, and the other coordinate has an arbitrary value as the position coordinate. Since the processor in 1 is coupled, the coordinate exchange operation and the coordinate value alignment are performed in one transfer. Therefore, when the transfer is performed twice, the coordinates are exchanged twice, so that the coordinates are arranged in the original order and the two coordinate values can be aligned, so that arbitrary interprocessor communication can be performed by the two transfers.

When the first invention is applied and the range of possible values of the X coordinate and the Y coordinate is matched, one processor is 2
While it is necessary to be able to receive optical information from N sets of optical sets, when the second invention is applied, it is only necessary to be able to receive optical information from N number of processors, so the structure of the optical device can be simplified. , The number of optical devices can be halved.

However, when the second invention is applied, the first
Different from the case of applying the invention of (1), the range of possible values of the X coordinate and the Y coordinate must match. Therefore, the number of processors is limited to square. Hypercube is not included.

If the third invention is applied, two-dimensional coordinates (a,
When information is transmitted from the processor located in b) to the processor located in coordinates (c, d), first, a, which is the X coordinate of (a, b), is changed to c, and X and Y
Since information (b, c) whose coordinates have been exchanged is optically connected to (a, b), information can be directly transferred. Next, the X coordinate b of (b, c) is changed to d, and the X coordinate and the Y coordinate are exchanged (c,
Since d) is optically connected to (b, c), information can be directly transferred. Thus, the information transmitted by (a, b) is (c,
can be transferred to d).

If the third invention is applied, two-dimensional coordinates (a,
When transmitting information from the processor located in b) to the processor located in coordinates (c, d), first, the Y coordinate b of (a, b) is changed to d, and the X coordinate and Y coordinate are set.
Since (d, a) whose coordinates have been exchanged is optically connected to (a, b), information can be directly transferred. Next, the Y coordinate a of (d, a) is changed to c, and the X coordinate and the Y coordinate are exchanged (c,
Since d) is optically connected to (d, a), information can be directly transferred. Thus, the information transmitted by (a, b) is (c,
can be transferred to d).

The operation of the optical ring bus in the optical interprocessor coupling network will be described below.

The optical ring bus uses an optical signal as a signal. Therefore, by using the good focusing property of the light beam, for example, even after the spatial propagation between the boards, 100 μ
It becomes possible to transmit a signal to only one of the light receiving elements arranged at intervals of m or less. Therefore, wiring between boards becomes unnecessary, leading to labor saving of wiring labor and improvement of maintainability. In addition, band limitation of electric wiring is likely to occur due to capacitance between wirings and crosstalk occurs between wirings. However, when optical wiring is used, these limitations do not occur, so high-speed and high-performance communication can be easily realized.

Further, by using a planar light emitting / receiving device in which light emitting / modulating elements are arranged on one surface and light receiving is collected on the other surface in a matrix form, light emitting / light receiving / modulation which is often required for inter-board communication is used. The mounting wiring of the element can be simplified and the device can be easily manufactured, and the increase in the number of wiring does not pose any problem. By increasing the wiring of the board, the flexibility of the communication method and procedure in this system can be improved. Furthermore, the technology of optical amplifiers and optical regenerators is used to combine the functions of light receiving elements, light emitting elements, and modulation to provide the function of relaying received signals, which requires less delay time than the communication between microprocessors on adjacent boards. realizable.

In the optical ring bus of the present invention, each row of the light emitting elements on the planar light receiving / light emitting device as defined in claim 1 is made to correspond to each microprocessor on the board, and at the same time the light receiving / modulating elements are arranged. By wiring each element in a one-to-one correspondence with each microprocessor on each board, it becomes possible to easily realize communication between the microprocessors between the boards.

Further, in the optical ring bus of the present invention, when the X column of the light emitting elements on the planar light emitting / receiving / modulating device as defined in claim 1 corresponds to the Xth microprocessor on the board. At the same time, the light receiving element portion is wired so that the element in the X row and the Y column corresponds to the Xth microprocessor on the Yth board, and the bth processor on the ath board to the dth processor on the cth board. When sending data to the processor, the c-th processor of the b-th board can be used as a relay to easily realize communication between arbitrary microprocessors between arbitrary boards.

[0040]

EXAMPLES Examples of the present invention will be described below.

FIG. 1 shows an embodiment of an m × n processor massively parallel computer using the present invention. n boards 1 to 0 with m microprocessors mounted on each board
(N-1) are arranged vertically. Each board has a planar light emitting / receiving device 2-0 to 2- (n-1). On this device, light emitting / receiving elements are arranged in an m × n matrix as shown in FIG.
By thus concentrating the optical elements required for inter-board communication in one place, it is possible to realize a device in which the elements are integrated, which facilitates mounting and wiring. The planar light emitting / receiving device can be placed at any position on the board, but considering the wiring with the microprocessor and how to connect the terminals from the device, it is best to place it in the center of the board as shown in the figure. There is. All the light emitting / receiving elements are connected to any of the m microprocessors 4 on the board. Communication between boards is performed by optical space propagation. Board 1-
The signal light emitted from the lower surface of 0 is received by the upper surface of the board 1-1 or is combined with the signal light generated in the board 1-1 and emitted from the lower surface of the board 1-1 toward the board 1-2. By repeating this, the signal is transmitted between the boards,
The signal light emitted from the last board 1- (n-1) is reflected by the mirrors 3-2, 3-3, 3-4, and 3-1 so as to be incident from the upper surface of the board 1-0. .. As a result, the bundle of signal lights between the light emitting / receiving elements can trace the locus 4 as shown in FIG. 1, and communication between the microprocessors can be realized.

The communication method between arbitrary microprocessors on different boards is as follows. Now board 1
-2 microprocessor No. 2 4 (hereinafter, referred to as processor [2, 4]) to the microprocessor Nos. 1 to 5 of the board 1-5. 3 (Similarly, data transmission to the processors [5, 3] will be taken as an example below.

Assuming that the device of FIG. 3 is a planar light emitting / receiving device on a board, each microprocessor [i,
j] is wired so as to control the light emitting element in a specific Tj column and the light receiving element in the position of (row Rj, column Ti) of the planar light emitting / light receiving device.

For example, if the device of FIG. 3 is a planar light emitting / receiving device on the board 1-2, the processor [2
4] can be transmitted using any element in the column T5 of the device, and can receive the optical signal reaching the light receiving element in (row R4, column T2). The processors [5, 3] are similarly connected to the surface emitting / receiving devices T3 and (row R3, column T5) on the board 1-5.

