WO1993003442A1 - Multiboard multiprocessor connector system - Google Patents

Multiboard multiprocessor connector system Download PDF

Info

Publication number
WO1993003442A1
WO1993003442A1 PCT/US1992/006345 US9206345W WO9303442A1 WO 1993003442 A1 WO1993003442 A1 WO 1993003442A1 US 9206345 W US9206345 W US 9206345W WO 9303442 A1 WO9303442 A1 WO 9303442A1
Authority
WO
WIPO (PCT)
Prior art keywords
coupled
circuit boards
slice
wiring
connections
Prior art date
Application number
PCT/US1992/006345
Other languages
French (fr)
Inventor
Roger Thorpe
Original Assignee
Cambridge Management Corporation
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Cambridge Management Corporation filed Critical Cambridge Management Corporation
Publication of WO1993003442A1 publication Critical patent/WO1993003442A1/en

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F13/00Interconnection of, or transfer of information or other signals between, memories, input/output devices or central processing units
    • G06F13/38Information transfer, e.g. on bus
    • G06F13/40Bus structure
    • G06F13/4063Device-to-bus coupling
    • G06F13/409Mechanical coupling
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F15/00Digital computers in general; Data processing equipment in general
    • G06F15/16Combinations of two or more digital computers each having at least an arithmetic unit, a program unit and a register, e.g. for a simultaneous processing of several programs
    • G06F15/163Interprocessor communication
    • G06F15/173Interprocessor communication using an interconnection network, e.g. matrix, shuffle, pyramid, star, snowflake
    • G06F15/17337Direct connection machines, e.g. completely connected computers, point to point communication networks
    • G06F15/17343Direct connection machines, e.g. completely connected computers, point to point communication networks wherein the interconnection is dynamically configurable, e.g. having loosely coupled nearest neighbor architecture
    • HELECTRICITY
    • H05ELECTRIC TECHNIQUES NOT OTHERWISE PROVIDED FOR
    • H05KPRINTED CIRCUITS; CASINGS OR CONSTRUCTIONAL DETAILS OF ELECTRIC APPARATUS; MANUFACTURE OF ASSEMBLAGES OF ELECTRICAL COMPONENTS
    • H05K1/00Printed circuits
    • H05K1/02Details
    • H05K1/14Structural association of two or more printed circuits

Definitions

  • This invention relates to a connector system for a multiple circuit board system. More specifically, this invention relates to a connector system in a multiboard multiprocessor system, in which the processors are coupled in a two-dimensional array, but the boards are coupled in a line without a backplane.
  • One method of the prior art is to connect the multiple circuit boards with a backplane, i.e., a wire connector which is joined to all of the circuit boards at a common location on each board. While this method of the prior art does connect all of the circuit boards, it has the drawbacks that (1) the backplane may supply only a limited number of connections, (2) the backplane may require a great deal of routing circuitry on each board, and (3) the backplane may be unsuitable for m ⁇ ltiprocessor systems with a rich topology.
  • SUBSTITUTESHEET boards in a line, but forming logical connections as in a two-dimensional array such as a square.
  • Each circuit board may comprise one or more connector ports, each of which may couple to a corresponding port on an adjacent circuit board, by means of a coupling element.
  • Nonadjacent circuit boards may be coupled by means of a logical helix of connection paths, in which data from a plurality of input ports is circularly shifted to a plurality of output ports in a plurality of circuit boards, so that the data is ultimately coupled from a connector port on one circuit board to a corresponding port on a nonadjacent circuit board.
  • a square array of components is coupled with east-west wiring, north-south wiring, row and column wiring, and global wiring.
  • East- west wiring may be made directly between adjacent circuit boards, and may be made by means of a logical helix coupling between nonadjacent circuit boards (such as at the two ends of a row) .
  • North-south wiring may be made by means of a logical helix coupling between circuit boards which are one row apart.
  • Row wiring may be made by means of a plurality of logical helix couplings; column wiring may be made directly and by reference to the row wiring.
  • each circuit board may comprise one or more processors (such as an integrated circuit comprising a square array of processors) , and the connector system may couple the circuit boards in a square array, so as to form a square array of interconnecting processors.
  • processors such as an integrated circuit comprising a square array of processors
  • the connector system may couple the circuit boards in a square array, so as to form a square array of interconnecting processors.
