FR2511216A1 - DATA PROCESSING DEVICE - Google Patents

Abstract

THE INVENTION RELATES TO A DATA PROCESSING DEVICE CONSTITUTED BY SEVERAL MODULES. EACH MODULE CONTAINS A SUB-MATRIX OF DATA ELEMENTS. IN EACH MODULE, THE CONNECTIONS TO AT LEAST TWO ELEMENTS ON ONE SIDE OF THE SUB-MATRIX SHARE A GROUP OF EXTERNAL TERMINALS, WITH THE SAME NUMBER OF ELEMENTS ON ANOTHER SIDE OF THE SUB-MATRIX. THE INVENTION APPLIES IN PARTICULAR TO A DATA PROCESSING DEVICE, INCLUDING SEVERAL DATA PROCESSING ELEMENTS CONNECTED TOGETHER IN A RECTANGULAR MATRIX.

Description

2 5 1 1 2 1 6

  The present invention relates to a data processing device of the type comprising a plurality of data processing elements connected together in one: rectangular matrix Jl area, such that each element can transfer the data to its four closest neighbors Devices of this kind are described for example in British Patent No. 1445714,

1536933, 2020457 and 2019620.

  A device of this kind can easily be made from several identical modules including

  each contains a sub-matrix of the processing elements

  These modules may, for example, be pellets

  advanced integration circuits, or circuit cards.

  printed food on which components are mounted

semi c onduc tors.

  In fact, the number of external terminals required

  on each module depends on the number of processing elements in the module. For example, if the module contains a sub-matrix of 4 x 4 processing elements, it appears that sixteen external terminals (four following each edge of the sub-matrix) are necessary to connect the module to the neighboring modules This comes in addition to the other terminals that are needed, for example

  for control signals, addresses,

  like, etc. however, the number of terminals available on a particular type of module is generally strictly limited (for example a push-in chip includes

  a limited number of pins) This imposes a limit of

  than the number of treatment elements that can

  be incorporated into the same module.

  The aforementioned Patent No. 2,020,457 describes a way

  This patent describes how to rectify this problem.

  to obtain the required number of terminals for connection to neighboring submatrices by combining two neighboring terminals on adjacent edges of the sub-matrix. For example, in the particular embodiment described in this patent, only one terminal (terminal N Wi) is used to transmit data to the north or west according to the value of a switch code In a similar way, another terminal (the terminal SE) transmits the data to the south or to 1 This is

  tion allows to save two terminals per pellet.

  An object of the invention is therefore to propose

  a way to get a bigger saving on the name-

  number of bridges that are needed to connect

neighboring sub-matrices.

  According to one aspect, the invention therefore relates to a data processing device module comprising a matrix N xm processing elements, each processing element being connected to neighboring elements in the matrix, and in which external connections to at least two elements on one side of the matrix

  shared common external terminals with a name

  equal number of elements on a neighboring side of the matrix.

  According to another aspect of the invention, a device

  tif of data processing has several elements

  connected together in a rectangular matrix

  such as each element works to transfer

  data to these neighboring elements, the matrix being sub-

  divided into several modules, each of which contains a sub-

  matrix of N x N processing elements in each model.

  dule, connections to at least two of the elements on

  one side of the sub-matrix sharing ex-

  common terminals with an equal number of elements on a neighboring side of the sub-array, and each shared terminal being externally connected to the two-terminal module

  not shared on the respective neighbor modules.

  Other features and advantages of the invention

  tion will be better understood by reading the description

  following an example of implementation and referring to

  FIG. 1 is an overall view of the device, FIG. 2 shows in more detail a module of the device, FIG.

2 5 1 1 2 1 6

  Figure 3 shows in more detail one of the data processing elements, and

  Figure 4 illustrates a variant of the module.

  FIG. 1 therefore represents a data processing device comprising a plurality of modules 10, each module consisting of a push-in integration chip. Each module comprises three groups of terminals.

  12, 14 and 16 to make a connection with

  neighbors, each group comprising four bounds.

  It should be noted that the llodules are represented

  only with a triangular shape, only for the convenience of representation; in reality, they are norulally arranged in a normal circuit cabinet.

advanced integration.

  The modules are connected together in a mnatrice

  rectangle of rows and co Lonnes, as represen-

  The terminal group 12 of each module is connected to

  by branch circuits 18 with group 14 of the neighboring module in the North direction, and also with group 16 of the neighboring module in the West direction (the terms "North", "East", "South" and "West" are

  used in this description simply to

  to write the logical relationships between modules and

  elements and should not be considered as

  as for a particular physical disposition).

  Figure 2 represents in more detail one of the

  This module contains a sus-matrix li 4 xt 4 of ele

  Each element may be similar to that described in the above-mentioned patents and will therefore not be described in detail. Each element (except for those on the edges of the submatrix) is connected to its four neighbors, the p Read close in the directions of North, South Este and Cuest

  to allow the transfer of data between them

  The transfer between the elements is governed by a routing code that is distributed in parallel to all elements. The routing code consists of two bits, the meaning of which is as follows: Switching Code Direction 0 O North 01 East 10 South 11 West

  As shown in Figure 3, each P elemen-

  holds a processing element 28 and a multiplexer 20 with four inputs 0, 1, 2, 3 connected to receive data from the four neighboring elements, respectively in

  the direction of South, West, North and East.

  The multiplexer 20 is controlled by the switch code on the line 30, so that it selects one of its four inputs as a function of the binary value of the code and the output data to the processing element 28. example, if the referral code is 00, entry 0 is selected so that each element receives

  data from its neighbor to the south.

data transfer to the North.