The processor [2, 4] performs conversion processing of data to be transmitted to the processor [5, 3] into transmission data to which recipient information and the like are added (row R5, column T).
Using the light emitting element at the position 4), the light is emitted onto the planar optical bus. FIG. 4 is a simple illustration of this situation. In the figure, 6-1 designates the transmission side processor [2, 4] as 6-
Reference numeral 2 denotes a receiving side processor [5, 3], 6-3 denotes a relay processor [4,5], and a dotted line 7 denotes signal light transferred between boards.

The transmitted signal light is relayed by the element at the position (R5, T4) of the planar light emitting / receiving device of the board 1-3 and (R5 of the planar light emitting / receiving device of the board 1-4. The light is received by the light receiving element at T4), and the data reaches the processor 6-3 [4,5]. Processor 6
-3 sees the recipient information of the received data and knows that it is data to the processor [5, 3]. Therefore, the received data is sent as it is onto the planar optical bus by using the light emitting element at the position of (row R3, column T5). The transmitted signal light is received by the light receiving element at (R3, T5) of the planar light emitting / light receiving device of the board 1-6, and the data reaches the processor 6-2 [5, 3].

The processor [3, 2] is relayed and transmitted to the processor [2, 4] to complete a series of data transmission.
This state is shown in FIG.

The processor [5, 3] performs conversion processing of the ACK signal to be transmitted to the processor [2, 4] into transmission data to which recipient information and the like are added (line R2, R2).
Light is emitted onto the planar optical bus by using the light emitting element in the position of the row T3). The transmitted signal light is relayed by the element at the position (R2, T3) of the planar light emitting / receiving device such as the board 1-2, and the (R2, T3) of the planar light emitting / receiving device of the board 1-3. The light is received by the light-receiving element at, and the data reaches the processor 6-4 [3, 4]. Processor 6
-4 sees the receiver information of the received data and knows that it is data to the processor [2, 4]. Therefore, the received data is sent as it is onto the planar optical bus using the light emitting element located at the position of (row R4, column T2). The transmitted signal light is received by the light receiving element at (R4, T2) of the planar light emitting / light receiving element device of the board 1-3, and the data reaches the processor 6-1 [2, 4].

In the above transmission method, in order to ensure that the signal is transmitted to the receiving side, it is necessary to make an agreement for signal collision in advance. Although it is a stage on the transmitting side, in the above method, the microprocessor No. 4 is a microprocessor No. of any board. Three
For data transmission (including relaying), an optical path is formed from the light emitting / receiving element at the same position (row R5, column T4) on each board to transmit signal light. When a plurality of such transmissions occur at the same time, a plurality of signal lights overlap each other, so that each receiving terminal cannot receive. In order to deal with this situation, the transmitting side prepares a timer, and if there is no reply from the receiving side within a certain time after data transmission, it is judged that a signal collision has occurred and the data is retransmitted. Further, the transmission side can further reduce the probability of signal collision by checking the corresponding optical path before transmission and confirming that there is no other signal light.

In the above description, the inter-microprocessor communication on another board is assumed, but the planar optical bus of the present invention can be used in the same manner as described above for the communication between microprocessors on the same board. It is possible. In this case, there arises a problem that the amount of information transmitted on the optical bus increases and unnecessary delay time increases, but there is also an advantage that wiring on the board and switching elements can be reduced.

FIG. 7 is a diagram showing an embodiment of an inter-processor engagement network of a parallel computer having a parallel optical ring bus to which the first, second and third inventions are applied. In this embodiment, a substrate having four processing elements is shown.

PE00, PE01, PE02, PE03
Is a processing element of X coordinate 0 to 3 of Y coordinate 0, PE10, PE11, PE12, PE13 is a processing element of X coordinate 0 to 3 of Y coordinate 1, PE2
0, PE21, PE22, PE23 are processing elements of X coordinate 0 to 3 on Y coordinate 2, PE30, PE.
31, PE32, PE33 are Y coordinate 3 X coordinate 0 to 3
Processing element.

Since the second invention is applied, the processing elements on the substrate are arranged so that the coordinates of X and Y are different. In this embodiment, the substrate 0 is PE00, PE
11, PE22, PE33 are mounted, and PE3 is mounted on the substrate 1.
0, PE01, PE12, PE23 are mounted, and PE20, PE31, PE02, PE13 are mounted on the substrate 2,
PE10, PE21, PE32, and PE03 are mounted on the substrate 3.

The above-mentioned substrates are fixed so that the positions of the optical devices between the adjacent substrates are matched with each other, and the light output from the optical device 0 on the substrate 0 hits the optical device 1 on the substrate 1 and the substrate 1 The light output from the upper optical device 1 strikes the optical device 2 on the substrate 2, and the light output from the optical device 2 on the substrate 2 strikes the optical device 3 on the substrate 3. The light output from the optical device 3 on the substrate 3 is guided so as to strike the optical device 0 on the substrate 0 via the lens and the mirror.

The optical device comprises a 4 × 4 array of optical elements, a transmission port and a reception port. Each optical element has a transmitting function that emits light according to the input data from the transmitting port, a relay function that outputs light incident in the vertical direction to the device surface from the opposite surface, and a light incident in the vertical direction to the device surface. Has a reception function of selectively capturing and outputting to the reception port.

Communication between the processing elements is performed via the optical devices arranged one by one on the substrate. Thus, each processing element has a port for communicating with the optical device. Since the first invention is applied in this embodiment, the X direction receiving port and the Y direction receiving port are provided in the number of processing elements on the substrate, that is, 4 units each. Furthermore, each optical device has four transmission ports. That is, each processing element is connected to two receiving ports and one transmitting port.

In this embodiment, the position of the optical element on the optical device corresponds to the processing element on the transmitting side.
FIG. 8 shows an example of allocation of processing elements to the optical element array and the receiving port. The number in FIG. 2 is the number of the processing element on the transmission side that the optical element at that position is responsible for transmitting information. In an optical device on a board, a processing element on the board causes the optical element with the same number to transmit optical information via a transmission port. The optical elements at other positions relay the light beam incident from the upper substrate to the lower substrate. In this way, the information output by all processing elements is collected on the optical devices of all substrates.

In this example, only the Y coordinate is different in the row direction and only the X coordinate is different in the column direction. The X direction receiving port selectively receives the optical information of each column, and the Y direction receiving port selectively receives the optical information of each row.

At this time, the X-direction receiving port i has the Y coordinate i.
Is connected to a processing element on the same substrate. Similarly, the Y-direction receiving port i is connected to the processing element on the same substrate whose X coordinate is i.

For example, the X direction receiving port 0 is 00, 01,
Since the optical information at positions 02 and 03 can be selectively received,
If this is connected to the processing element having the Y coordinate of 0 on each substrate, crossbar coupling between the processing elements having the Y coordinate of 0 can be realized.