  • at least 4096 processors may be coupled in a 64 x 64 array in a diameter of no more than 13 cm and a volume of no more than 1000 ⁇ - 3 .
  • Figure 1 is a diagram of a multiprocessor system comprising an embodiment of the invention.
  • FIG. 2A is a diagram of a set of global wiring and a set of E-W wiring.
  • Figure 2B is a diagram of a set of E-W wiring in an alternative embodiment.
  • Figure 3A is a diagram of a set of N-S wiring.
  • Figure 3B is a diagram of a set of N-S wiring in an alternative embodiment.
  • Figure 4A is a diagram of a set of row wiring and a set of column wiring.
  • Figure 4B is a diagram of a set of row wiring and a set of column wiring in an alternative embodiment.
  • Figure 5 is a diagram of a set of self-identifi ⁇ cation circuits.
  • Figure 1 is a diagram of a multiprocessor system comprising an embodiment of the invention.
  • a multiprocessor system 101 may comprise a cylindrical processor platform 102 having a plurality of cylindrical slices 103.
  • Each cylindrical slice 103 may comprise a plurality of processors 104 (e.g., embodied in an integrated circuit 105) in a square subarray 106 within a square array 107 of those processors 104.
  • the platform 102 may comprise sixteen slices 103 with logical addresses 0,0 0,1 0,2 0,3 1,0 1,1 1,2 1.3 2,0 2,1 2,2 2,3 3,0 3,1 3,2 3,3 and physically disposed as shown in this table:
  • the slices 103 may comprise an aluminum alloy or similar sturdy heat- resistant material, and may be coupled together with rods or screws 108.
  • Each slice 103 may comprise a plurality of connector slots 109 on each side 110, each of which may be empty or may be filled with an electrical connector 111.
  • a connector 111 may comprise a gold- plated berylliumcopper springy "button connector" (or a “fuzzbutton”, as they are also known in the art), such as a part named “CINAPSE” made by Cinch Corporation of Elk Grove Village, Illinois.
  • CINAPSE Cinch Corporation of Elk Grove Village, Illinois.
  • the connectors 111 may comprise small metal plugs, springs, or tubes.
  • the slots 109 on opposite sides 110 of each slice 103 are not necessarily coupled.
  • the slots 109 of adjacent slices 103 may be aligned to conform to a set of slot positions 112, so that when a first slice 103 and a second slice 103 are in physical proximity, the slots 109 of the two slices 103 may be coupled if both slots 109 are filled with connectors 111 and may be decoupled if one or both slots 109 are empty.
  • Each slice 103 may also comprise a set of coupling circuits 113 for electrically coupling selected slots 109 together, or for coupling selected slots 109 to a set of circuits 114 (e.g., processors 104) on the slice 103.
  • circuits 114 on one slice 103 may be coupled to circuits 114 on an adjacent slice 103 by means of a pair of slots 109 which are filled with connectors ill.
  • circuits 114 on one slice 103 may be coupled to
  • the slices 103 may be disposed physically in line within the platform 102, but the processors 104 on the slices 103 may be disposed logically in a square array within the array 107 of processors 104. It is desirable for each slice 103 to be electrically coupled to each other slice 103 with which it is logically coupled, such as E-W neighbors, N-S neighbors, other slices 103 in the same row or column, and among substantially all slices 103 globally.
  • wiring means electrical coupling between a slice 103 and another slice 103
  • global wiring means wiring among substantially all slices 103
  • E-W wiring means wiring between slices 103 which are E-W neighbors
  • N-S wiring means wiring between slices 103 which are N-S neighbors
  • row wiring means wiring between slices 103 which are in a row
  • column wiring means wiring between slices 103 which are in a column.
  • Figure 2A is a diagram of a set of global wiring and a set of E-W wiring.
  • substantially all slices 103 may be coupled by means of a single slot position 112.
  • the slot 109 for that slot position 112 on each side 110 of each slice 103 may be filled, and the circuits 114 on each slice 103 may couple the slot 109 on the left side 110 of each slice 103 to the slot 109 on the right side 110 of each slice 103, so as to couple all slots 109 for that slot position 112 in each slice 103 together into a global wire 201.
  • sixteen such global wires 201 may carry sixteen bits of information to substantially all slices 103.