  The processing element 28 transmits data on the

  32 connection to one of the inputs of each of its ele-

  When the connections transmit

  In both directions, a door is included between each bidirectional connection and the output of the processing element. A door of this type is shown as an example at 34 at the exit of a connection directed towards the East. gate is controlled by the Llage signal code signals on a line 36 so that it is active only to shift the data in one direction

to the east.

  In general, the connections between elements P of a nodule 10 may consist of two unidirectional connections, in which case the doors 34 are not provided in these connections between the elements P Si of

  two-way connections are planned between

  P, doors 341 are provided at each output of the processing element The connections between the modules

1 1216

  are bidirectional and therefore doors 34 are provided between the outputs of the elements P on the edges of the module and the connected terminals 12, 14 and 16

  to bidirectional connections 18 between modules.

  Returning to Figure 2, the four elements

  P on the East side of the sub-matrix are connected

  respectively at terminals 16 while those on the South edge are respectively connected to

  terminals 14 The terminals of the North and West sides of the sub-

  matrix are respectively connected to terminals 12 by

  a group of four bilateral switches 22.

  L Switches 22 have two states, controlled by the second bit of the switch code When the switch code represents North or South (ie

  the second bit -O 0), the switches connect the terminals

  12 On the North side of the sub-matrix Conversely, when the routing code represents West or East (second bit = 1) the switches connect the terminals

12 to 8 West.

  Switches 22 are represented symbolically

  like mechanical switches but in practice

  tick, they are electronic switches made

  in a common way with logical components con-

  naked. For example, when the pointer code

  is O (representing the "North"), the data are

  Each of the elements is shifted from the elements in the north edge of each sub-matrix by the switches 22, the terminals 12 and the branch connections 18 to the sub-matrix. terminals 14 of the neighboring module in the North direction The multiplexers 20 in the elements on the southern edge of the neighboring unit are set to

accept these data.

  It appears that the group of terminals 12 is ef-

  fectively shared between the four elements on the north edge of the submatrix and the four elements on the

11216

  West edge This reduces the number of terminals required for connection with neighboring modules from sixteen to twelve This terminal sharing requires connections from

  connecting the modules so that each

  connection connection connects three module terminals together

  different, but there is no ambience to the direction

  in which the data must circulate because at a given moment, only one of the three terminals transmits

data and only one receives one.

  In the device described above, each of the modules 10 is a large integration chip. In one embodiment, each module may consist of a printed circuit board on which integrated circuit components forming the elements of the circuit are mounted. P treatment and switches 22 In this case, the

  treatment elements P can be made from the

  described in Brebet No. 2,020,457, according to which four treatment elements are incorporated in

  a single integration pellet pushed.

  FIG. 4 shows a module containing four pellets 24 of this kind. As shown in the figure, each chip has six terminals instead of the eight that would normally be necessary for a pellet of this kind. This result is obtained by combining the function of the two terminals. a simple North-West Nile pin and similarly combining the functions of two other terminals into a simp The South West pin S If The way

  this is done is described in more detail in the pre-patent

  city. The module also comprises a group of four switches 26, of equivalent function to those of the switches 22 of FIG. 2, although in this case the switches 26 consist of separate components mounted on the printed circuit board whereas according to FIG. Figure 2, the switches are incorporated on the same

  lozenge as the treatment elements.

  The connections between the processing elements and between the commutators 26 and the edges of the sub-matrix

  treatment elements are equivalent to those

  shown in FIG. 2 It should be noted that the NM pin of the North-West patch is connected to two different switches 26 because this pin is shared and

  therefore, it is at the same time a bro-

  On the northern edge of the sub-mast and to the south as a boundary mark on the west edge, as understood, various modifications may be made by the skilled person to the embodiments

  described and illustrated as examples in no way Lim: i ta-

  without departing from the scope of the invention.

Claims (5)

  1 coo data processing device module: taking a matrix of N x N processing elements, each processing element being connected to the elements   neighbors in the matrix, characterized in that   external nexions to at least two of the elements (P)   on one side of the matrix are shared external bounds   (12) with an equal number of elements (P) on a neighbor side of the matrix.   Data processing device comprising a plurality of processing elements connected together in a rectangular matrix such that each element transfers   data to its neighboring elements, the matrix being sub-   divided into several modules, each of which contains a sub-   matrix N x N of processing elements, characterized in that   at least two of the elements (P) on one side of the sub-   matrix jointly share external terminals (12) with an equal number of elements (P) on a neighboring side of the submatrix and in that each shared terminal (12) is externally connected to the module (b O) at two terminals (14,   16) not shared by respective neighbor modules (l O).   3 Device according to claim 2, characterized   in that the shared external terminals (12) are   connected by switches (26) to the processing elements   (P) on the adjoining edges of the subma-   mutators (26) whose function is to establish a circuit for data, selectively to the elements (P) on   one side and the elements (P) on the neighboring side.   4 Device according to claim 3, characterized   in that the processing elements (P) and the switching   (26) are constituted by an integrated circuit.   Device according to any one of the claims   2 to 4, characterized in that each processing element   (P) comprises a multiplexer (20) for selecting   enter a data entry from the selected one neighboring elements (P).   6 Device according to any one of the   Claims 2 to 5, characterized in that the connections between the processing elements (P) serve to transmit the data in one direction and in the other, doors (34) being provided to allow the transmission of data. data only in selected directions.