Similarly, for example, the Y direction receiving port 0 is 00,
Since the optical information at the positions of 10, 20, and 30 can be selectively received, if this is connected to the processing element having the X coordinate of 0 of each substrate, the crossbar coupling between the processing elements having the X coordinate of 0 can be achieved. Can be realized.

With the above construction, the 16 processing elements are base-4 2-cube.
Will be connected in the topology of.

In FIG. 7, the PE 12 is applied by applying the third invention.
3 shows a propagation path of a message when the message is transmitted from PE to PE31. The message output from the transmission port of the PE 12 is input to the transmission port 2 of the optical device on the board 1 to cause the element at the position (1, 2) to emit light. The light is emitted from the optical device on the substrate 2, the optical device on the substrate 3, the lens,
It is transmitted to the optical element at the position (1, 2) of the optical device on the substrate 0 via the mirror. Substrate 0 optical device (1, 2)
The optical element at the position of takes in light to the X direction receiving port 1.
The X-direction receiving port 1 connected to the optical element at the position (1, 2) of the optical device on the substrate 0 is connected to the X-direction receiving port of the PE 11. The PE 11 relays this message and outputs it to the transmission port 1 of the optical device on the board 0. The optical element at the (1,1) position of the optical device on the substrate 0 emits light and is transmitted to the optical element at the (1,1) position of the optical device on the substrate 2 via the optical device on the substrate 1. The optical element at the position (1, 1) of the optical device on the substrate 2 is the Y-direction reception port 1
Capture light into. The Y-direction receiving port 1 connected to the optical element at the (1,1) position of the optical device on the substrate 2 is PE3.
Since it is connected to the Y direction receiving port of 1, P from there
The message is transmitted to E31.

As described above, according to the third aspect of the invention, the processor positioned at the coordinate (c, b) is routed so as to be relayed, so that the arbitrary coordinate (a,
From the processor located in b) any coordinates (c,
Information can be transmitted between the processors located in d).

In FIG. 9, the PE 12 is applied by applying the second invention.
3 shows a transmission path of a message when the message is transmitted from PE to PE31. The message output from the transmission port of the PE 12 is input to the transmission port 2 of the optical device on the board 1 to cause the element at the position (1, 2) to emit light. The light is transmitted to the optical element at the position (1, 2) of the optical device on the substrate 3 via the optical device on the substrate 2. The optical element at the position (1, 2) of the optical device on the substrate 3 takes in light to the Y-direction receiving port 1. The Y direction receiving port 1 connected to the optical element at the position (1, 2) of the optical device on the substrate 3 is PE 32.
It is connected to the Y direction receiving port. The PE 32 relays this message and sends the optical device on the substrate 3 to the transmission port 2
Output to. The optical element at the position (3,2) of the optical device of the substrate 3 emits light, and passes through the lens, the mirror, the optical device of the substrate 0, the optical device of the substrate 1 and the (3,2) of the optical device of the substrate 2. Is transmitted to the optical element at the position. The optical element at the position (3, 2) of the optical device on the substrate 2 is the X-direction reception port 3
Capture light into. The X direction receiving port 1 connected to the optical element at the position (3, 2) of the optical device on the substrate 2 is PE3.
Since it is connected to the X direction receiving port of 1, P from there
The message is transmitted to E31.

As described above, when the second invention is used, the processor positioned at the coordinate (a, d) is routed so as to be relayed, so that the arbitrary coordinate (a,
From the processor located in b) any coordinates (c,
Information can be transmitted between the processors located in d).

FIG. 10 is a diagram showing an embodiment of an interprocessor coupling network of a parallel computer having a parallel optical ring bus to which the first, second and third inventions are applied. Each substrate is equipped with four processing elements and two optical element arrays. In this embodiment, the optical element array is provided with two systems for X-direction routing and Y-direction routing.

FIG. 11 is a diagram showing an example of allocation of processing elements in an optical element array for X-direction routing. The receive port i connects to a processing element on the same board whose Y coordinate is i.

The connection pattern is different for each substrate, and a wiring switching mechanism is required to obtain substrates having the same structure. FIG. 12 shows the structure of a connector type optical rotary switch to which the fourth invention is applied. Four sets of optical fibers are attached to the concave optical connector with the alignment rotary shaft and the convex optical connector with the alignment rotary shaft at equal intervals around the alignment rotary shaft, respectively, and they are mated with each other. Then, the positions of the optical fibers are switched by changing the mutual positions around the rotation axis. A channel alignment protrusion and a mating hole are provided so that the positions of the optical fibers do not shift when mated. A mark indicating the position of the fiber and its channel number is attached to the outer circumference of the connector.

By applying the fourth invention, the X direction connection of the substrate of FIG. 11 can be achieved by applying the fourth invention to those of FIG. 13 (substrate 0), FIG. 14 (substrate 1), FIG. 15 (substrate 2), and FIG. 16 (substrate 3). It can be easily realized by switching the optical rotary switch.

FIG. 17 is a diagram showing an example of allocation of processing elements in an optical element array for routing in the Y direction. The receive port i connects to a processing element on the same board whose X coordinate is i.

As is apparent from FIGS. 11 and 17, the optical element array for X-direction routing and the optical element array for Y-direction routing are structurally the same except for the allocation.

Further, as is clear from the comparison between FIG. 8 and FIGS. 11 and 17, this embodiment simplifies the construction of the optical device as the receiving port of the optical device. That is, this embodiment is more advantageous when it is difficult to manufacture an optical device having a complicated structure. However, the number of transmission ports of the processing element and the number of optical element arrays are larger in this embodiment. As the amount of hardware increases, the processing element can perform transmission in the X direction and transmission in the Y direction independently, so that the communication performance of this embodiment is superior.

In FIG. 10, PE1 is applied by applying the third invention.
2 shows a propagation path of a message when transmitting from 2 to the PE 31. The message output from the X-direction transmission port of the PE 12 is input to the transmission port 1 of the X-direction routing optical device of the board 1 to cause the element at the position (1, 2) to emit light. The light passes through the X-direction routing optical device of the substrate 2, the X-direction routing optical device of the substrate 3, the lens, and the mirror, and the optical element at the position (1, 2) of the X-direction routing optical device of the substrate 0. Be transmitted to. The optical element at the position (1, 2) of the optical device on the substrate 0 takes in light to the X-direction receiving port 1. The receiving port 1 connected to the optical element at the position (1, 2) of the optical device on the substrate 0 is connected to the X-direction receiving port of the PE 11.
The PE 11 relays this message and outputs it to the transmission port 1 of the optical device for Y-direction routing on the board 0. The optical element at the position (1, 1) of the Y-direction routing optical device of the substrate 0 emits light, and passes through the Y-direction routing optical device of the substrate 1 to the (1, 1) of the Y-direction routing optical device of the substrate 2. It is transmitted to the optical element at the position of 1). Board 2
The optical element at the position (1, 1) of the Y-direction routing optical device of (1) takes in light to the reception port 1. Since the receiving port 1 connected to the optical element at the position (1,1) of the optical device for Y-direction routing on the board 2 is connected to the Y-direction receiving port of the PE 31, a message is transmitted from there to the PE 31. ..