  • each slice 103 may be coupled to a left neighbor 202 and to a right neighbor 203 by means of a set of four helical E-W wires 204 in a set of
  • the left side 110 of each slice 103 may be designated "E” and the right side 110 of each slice 103 may be designated "W”.
  • circuits 114 may couple each helical E-W wire 204 from one slot position 112 on the E side 110 to the next modulo-four on the W side 110; thus E-W wire 204 0E is coupled to E-W wire 204 IW, E-W wire 204 IE is coupled to E-W wire 204 2W, E-W wire 204 2E is coupled to E-W wire 204 3W and E-W wire 204 3E is coupled to E-W wire 204 OW, so as to fully couple all four helical E-W wires 204 on each slice 103 with its left neighbor 202 and its right neighbor 203.
  • E-W neighbors occupy adjacent slices 103 and may be coupled directly by means of one of the four slot positions 112.
  • the fourth pair of slices 103 may be coupled by means of one of the four helical E-W wires 204 cycling through all four slot positions 112.
  • slice 103 0,0 is coupled to slice 103 0,1 is coupled to slice 103 0,2 is coupled to slice 103 0,3.
  • the next four slices 103 may use a different one of the four helical E-W wires 204 so that slice 103 0,3 is not coupled to slice 103 1,0. .
  • sixteen sets of E-W wiring may carry sixteen bits of information between E-W neighbor slices 103.
  • Figure 2B is a diagram of a set of E-W wiring in an alternative embodiment.
  • the platform 102 may comprise only four slices 103, with logical addresses 0,0 0,1 1,0 L,l and physically disposed as shown in this table:
  • a set of E-W wiring for an alternative embodiment may be similarly constructed, with two changes. First, no couplings are made between slice 103 0,1 and slice 103
  • circuits 114 on slice 103 0,0 couple E-W wire 204 0E with E-W wire 204 2E and couple E-W wire 204 IE with E-W wire 204 3E, while circuits 114 on slice 103 1,1 couple E-W wire 204 OW with E-W wire 204 2W and couple E-W wire 204 IW with E-W wire 204 3W.
  • Figure 3A is a diagram of a set of N-S wiring.
  • each slice 103 may be coupled to its left neighbor 202 and to its right neighbor 203 by means of a set of four helical N-S wires 301 in a set of four N-S slot positions 112.
  • the left side 110 of each slice 103 may be designated "S” and the right side 110 of each slice 103 may be designated "N".
  • circuits 114 may couple each helical N-S wire 301 from one slot position 112 on the S side 110 to the next on the N side 110; thus N-S wire 301 OS is coupled to N-S wire 301 IN, N-S wire 30115 is coupled to N-S wire 3012N and N-S wire 301 2S is coupled to N-S wire 301 3N, so as to couple three out of four helical N-S wires 301 on each slice 103 with its left neighbor 202 and to couple three out of four helical N-S wires 301 on each slice 103 with its right neighbor 203.
  • N-S neighbors do not occupy adjacent slices 103 and thus may be coupled by means of the four helical N-S wires 301 cycling through all four slot positions 112.
  • slice 103 0,0 is coupled to slice 103 0,1 is coupled to slice 103 0,2 is coupled to slice 103 0,3 is coupled to slice 103 1,0.
  • This pattern is continued unchanged throughout the entire platform 102, so that slice 103 0,0 is coupled to slice 103 1,0 is coupled to slice 103 2,0 is coupled to slice 103 3,0.
  • Three global wires 201 may couple slice 103 3,0 with slice 103 0,0 as if they were left neighbor 202 and right neighbor 203 respectively.
  • sixteen sets of N-S wiring may carry sixteen bits of information between N-S neighbor slices 103.
  • Figure 3B is a diagram of a set of N-S wiring in an alternative embodiment.
  • a set of N-S wiring for an alternative embodiment may be similarly constructed, with one change.
  • two global wires 201 couple N-S wire 301 OS on slice 103 0,0 with N-S wire 3013N on slice 103 1,1 and couple NS wire 301 IS on slice 103 0,0 with N- S wire 301 2N on slice 103 1,1.
  • Figure 4A is a diagram of a set of row wiring and a set of column wiring.
  • each slice 103 may be coupled to its left neighbor 202 and to its right neighbor 203 by means of a set of four helical row wires 401 in a set of four row slot positions 112, in like manner as in a set of E-W wiring, except that all connector slots 109 in all slices 103 are filled, making a fully connected set of four helical row wires 401 which cycle through the four row slot positions 112.
  • Each row comprises one fourth of all the processors 104 and one of the four helical row wires 401.
  • sixteen sets of row wiring may carry sixteen bits of information to each of four rows.
  • the array 107 of processors 104 may be controlled by a control unit 402, which may couple sixteen bits to each of four rows in the array 107, for a total of 64 bits.
  • the control unit 402 may couple the same 64 bits as sixteen bits to each of four columns in the array 107.
  • a set of column wiring may be coupled to a set of row wiring and may comprise two additional slot positions 112 for each slice 103.
  • a set of first column wires 403 may couple all four slices 103 in each column, in like manner as in a set of global wiring, except that every fourth connector slot 109 remains unfilled.
  • slices 103 0,0 0,1 0,2 and 0,3 are coupled; slices 103 1,0 1,1 1,2 and
  • a set of second column wires 404 may couple a selected one of the four helical row wires 401 to the first column wires 403, so as to couple information supplied on the row wires 401 to the four columns.
  • circuits 114 couple the selected one row wire 401 with the second column wire 404 on the right, and couple the second column wire 404 on the left to the first column wire 403 on the right.
  • only one connector slot 109 is filled for each diagonal position in the array 107; thus the connector slot following slices 103 0,01,12,2 and 2,3 are filled (slice 103 2,3 is used because there is no connector slot following slice 103 3,3).
  • column 0 is coupled with row 0
  • column 1 is coupled with row 1
  • column 2 is coupled with row 2
  • column 3 is coupled with row 3.
  • Figure 4B is a diagram of a set of row wiring and a set of column wiring in an alternative embodiment.
  • a set of row wiring for an alternative embodiment may be similarly constructed, with two changes. First, circuits in slice 103 1,1 couple slot position 112 0 with slot position 112 2 and couple slot position 112 1 with slot position 112 3. Second, in the set of second column wires 404, only the connector slots 109 following slice 103 0,0 and slice 103 0,1 are filled.
  • Figure 5 is a diagram of a set of self-identification circuits.
  • each slice 103 may comprise a self-ID circuit 501, each of which is coupled both to circuits 114 on the slice 103 and to the self-ID circuits 501 on the left neighbor 202 (if one exists) and the right neighbor 203 (if one exists) .
  • each self-ID circuit 501 may receive a hexadecimal digit which represents the ID of its slice 103 from a set of four self-ID slot positions 502 coupled to the right neighbor 203, and may subtract one from that ID to
  • SUBSTITUTESHEET determine the ID of its left neighbor 202.
  • Each self-ID circuit 501 may transmit the ID of its own slice 103 to circuits 114 on its own slice 103 and the ID of its left neighbor 202 to the four self-ID slot positions 502 coupled to the left neighbor 202.
  • the four self-ID slot positions 502 for slice 103 3,3 may be coupled to the constant hexadecimal digit "F"
  • Each slice 103 may identify its logical position in the array 107 by noting the two most significant bits of its four-bit ID as its row address and the two least significant bits of its four-bit ID as its column address.
  • connections between slices which are physically made or broken
  • connections between slices could be made with active circuits such as transceiver gates, so as to make those connections dynamically reconfigurable in the face of dynamic conditions.
  • circuits on each slice which are physically made or broken, it would be clear to one of ordinary skill in the art, after perusal of the specification, drawings and claims herein, that circuits on each slice could be made with active circuitry such as transceiver gates, so as to
  • SUBSTITUTESHEET make those circuits dynamically reconfigurable (and so as to make the connectivity of the platform dynamically reconfigurable) in the face of dynamic conditions.

Abstract

A connector system for coupling a plurality of substantially identical circuit boards (103) in a line, but forming logical connections as in a two-dimensional array such as a square. Each circuit board may comprise one or more connector ports (109), each of which may couple to a corresponding port (109) on an adjacent circuit board, by means of a coupling element (111). Nonadjacent circuit boards may be coupled by means of a logical helix of connection paths, in which data from a plurality of input ports is circularly shifted to a plurality of output ports in a plurality of circuit boards, so that the data is ultimately coupled from a connector port on one circuit board to a corresponding port on a nonadjacent circuit board. A square array of components is coupled with east-west wiring, north-south wiring, row and column wiring, and global wiring. Each circuit board may comprise one or more processors (104) (such as an integrated circuit comprising a square array (106) of processors), and the connector system may couple the circuit boards in a square array. At least 4096 processors may be coupled in a 64 x 64 array in a diameter of no more than 13 cm and a volume of no more than 1000 cm3.