As described above, according to the third aspect of the invention, the processor positioned at the coordinate (c, b) is routed so as to be relayed, so that the arbitrary coordinate (a,
From the processor located in b) any coordinates (c,
Information can be transmitted between the processors located in d). By using the first invention and the second invention together, it is possible to transfer information between arbitrary processing elements in substantially the same manner.

In this example, the PE 11 transmits a message in the Y direction for relaying, but since the transmission ports in the X direction and the Y direction are independent, the PE 11 may transmit a message in the X direction simultaneously with the relay operation. it can. In the case of this embodiment, since the number of processing elements is as small as 16, the effect is not so remarkable, but when the number of processing elements is large, such as 128 × 128 = 16384, it can be seen from a certain processing element. Processing element with communication distance 1 is 127 × 2 = 254.
On the other hand, since there are as many as 16384-1-254 = 16129 processing elements with a communication distance of 2, the majority of communication between arbitrary processing elements involves a relay operation. Therefore, the effect that the relay and the transmission in the processing element can be performed in parallel becomes remarkable.

The adjacency relationship between the processing elements of the connection network to which the above-described first invention is applied is expressed by a matrix as shown in FIG. The processing elements corresponding to the positions marked with 1 are adjacent to each other, and communication can be performed without relaying other processing elements.
The hypercube combination shown in FIG. 19 is a highly versatile topology that can be applied to parallel processing of various problems such as FFT due to its mathematical beauty, and many parallel algorithms for hypercubes have been developed. When the adjacency relation of the hypercube connection of FIG. 13 is expressed by a matrix, FIG.
However, all the positions marked with 1 in FIG. 20 are also 1 in FIG. That is, it can be said that the connection network to which the first invention is applied is a more versatile connection network including a hypercube connection.

FIG. 21 is a diagram showing an embodiment of an interprocessor coupling network of a parallel computer having parallel optical ring paths to which the second and third inventions are applied. In this embodiment, a substrate having four processing elements is shown.

In this embodiment, PE00 and PE0 are used for the substrate 0.
1, PE02, PE03 are mounted, and PE1 is mounted on the substrate 1.
0, PE11, PE12, PE13 are mounted, and PE20, PE21, PE22, PE23 are mounted on the substrate 2,
PE30, PE31, PE32, and PE33 are mounted on the substrate 3.

In this embodiment, the position of the optical element on the optical device corresponds to the processing element on the transmitting side.
FIG. 22 shows an example of allocation of processing elements to the optical element array and the receiving port. The numbers in FIG. 22 are the numbers of the processing elements on the transmission side that the optical element at that position is responsible for transmitting information. In an optical device on a board, a processing element on the board causes the optical element with the same number to transmit optical information via a transmission port. The optical elements at other positions relay the light beam incident from the upper substrate to the lower substrate. In this way, the information output by all processing elements is collected on the optical devices of all substrates.

In this example, only the Y coordinate is different in the row direction and only the X coordinate is different in the column direction. The receiving port selectively receives the optical information of each column.

At this time, the reception port i is connected to the processing element on the same substrate whose X coordinate is i. The transmit port i connects to a processing element on the same board whose X coordinate is i.

For example, the receiving port 0 is 00, 01, 02,
Since the optical information at the position 03 can be selectively received, if this is connected to the processing element having the X coordinate of 0 of each substrate, the processing element having the X coordinate of 0 can output all the information having the Y coordinate of 0. Optical information from the processing element can be received.

In FIG. 21, PE2 is applied by applying the third invention.
2 shows a propagation path of a message when transmitting from 2 to the PE 31. The message output from the transmission port of the PE 22 is input to the transmission port 2 of the optical device on the board 2 and causes the element at the position (2, 2) to emit light. The light is transmitted to the optical element at the position (2, 2) of the optical device on the substrate 1 via the optical device on the substrate 3, the lens, the mirror, and the optical device on the substrate 0. Substrate 1 optical device (2,2)
The optical element at the position of takes in light to the reception port 2. Board 1
The receiving port 2 connected to the optical element at the position (2, 2) of the optical device is connected to the receiving port of the PE 12. The PE 12 relays this message and outputs it to the transmission port 2 of the optical device on the board 1. The optical element at the position (1,2) of the optical device on the substrate 1 emits light and is transmitted to the optical element at the position (1,2) of the optical device on the substrate 3 via the optical device on the substrate 2. The optical element at the position (1, 2) of the optical device on the substrate 3 takes in light to the reception port 1. Since the receiving port 1 connected to the optical element at the position (1, 2) of the optical device on the board 3 is connected to the receiving port of the PE 31, a message is transmitted from there to the PE 31.

As described above, according to the sixteenth aspect of the invention, the processor positioned at the coordinates (b, c) is routed so as to be relayed, so that the arbitrary coordinates (a,
From the processor located in b) any coordinates (c,
Information can be transmitted between the processors located in d).

The matrix showing the adjacency relationship between the processing elements of this embodiment is as shown in FIG. 23, which is sparser than that of FIG. 18, and is not related to the hypercube coupling of FIG. It is inferior to the connection network to which the first invention is applied. Furthermore, if we focus on two sets of processing elements due to the lack of matrix symmetry, the reverse communication requires different relay processing elements and the paths are different.

Further, since it is more difficult for humans to imagine the topology than in the connection network to which the first invention is applied, it is somewhat difficult to program while being aware of the topology.

However, if the connection is made as in the present embodiment, FIG.
As shown in FIG. 1, the connections between the processing elements and the optical element arrays in all the substrates have the same pattern, which makes the substrates easier to manufacture than the connection network to which the first invention is applied. The structure of the array of optical elements is also simpler than that of the first embodiment, which is advantageous in terms of manufacturing an optical device.

FIG. 24 is a diagram showing an embodiment of interprocessor coupling of a closed computer having a parallel optical ring bus to which the second and third inventions are applied.

In this embodiment, the substrate 0 resembles PE00 and PE1.
0, PE20, PE30 are mounted, and PE0 is mounted on the substrate 1.
1, PE11, PE21, PE31 are mounted, and PE02, PE12, PE22, PE32 are mounted on the substrate 2,
PE03, PE13, PE23, PE33 are mounted on the substrate 3.