Description

DESCRIPTION
Multiboard Multiprocessor Connector System
Background of the Invention
1. Field of the invention
This invention relates to a connector system for a multiple circuit board system. More specifically, this invention relates to a connector system in a multiboard multiprocessor system, in which the processors are coupled in a two-dimensional array, but the boards are coupled in a line without a backplane.
2. Description of Related Art In computer systems with multiple circuit boards, it is advantageous to provide a connector system which does not use too much of the circuit board space for making connections. Where the multiboard system comprises a plurality of processors, the topology may require a large number of connections and thus exacerbate the problem. For example, multiprocessor systems with the processors connected in a square array, in a hypercube, or in other rich topologies have been proposed.
One method of the prior art is to connect the multiple circuit boards with a backplane, i.e., a wire connector which is joined to all of the circuit boards at a common location on each board. While this method of the prior art does connect all of the circuit boards, it has the drawbacks that (1) the backplane may supply only a limited number of connections, (2) the backplane may require a great deal of routing circuitry on each board, and (3) the backplane may be unsuitable for mμltiprocessor systems with a rich topology.
Summary of the Invention The invention provides a connector system for coupling a plurality of substantially identical circuit
SUBSTITUTESHEET boards in a line, but forming logical connections as in a two-dimensional array such as a square. Each circuit board may comprise one or more connector ports, each of which may couple to a corresponding port on an adjacent circuit board, by means of a coupling element. Nonadjacent circuit boards may be coupled by means of a logical helix of connection paths, in which data from a plurality of input ports is circularly shifted to a plurality of output ports in a plurality of circuit boards, so that the data is ultimately coupled from a connector port on one circuit board to a corresponding port on a nonadjacent circuit board.
In a preferred embodiment, a square array of components is coupled with east-west wiring, north-south wiring, row and column wiring, and global wiring. East- west wiring may be made directly between adjacent circuit boards, and may be made by means of a logical helix coupling between nonadjacent circuit boards (such as at the two ends of a row) . North-south wiring may be made by means of a logical helix coupling between circuit boards which are one row apart. Row wiring may be made by means of a plurality of logical helix couplings; column wiring may be made directly and by reference to the row wiring. In a preferred embodiment, each circuit board may comprise one or more processors (such as an integrated circuit comprising a square array of processors) , and the connector system may couple the circuit boards in a square array, so as to form a square array of interconnecting processors. With known technology, at least 4096 processors may be coupled in a 64 x 64 array in a diameter of no more than 13 cm and a volume of no more than 1000 αα-3.
Brief Description of the Drawings
Figure 1 is a diagram of a multiprocessor system comprising an embodiment of the invention.
SUBSTITUTESHEET Figure 2A is a diagram of a set of global wiring and a set of E-W wiring. Figure 2B is a diagram of a set of E-W wiring in an alternative embodiment.
Figure 3A is a diagram of a set of N-S wiring. Figure 3B is a diagram of a set of N-S wiring in an alternative embodiment.
Figure 4A is a diagram of a set of row wiring and a set of column wiring. Figure 4B is a diagram of a set of row wiring and a set of column wiring in an alternative embodiment.
Figure 5 is a diagram of a set of self-identifi¬ cation circuits.
Description of the Preferred Embodiment
An embodiment of this invention may be used together with inventions which are disclosed in a copending application titled "CLOSE-PACKED IMPEDANCE-CONTROLLED CONNECTORS", United States application Serial No. 738,051, Lyon & Lyon Docket No. 192/192, filed the same day in the name of Robert P. Stimson and Richard Jay Lind an, both assignors to the same assignee, hereby incorporated by reference as if fully set forth herein.
Figure 1 is a diagram of a multiprocessor system comprising an embodiment of the invention.