In this embodiment, the position of the optical element on the optical device corresponds to the processing element on the transmitting side.
FIG. 25 shows an example of allocation of processing elements to the optical element array and the receiving board. The numbers in FIG. 25 are the numbers of the processing elements on the transmission side that the optical element at that position is responsible for transmitting information. In an optical device on a board, a processing element on the board causes the optical element with the same number to transmit optical information via a transmission port. The optical elements at other positions relay the light beam incident from the upper substrate to the lower substrate. In this way, the information output by all processing elements is collected on the optical devices of all substrates.

In this example, only the X coordinate is different in the row direction and only the Y coordinate is different in the column direction. The receiving port selectively receives the optical information of each column.

At this time, the reception port i is connected to the processing element on the same substrate whose Y coordinate is i. The transmit port i connects to a processing element on the same board whose Y coordinate is i.

For example, the receiving port 0 is 00, 10, 20,
Since the optical information at the positions of 30 can be selectively received, if this is connected to the processing element having the Y coordinate of 0 of each substrate, the processing element having the Y coordinate of 0 can output all the information having the Y coordinate of 0. Optical information from the processing element can be received.

In FIG. 24, PE2 is applied by applying the third invention.
2 shows a propagation path of a message when transmitting from 2 to the PE 31. The message output from the transmission port of the PE 22 is input to the transmission port 2 of the optical device on the board 2 and causes the element at the position (2, 2) to emit light. The light is transmitted to the optical element at the (2, 2) position of the optical device on the substrate 3. The optical element at the position (2, 2) of the optical device on the substrate 3 takes in light to the reception port 2. The reception port 2 connected to the temporary element at the position (2, 2) of the device on the substrate 3 is connected to the reception port of the PE 23. The PE 23 relays this message and outputs it to the transmission port 2 of the optical device on the board 3. The optical element at the position (2,3) of the optical device on the substrate 3 emits light, and is passed through the lens, the mirror, and the optical device at the substrate 0 to the optical element at the position (2,3) of the optical device on the substrate 1. It is transmitted. Substrate 1 optical device (2,3)
The optical element at the position of takes in light to the reception port 3. Board 1
Since the receiving port 3 connected to the optical element at the position (2,3) of the optical device is connected to the receiving port of the PE 31, the message is transmitted from there to the PE 31.

As described above, according to the eleventh aspect of the invention, by routing so that the processor whose position is bloomed at the coordinates (d, a) is relayed, the arbitrary coordinates (a,
From the processor located in b) any coordinates (c,
Information can be transmitted between the processors located in d).

The matrix showing the adjacency relationship between the processing elements of this embodiment is as shown in FIG. 26, which is sparser than that in FIG. 18, and is not related to the hypercube coupling in FIG. It is inferior to the connection network to which the first invention is applied. Further, when there is no symmetry of the matrix, when attention is paid to a set of two processing elements, the communication in the reverse direction requires different relay processing elements, and the path is different.

Further, as compared with the connection network to which the first aspect of the invention is applied, it is difficult for humans to imagine the topology, and thus it is difficult to perform programming while being aware of the topology.

However, if the connection is made as in the present embodiment, FIG.
As shown in FIG. 4, the connections between the processing elements and the optical element arrays in all the substrates have the same pattern, and the substrates are easier to manufacture than the connection network to which the first invention is applied. Since the structure of the optical element array is simpler than that of the first embodiment, it is advantageous from the aspect of manufacturing an optical device.

In the above embodiments, the processing element performs the relay process when the communication distance is 2, but the optical element array may have a configuration in which the light beams are changed. When the optical element array is divided for each direction like the example shown in FIG. 4, the receiving port of the X-direction optical element array is not connected to the processing element, but is connected to the transmitting port of the Y-direction optical element array. It is also possible to take the form of a multistage connection network.

[0101]

As described above, according to the present invention, the restrictions on wiring between boards are greatly relaxed, and a parallel computer-processor connection network having a close connection between processors close to a full crossbar switch and a high transfer speed is constructed. it can. Further, by constructing an optical ring bus using devices in which light emitting / light receiving elements of surface elements are arranged in a matrix, coupling between the elements can be easily obtained, and a high speed and high performance optical ring bus can be realized. Moreover, since the device can be made into a device having a large number of elements by using the integration technique, the number of wirings can be increased and the flexibility of the optical ring bus can be improved. It is possible to realize a highly versatile connection network having a maximum interprocessor distance of 2 including the hypercube connection while having the above-mentioned characteristics, and a connection network having a maximum interprocessor distance of 2 with common hardware between boards in a board.

[Brief description of drawings]

FIG. 1 is a model diagram showing an embodiment of a massively parallel computer using an optical bus according to the present invention.

FIG. 2 is an external view showing an example of details of a microprocessor board in FIG.

FIG. 3 is a diagram showing an array of light emitting / light receiving elements that constitute the planar device in the central portion of FIG.

FIG. 4 is a view showing a microprocessor 6-1 on a board 1-6 from a microprocessor 6-1 on a boat 1-3.
It is the figure which showed the flow of the data to 2.

FIG. 5 shows the microprocessor 6-2 on the board 1-3 from the microprocessor 6-2 on the boat 1-6.
FIG. 3 is a diagram showing a flow of an ACK signal to 1;

FIG. 6 is a diagram showing a conventional multi-microprocessor connection between boats by electrical wiring.

FIG. 7 is a diagram showing an embodiment of routing to which the second invention is applied in an interprocessor coupling network of a parallel computer including a parallel optical ring bus to which the first invention is applied.

FIG. 8 is a diagram showing an example of allocation of processing elements to optical element arrays and reception ports of the coupling networks of FIGS. 7 and 9;

FIG. 9 is a diagram showing an embodiment of routing to which the second invention is applied in an interprocessor coupling network of a parallel computer including a parallel optical ring bus to which the first invention is applied.

FIG. 10 is a diagram showing an embodiment of a sorting system to which the second invention is applied, with the interprocessor processor network of the parallel computer including the parallel optical ring bus to which the first invention is applied.

11 is a diagram showing an example of allocation of processing elements to an optical element array for X-direction routing and a receiving port of the coupling network of FIG.

12 is a view showing the structure of a connector type optical rotary switch to which the invention of FIG. 4 is applied.

FIG. 13 is a diagram showing settings and coupling patterns of the optical rotary switch on the substrate 0 to which the fourth invention is applied.

FIG. 14 is a diagram showing settings and coupling patterns of optical rotary switches on the substrate 1 to which the fourth invention is applied.

FIG. 15 is a diagram showing settings and coupling patterns of optical rotary switches on the substrate 2 to which the fourth invention is applied.

FIG. 16 is a diagram showing settings and coupling patterns of optical rotary switches on a substrate 3 to which the fourth invention is applied.