A multiprocessor system 101 may comprise a cylindrical processor platform 102 having a plurality of cylindrical slices 103. Each cylindrical slice 103 may comprise a plurality of processors 104 (e.g., embodied in an integrated circuit 105) in a square subarray 106 within a square array 107 of those processors 104. In a preferred embodiment, the platform 102 may comprise sixteen slices 103 with logical addresses 0,0 0,1 0,2 0,3 1,0 1,1 1,2 1.3 2,0 2,1 2,2 2,3 3,0 3,1 3,2 3,3 and physically disposed as shown in this table:
SUBSTITUTESHEET 0,0 0,1 0,2 0,3
1,0 1,1 1,2 1,3
2,0 2,1 2,2 2,3
3,0 3,1 3,2 3,3 In a preferred embodiment, the slices 103 may comprise an aluminum alloy or similar sturdy heat- resistant material, and may be coupled together with rods or screws 108.
Each slice 103 may comprise a plurality of connector slots 109 on each side 110, each of which may be empty or may be filled with an electrical connector 111. In a preferred embodiment, a connector 111 may comprise a gold- plated berylliumcopper springy "button connector" (or a "fuzzbutton", as they are also known in the art), such as a part named "CINAPSE" made by Cinch Corporation of Elk Grove Village, Illinois. However, it would be clear to one of ordinary skill in the art, after perusal of the specification, drawings and claims herein, that numerous other types of connectors would also be workable, and are within the scope and spirit of the invention. For example, the connectors 111 may comprise small metal plugs, springs, or tubes. The slots 109 on opposite sides 110 of each slice 103 are not necessarily coupled.
The slots 109 of adjacent slices 103 may be aligned to conform to a set of slot positions 112, so that when a first slice 103 and a second slice 103 are in physical proximity, the slots 109 of the two slices 103 may be coupled if both slots 109 are filled with connectors 111 and may be decoupled if one or both slots 109 are empty. Each slice 103 may also comprise a set of coupling circuits 113 for electrically coupling selected slots 109 together, or for coupling selected slots 109 to a set of circuits 114 (e.g., processors 104) on the slice 103. Thus, circuits 114 on one slice 103 may be coupled to circuits 114 on an adjacent slice 103 by means of a pair of slots 109 which are filled with connectors ill. Moreover, circuits 114 on one slice 103 may be coupled to
SUBSTITUTE SHEET circuits 114 on a nonadjacent slice 103 by means of pairs of slots 109 between a plurality of pairs of adjacent slices 103 and by means of circuits 114 on those slices 103. In a preferred embodiment, the slices 103 may be disposed physically in line within the platform 102, but the processors 104 on the slices 103 may be disposed logically in a square array within the array 107 of processors 104. It is desirable for each slice 103 to be electrically coupled to each other slice 103 with which it is logically coupled, such as E-W neighbors, N-S neighbors, other slices 103 in the same row or column, and among substantially all slices 103 globally. As used herein, "wiring" means electrical coupling between a slice 103 and another slice 103, "global wiring" means wiring among substantially all slices 103, "E-W wiring" means wiring between slices 103 which are E-W neighbors, "N-S wiring" means wiring between slices 103 which are N-S neighbors, "row wiring" means wiring between slices 103 which are in a row, and "column wiring" means wiring between slices 103 which are in a column.
Figure 2A is a diagram of a set of global wiring and a set of E-W wiring.
In a set of global wiring, substantially all slices 103 may be coupled by means of a single slot position 112. The slot 109 for that slot position 112 on each side 110 of each slice 103 may be filled, and the circuits 114 on each slice 103 may couple the slot 109 on the left side 110 of each slice 103 to the slot 109 on the right side 110 of each slice 103, so as to couple all slots 109 for that slot position 112 in each slice 103 together into a global wire 201. In a preferred embodiment, sixteen such global wires 201 may carry sixteen bits of information to substantially all slices 103. In a set of E-W wiring, each slice 103 may be coupled to a left neighbor 202 and to a right neighbor 203 by means of a set of four helical E-W wires 204 in a set of
SUBSTITUTE SHEET four E-W slot positions 112. The left side 110 of each slice 103 may be designated "E" and the right side 110 of each slice 103 may be designated "W". Within each slice 103, circuits 114 may couple each helical E-W wire 204 from one slot position 112 on the E side 110 to the next modulo-four on the W side 110; thus E-W wire 204 0E is coupled to E-W wire 204 IW, E-W wire 204 IE is coupled to E-W wire 204 2W, E-W wire 204 2E is coupled to E-W wire 204 3W and E-W wire 204 3E is coupled to E-W wire 204 OW, so as to fully couple all four helical E-W wires 204 on each slice 103 with its left neighbor 202 and its right neighbor 203.