FIG. 17 is a diagram showing an example of allocation of processing elements to Y-direction routing optical element arrays and reception ports of the connection network of FIG. 10;

FIG. 18 is a matrix showing a connection relationship between processors in the connection network of FIGS. 7, 9 and 10;

FIG. 19 is a diagram showing connections between processors of a hibercube coupling.

FIG. 20 is a matrix showing connections between processors of a hibercube connection.

FIG. 21 is a diagram showing an inter-processor coupling network of a parallel computer including a parallel optical ring bus according to the fourth invention;
Showing an example of routing to which the invention of FIG.

22 is a diagram showing an example of allocation of processing elements to optical element arrays and reception ports of the coupling network of FIG.

FIG. 23 is a matrix showing a connection relationship between processors in the connection network of FIG. 21.

FIG. 24 is a diagram showing an inter-processor coupling network of a parallel computer including a parallel optical ring bus to which the fourth invention is applied;
FIG. 6 is a diagram showing an example of routing according to the present invention.

FIG. 25 is a diagram showing an example of allocation of processing elements to optical element arrays and reception ports of the coupling network of FIG. 24.

FIG. 26 is a matrix showing a connection relationship between processors of the connection network of FIG.

[Explanation of symbols]

 1-0 Board (Microprocessor Board) 0 1-1 Board (Microprocessor Board) 1 1-2 Board (Microprocessor Board) 2 1-3 Board (Microprocessor Board) 3 1-4 Board (Microprocessor Board) 4 1-5 Substrate (Microprocessor Board) 5 1-6 Substrate (Microprocessor Board) 6 1-7 Substrate (Microprocessor Board) 7 2-0 Optical Device 0 2-1 Optical Device 1 2-2 Optical Device 2 2- 3 Optical Device 3 2-4 Optical Device 4 2-5 Optical Device 5 2-6 Optical Device 6 2-7 Optical Device 7 3-1 Mirror 1 3-2 Mirror 2 3-3 Mirror 3 3-4 Mirror 4 4 Optical Path 5-00 Processing Element PE00 5-01 Processing Element PE01 5-02 Process Processing element PE02 5-03 Processing element PE03 5-10 Processing element PE10 5-11 Processing element PE11 5-12 Processing element PE12 5-13 Processing element PE13 5-20 Processing element PE20 5-21 Processing element PE21 5-22 Processing element PE22 5-23 Processing element PE23 5-30 Processing element PE30 5-31 Processing element PE31 5-32 Processing element PE32 5-33 Processing element PE33 6-1 Microprocessor [2,4] 6-2 Microprocessor [5,3 ] 6-3 Microprocessor [4,5] 6-4 Microprocessor [ , 2] 7 Signal light 8 Recessed optical connector 9 Convex optical connector 10 Optical fiber 11 Rotation shaft for alignment 12 Rotation shaft fitting hole 13 Channel alignment protrusion 14 Channel alignment protrusion Fitting hole 15 Position alignment mark when switching channels 17 Optical Rotary Switch 18 Processing Element 19-1 Relay Processor 19-2 Transmit Processor 19-3 Receive Processor 20 Optical Element Array 20-1 X-Directed Optical Element Array for Routing 20-2 Y-Directed Optical Element Array 21 Lens 22- 1 transmission port 1 22-2 transmission port 2 22-3 transmission port 3

─────────────────────────────────────────────────── ───

[Procedure amendment]

[Submission date] October 30, 1991

[Procedure Amendment 1]

[Document name to be amended] Statement

[Name of item to be amended] Claims

[Correction method] Change

[Correction content]

[Claims]

[Procedure Amendment 2]

[Document name to be amended] Statement

[Correction target item name] 0011

[Correction method] Change

[Correction content]

There is also an example of constructing a broadcast bus by reflecting a light beam having a spread that propagates in a free space with a cylindrical mirror , but since the reflection angle is limited, there is not much space between substrates unless the mirror is placed far away. However, there is a drawback in that the alignment becomes difficult as the substrate is separated from the mirror because the substrate cannot be connected. In addition, this system has a bus configuration in which the logical sum of the optical signals output from the processor is taken, and its performance is much lower than that of crossbar coupling.

[Procedure 3]

[Document name to be amended] Statement

[Correction target item name] 0012

[Correction method] Change

[Correction content]

[0012]

As described above, the number of wirings between the boards by electricity is limited, and there are drawbacks such as low reliability due to noise and difficulty in speeding up in terms of frequency. An object of the present invention is to increase the inter-substrate transfer frequency as compared with the electric type, suppress the reliability deterioration due to noise, and dramatically increase the number of inter-substrate wirings. further,
The number of connectable boards is increased by using a connection network including hypercubes and a small amount of hardware, and then a connection network with a small distance between processors is realized. A high-density connection network for massively parallel computers with a dense topology close to a crossbar network is provided, and high-speed communication and flexible communication between processors are also possible.
In addition, broadband ISDN that requires more than 10 gigabits per second
Development of super parallel type InGaAs / InAlAs APD (Avalanche multiplication photodetector) and GaAlAs / GaAs superlattice laminated structure active layer forming surface active laser for ultra high speed image information optical communication in (integrated digital communication network) Optical computer image information Solving the problem of optical communication.

[Procedure amendment 4]

[Document name to be amended] Statement

[Correction target item name] 0013

[Correction method] Change

[Correction content]

[0013]

According to a first aspect of the present invention, in an optical interprocessor coupling network, an optical device having an array of planar optical elements arranged on a board and having a light transmitting / receiving / modulating function is provided. , An optical device capable of outputting the received information from the back surface of the device, the optical device being mounted on a substrate area through which light can pass in a direction perpendicular to the substrate surface, and a predetermined position of the first optical device. At least one of the phase, wavelength, and intensity of the input light beam from the surface-type optical element is controlled by input information to the first optical device, and a light beam that is modulated by applying a bias is output and faced. The optical input signal is received as it is or relayed while receiving light by the light receiving / modulating element at the predetermined position of the second optical device.
A bundle of light beams that is modulated and at least increased in intensity to output a light beam to a third optical device is directed to the first optical device and back to the first optical device. To form an optical ring bus,
Using this optical ring bus, the processor positioned in two-dimensional coordinates (X, Y) can receive optical information communication from the processor in any position coordinate where either one of the X coordinate or the Y coordinate matches. And that at least one of the X-coordinate and the Y-coordinate is exchanged and the remaining other coordinate has a position coordinate of an arbitrary value, which makes it possible to receive optical information from the processor. And

[Procedure Amendment 5]

[Document name to be amended] Statement

[Correction target item name] 0014

[Correction method] Change

[Correction content]

According to a second aspect of the present invention, the optical inter-processor coupling network is provided with a port for receiving information received by the light receiving element in the Xth row and a port for receiving information received by the light receiving element in the Yth column. The optical device for reading in the X direction and the optical device for routing in the Y direction are arranged by using the optical device to be arranged so that the coordinates of the processors on the same substrate are different in both X and Y. The same type of optical device is used for Y-direction routing.