For three out of each four pairs of slices 103, E-W neighbors occupy adjacent slices 103 and may be coupled directly by means of one of the four slot positions 112. The fourth pair of slices 103 may be coupled by means of one of the four helical E-W wires 204 cycling through all four slot positions 112. Thus, slice 103 0,0 is coupled to slice 103 0,1 is coupled to slice 103 0,2 is coupled to slice 103 0,3. The next four slices 103 may use a different one of the four helical E-W wires 204 so that slice 103 0,3 is not coupled to slice 103 1,0. . In a preferred embodiment, sixteen sets of E-W wiring may carry sixteen bits of information between E-W neighbor slices 103.
Figure 2B is a diagram of a set of E-W wiring in an alternative embodiment.
In an alternative embodiment, the platform 102 may comprise only four slices 103, with logical addresses 0,0 0,1 1,0 L,l and physically disposed as shown in this table:
0,0 0,1
1,0 1,1
A set of E-W wiring for an alternative embodiment may be similarly constructed, with two changes. First, no couplings are made between slice 103 0,1 and slice 103
1,0. Second, circuits 114 on slice 103 0,0 couple E-W wire 204 0E with E-W wire 204 2E and couple E-W wire 204 IE with E-W wire 204 3E, while circuits 114 on slice 103 1,1 couple E-W wire 204 OW with E-W wire 204 2W and couple E-W wire 204 IW with E-W wire 204 3W. Figure 3A is a diagram of a set of N-S wiring.
In a set of N-S wiring, each slice 103 may be coupled to its left neighbor 202 and to its right neighbor 203 by means of a set of four helical N-S wires 301 in a set of four N-S slot positions 112. The left side 110 of each slice 103 may be designated "S" and the right side 110 of each slice 103 may be designated "N". Within each slice 103, circuits 114 may couple each helical N-S wire 301 from one slot position 112 on the S side 110 to the next on the N side 110; thus N-S wire 301 OS is coupled to N-S wire 301 IN, N-S wire 30115 is coupled to N-S wire 3012N and N-S wire 301 2S is coupled to N-S wire 301 3N, so as to couple three out of four helical N-S wires 301 on each slice 103 with its left neighbor 202 and to couple three out of four helical N-S wires 301 on each slice 103 with its right neighbor 203.
N-S neighbors do not occupy adjacent slices 103 and thus may be coupled by means of the four helical N-S wires 301 cycling through all four slot positions 112. Thus, slice 103 0,0 is coupled to slice 103 0,1 is coupled to slice 103 0,2 is coupled to slice 103 0,3 is coupled to slice 103 1,0. This pattern is continued unchanged throughout the entire platform 102, so that slice 103 0,0 is coupled to slice 103 1,0 is coupled to slice 103 2,0 is coupled to slice 103 3,0. Three global wires 201 may couple slice 103 3,0 with slice 103 0,0 as if they were left neighbor 202 and right neighbor 203 respectively. In a preferred embodiment, sixteen sets of N-S wiring may carry sixteen bits of information between N-S neighbor slices 103. Figure 3B is a diagram of a set of N-S wiring in an alternative embodiment.
SUBSTITUTESHEET A set of N-S wiring for an alternative embodiment may be similarly constructed, with one change. Instead of three global wires 201 which couple slice 103 3,0 with slice 103 0,0 as if they were left neighbor 202 and right neighbor 203 respectively, two global wires 201 couple N-S wire 301 OS on slice 103 0,0 with N-S wire 3013N on slice 103 1,1 and couple NS wire 301 IS on slice 103 0,0 with N- S wire 301 2N on slice 103 1,1.
Figure 4A is a diagram of a set of row wiring and a set of column wiring.
In a set of row wiring, each slice 103 may be coupled to its left neighbor 202 and to its right neighbor 203 by means of a set of four helical row wires 401 in a set of four row slot positions 112, in like manner as in a set of E-W wiring, except that all connector slots 109 in all slices 103 are filled, making a fully connected set of four helical row wires 401 which cycle through the four row slot positions 112. Each row comprises one fourth of all the processors 104 and one of the four helical row wires 401. In a preferred embodiment, sixteen sets of row wiring may carry sixteen bits of information to each of four rows.