[Procedure correction 6]

[Document name to be amended] Statement

[Correction target item name] 0015

[Correction method] Change

[Correction content]

According to a third aspect of the present invention, in the optical interprocessor network, the coordinates (c) are transmitted when information is transmitted from the processor located at the two-dimensional coordinates (a, b) to the processor located at the coordinates (c, d). , D) are routed to relay the processor located at coordinates (a, b), and when the information is transmitted from the processor located at two-dimensional coordinates (a, b) to the processor located at coordinates (c, d). It is possible to route to relay the processor located in d).

[Procedure Amendment 7]

[Document name to be amended] Statement

[Correction target item name] 0016

[Correction method] Change

[Correction content]

According to a fourth aspect of the present invention, in an optical inter - processor coupling network, two optical fiber support portions, each of which has an end face of a plurality of optical waveguides arranged circumferentially around a rotation axis, have their rotation axes aligned with each other and face each other. It has an optical switch for switching the coupling mode of the optical fiber group by coupling the combined optical fiber groups and rotating them about the rotation axis, and the optical switch is provided in the coupling section between the processor and the optical device. Characterize.

[Procedure Amendment 8]

[Document name to be amended] Statement

[Correction target item name] 0017

[Correction method] Change

[Correction content]

[0017]

[Action] immediately an array of coordinated surface-type optical device on the board
The provided optical device having a light receiving / transmitting / modulating function is an optical device capable of outputting the received information from the back surface of the device, and the optical device is mounted on a substrate region through which light can pass in a direction perpendicular to the substrate surface. Then, the phase of the input light beam from the surface type optical element at a predetermined position of the first optical device, the wavelength,
At least any one or more of the intensity is modulated by the bias application controlled by the input information to the first optical device to output, and the light beam is received and modulated at the predetermined position of the second optical device. The light beam bundle formed by repeating the mode in which the optical input signal is received as it is, or the optical input signal is relayed / modulated and at least the intensity is increased and the light beam is output to the third optical device while being received by the element, Forming an optical ring bus by guiding the optical device back to the first optical device, and this optical ring bus can also be used to source any device in the device on the board. When the first device is used as the transmission source, the second optical device can output the received information from the back surface of the device, so that the optical device on the third substrate facing the second substrate can receive the information from the first device. Information is transmitted.

[Procedure Amendment 9]

[Document name to be amended] Statement

[Name of item to be corrected] 0023

[Correction method] Change

[Correction content]

[0023] With the first invention, the processor (X,
When the element (X, Y) on the optical device is caused to emit light when Y) transmits, the light emitting element in the Xth column is the processor (X, Y).
The optical information transmitted by *) is transmitted, and the light receiving elements in the Yth row receive the optical information transmitted by the processor (*, Y). Therefore, if the port for taking out the optical information received by the light receiving element in the Xth row and the port for taking out the optical information received by the light receiving element in the Yth column are connected to the processor (X, Y), the coordinates of one of them coincide. Since the optical information from the processor having the position coordinates can be received, the communication to an arbitrary processor can be realized by two transfers, that is, a transfer for adjusting the X coordinate and a transfer for adjusting the Y coordinate. However, since the number of receiving ports is twice as many as the number of transmitting ports, it is difficult to realize when the scale is large.

[Procedure Amendment 10]

[Document name to be amended] Statement

[Correction target item name] 0028

[Correction method] Change

[Correction content]

According to the fourth aspect of the invention, the n sets of optical waveguides are modified so that the two optical fiber support portions arranged on the circumference around the rotation axis have their rotation axes aligned with each other and end faces face each other. A pair of optical fiber groups can be combined, and the optical fiber channel i can be switched so as to shift to the channel module (i + k) by rotating about the rotation axis.

[Procedure Amendment 11]

[Document name to be amended] Statement

[Name of item to be corrected] 0029

[Correction method] Change

[Correction content]

If the fourth invention is applied, since the optical switch of the fourth invention is provided in the coupling portion between the processor and the optical device, the channel i is changed to the channel module (i + k) by changing k for each substrate. It is possible to switch to shift, and even when the first invention is applied, the same type of substrate should be manufactured, which reduces the manufacturing cost of the substrate,
It is easy to change the position of the board.

[Procedure Amendment 12]

[Document name to be amended] Statement

[Correction target item name] 0036

[Correction method] Change

[Correction content]

The optical ring bus uses an optical signal as a signal. Therefore, by using the good focusing property of the light beam, it becomes possible to transmit the signal only to one of the light receiving elements arranged at intervals of, for example, 100 μm or less even after the spatial propagation between the boards. Therefore, wiring between boards becomes unnecessary,
This saves wiring labor and improves maintainability. In addition, band limitation of electric wiring is likely to occur due to capacitance between wirings and crosstalk occurs between wirings. However, when optical wiring is used, these limitations do not occur, so high-speed and high-performance communication can be easily realized.

[Procedure Amendment 13]

[Document name to be amended] Statement

[Name of item to be corrected] 0037

[Correction method] Change

[Correction content]

Further, by using a planar light emitting / receiving device in which a light emitting / modulating element is arranged on one side and a light receiving element is collected on the other side in a matrix form, light emitting / light receiving / reception which is often required for inter-board communication is performed. The mounting wiring of the modulation element can be simplified and the device can be easily manufactured, and the increase in the number of wiring does not pose any problem. By increasing the wiring of the board, the flexibility of the communication method and procedure in this system can be improved. Furthermore, by using the technology of optical amplifiers and optical regenerators to combine the functions of light receiving elements, light emitting elements, and modulation elements to provide a relay function for received signals, the delay time is less than the communication between microprocessors on adjacent boards. Can be achieved with.

[Procedure Amendment 14]

[Document name to be amended] Statement

[Correction target item name] 0041

[Correction method] Change

[Correction content]

FIG. 1 shows an embodiment of an m × n processor massively parallel computer using the present invention. n boards 1 to 0 with m microprocessors mounted on each board
(N-1) are arranged vertically. Each board has a planar light emitting / receiving device 2-0 to 2- (n-1). On this device, light emitting / receiving elements are arranged in an m × n matrix as shown in FIG.
By thus concentrating the optical elements required for inter-board communication in one place, it is possible to realize a device in which the elements are integrated, which facilitates mounting and wiring. The planar light emitting / receiving device can be placed at any position on the board, but considering the wiring with the microprocessor and how to connect the terminals from the device, it is best to place it in the center of the board as shown in the figure. There is. All the light emitting / receiving modulators are connected to any of the m microprocessors 4 on the board. Communication between boards is performed by optical space propagation. The signal light emitted from the lower surface of the board 1-0 is received by the upper surface of the board 1-1 or is combined with the signal light generated in the board 1-1 and directed from the lower surface of the board 1-1 to the board 1-2. And emit. By repeating this, the signal is transmitted between the boards, and the signal light emitted from the last board 1- (n-1) is reflected by the mirrors 3-2, 3-3, 3-4, 3-1 and is reflected by the board 1. It should be incident from the upper surface of −0. Thus bundle of signal light between the light-emitting / light-receiving element and Fig. 1
By tracing the locus 4 as shown in FIG. 7 , communication between the microprocessors can be realized.

[Procedure Amendment 15]

[Document name to be amended] Statement

[Correction target item name] 0057

[Correction method] Change

[Correction content]

In this embodiment, the position of the optical element on the optical device corresponds to the processing element on the transmitting side.
FIG. 8 shows an example of allocation of processing elements to the optical element array and the receiving port. The number in FIG. 8 is the number of the processing element on the transmission side that the optical element at that position is responsible for transmitting information. In an optical device on a board, a processing element on the board causes the optical element with the same number to transmit optical information via a transmission port. The optical elements at other positions relay the light beam incident from the upper substrate to the lower substrate. In this way, the information output by all processing elements is collected on the optical devices of all substrates.

[Procedure 16]

[Document name to be amended] Statement

[Correction target item name] Explanation of code

[Correction method] Change

[Correction content]

[Explanation of Codes] 1-0 Board (Microprocessor Board) 0 1-1 Board (Microprocessor Board) 1 1-2 Board (Microprocessor Board) 2 1-3 Board (Microprocessor Board) 3 1-4 Board ( Microprocessor board) 4 1-5 Substrate (microprocessor board) 5 1-6 Substrate (microprocessor board) 6 1-7 Substrate (microprocessor board) 7 2-0 Optical device 0 2-1 Optical device 1 2-2 Optical device 2 2-3 Optical device 3 2-4 Optical device 4 2-5 Optical device 5 2-6 Optical device 6 2-7 Optical device 7 3-1 Mirror 1 3-2 Mirror 2 3-3 Mirror 3 3- 4 Mirror 4 4 Optical Path 5-00 Processing Element PE00 5-01 Processing Element PE01 5 02 Processing element PE02 5-03 Processing element PE03 5-10 Processing element PE10 5-11 Processing element PE11 5-12 Processing element PE12 5-13 Processing element PE13 5-20 Processing element PE20 5-21 Processing element PE21 5-22 Processing Element PE22 5-23 Processing element PE23 5-30 Processing element PE30 5-31 Processing element PE31 5-32 Processing element PE32 5-33 Processing element PE33 6-1 Microprocessor [2,4] 6-2 Microprocessor [5 3] 6-3 Microprocessor [4,5] 6-4 Microphone Processor [3, 2] 7 Signal light 8 Recessed optical connector 9 Convex optical connector 10 Optical fiber 11 Rotation shaft for alignment 12 Rotation shaft fitting hole 13 Channel alignment protrusion 14 Channel alignment protrusion Fitting hole 15 When switching channels Alignment mark 16 Electrical wiring 17 Optical rotary switch 18 Processing element 19-1 Relay processor 19-2 Transmit processor 19-3 Receive processor 20 Optical element array 20-1 Optical element array for X-direction routing 20-2 Optical light for Y-direction routing Element array 21 Lens 22-1 Transmission port 1 22-2 Transmission port 2 22-3 Transmission port 3

[Procedure Amendment 17]

[Document name to be corrected] Drawing

[Name of item to be corrected] Fig. 25

[Correction method] Change

[Correction content]

FIG. 25

 ─────────────────────────────────────────────────── ─── Continuation of front page (72) Inventor Noboru Tanabe 1 Komukai Toshiba-cho, Sachi-ku, Kawasaki-shi, Kanagawa Toshiba Research Institute, Inc. (72) Inventor Yu Nakamura 1 Komukai-shiba, Ko-ku, Kawasaki-shi, Kanagawa Stock company Toshiba Research Institute

Claims (4)

[Claims] 1. An optical device having a function of transmitting / receiving / modulating light, comprising an array of a plurality of surface-type optical elements arranged in a marix shape at one place on a board, wherein the received information is the back surface of the device. Is an optical device capable of outputting from the optical device, the optical device is mounted on a substrate region through which light can pass in a direction perpendicular to the substrate surface, and an input light beam from a planar optical element at a predetermined position of the first optical device. At least one of at least one of phase, wavelength, and intensity is controlled by input information to the first optical device, outputs a light beam that is modulated by applying a bias, and outputs the light beam at a predetermined position facing the second optical device. The optical input signal is formed by repeating the method of receiving the optical input signal as it is or repeating / modulating the optical input signal while receiving the light, and at least increasing the intensity and outputting the optical beam toward the third optical device. Forming a light ring bus by directing the bundle of beams to the first light device and back to the first light device, which is used to create a two-dimensional coordinate (X, X, The processor positioned in Y) is capable of receiving optical information communication from the processor having an arbitrary position coordinate in which one of the X coordinate and the Y coordinate matches, and at least one of the X coordinate and the Y coordinate. An optical inter-processor coupling network characterized in that optical information can be received from a processor which has a position coordinate of an arbitrary value while the remaining coordinate is exchanged with one of the coordinates. 2. The optical inter-processor coupling network according to claim 1, wherein a port for receiving information received by the light receiving element in the Xth row (digit) and a information received by the light receiving element in the Yth column are provided. An optical device for reading in the X direction and an optical device for Y direction routing are used by using an optical device having a receiving port, and the processors on the same substrate are arranged so that the coordinates of X and Y are different, and the routing in the X direction is performed. Optical inter-processor network that uses the same type of optical device for the Y direction and the optical device for the Y direction routing. 3. The optical inter-processor coupling network according to claim 1, wherein coordinates are transmitted when information is transmitted from a processor located at two-dimensional coordinates (a, b) to a processor located at coordinates (c, d). (c, d) is routed so as to relay the processor located at (c, d), and when the information is transmitted from the processor located at two-dimensional coordinates (a, b) to the processor located at coordinates (c, d), the coordinates (a , D) is a routing network for relaying the processor located in step d). 4. The inter-optical-processor coupling network according to claim 1, wherein the two optical fiber support portions, which are arranged circumferentially around the rotation axis of the end faces of the plurality of optical waveguides, have their rotation axes aligned with each other. The optical fiber group is coupled to each other, and an optical switch that switches the coupling mode of the optical fiber group by rotating about the rotation axis is provided.
An optical processor interconnection network comprising the optical switch in a coupling portion between a processor and an optical device.