In a preferred embodiment, the array 107 of processors 104 may be controlled by a control unit 402, which may couple sixteen bits to each of four rows in the array 107, for a total of 64 bits. In a preferred embodiment, the control unit 402 may couple the same 64 bits as sixteen bits to each of four columns in the array 107. A set of column wiring may be coupled to a set of row wiring and may comprise two additional slot positions 112 for each slice 103. A set of first column wires 403 may couple all four slices 103 in each column, in like manner as in a set of global wiring, except that every fourth connector slot 109 remains unfilled. Thus, slices 103 0,0 0,1 0,2 and 0,3 are coupled; slices 103 1,0 1,1 1,2 and
8 UBSTITUTΞSHEET 1,3 are coupled; slices 103 2,0 2,1 2,2 and 2,3 are coupled; and slices 103 3,0 3,1 3,2 and 3,3 are coupled.
A set of second column wires 404 may couple a selected one of the four helical row wires 401 to the first column wires 403, so as to couple information supplied on the row wires 401 to the four columns. In each slice 103, circuits 114 couple the selected one row wire 401 with the second column wire 404 on the right, and couple the second column wire 404 on the left to the first column wire 403 on the right. In a preferred embodiment, only one connector slot 109 is filled for each diagonal position in the array 107; thus the connector slot following slices 103 0,01,12,2 and 2,3 are filled (slice 103 2,3 is used because there is no connector slot following slice 103 3,3). Thus column 0 is coupled with row 0, column 1 is coupled with row 1, column 2 is coupled with row 2, and column 3 is coupled with row 3.
Figure 4B is a diagram of a set of row wiring and a set of column wiring in an alternative embodiment. A set of row wiring for an alternative embodiment may be similarly constructed, with two changes. First, circuits in slice 103 1,1 couple slot position 112 0 with slot position 112 2 and couple slot position 112 1 with slot position 112 3. Second, in the set of second column wires 404, only the connector slots 109 following slice 103 0,0 and slice 103 0,1 are filled.
Figure 5 is a diagram of a set of self-identification circuits.
In a preferred embodiment, each slice 103 may comprise a self-ID circuit 501, each of which is coupled both to circuits 114 on the slice 103 and to the self-ID circuits 501 on the left neighbor 202 (if one exists) and the right neighbor 203 (if one exists) . In each slice 103, each self-ID circuit 501 may receive a hexadecimal digit which represents the ID of its slice 103 from a set of four self-ID slot positions 502 coupled to the right neighbor 203, and may subtract one from that ID to
SUBSTITUTESHEET determine the ID of its left neighbor 202. Each self-ID circuit 501 may transmit the ID of its own slice 103 to circuits 114 on its own slice 103 and the ID of its left neighbor 202 to the four self-ID slot positions 502 coupled to the left neighbor 202.
The four self-ID slot positions 502 for slice 103 3,3 may be coupled to the constant hexadecimal digit "F"
(decimal value fifteen), so as to identify slice 103 3,3 with that ID. Successive slices 103 will therefore each be identified with successively smaller IDs, down to hexadecimal digit "0" (decimal value zero) for slice 103
0,0. Each slice 103 may identify its logical position in the array 107 by noting the two most significant bits of its four-bit ID as its row address and the two least significant bits of its four-bit ID as its column address.
Alternative Embodiments
While preferred embodiments are disclosed herein, many variations are possible which remain within the concept and scope of the invention, and these variations would become clear to one of ordinary skill in the art after perusal of the specification, drawings and claims herein.
For example, while the invention has been disclosed with connections between slices which are physically made or broken, it would be clear to one of ordinary skill in the art, after perusal of the specification, drawings and claims herein, that connections between slices could be made with active circuits such as transceiver gates, so as to make those connections dynamically reconfigurable in the face of dynamic conditions.
Moreover, while the invention has been disclosed with circuits on each slice which are physically made or broken, it would be clear to one of ordinary skill in the art, after perusal of the specification, drawings and claims herein, that circuits on each slice could be made with active circuitry such as transceiver gates, so as to
SUBSTITUTESHEET make those circuits dynamically reconfigurable (and so as to make the connectivity of the platform dynamically reconfigurable) in the face of dynamic conditions.
BSTITUTE SHEET

Claims

Claims
1. A connector system for a multiple circuit board system, comprising a plurality of circuit boards, said circuit boards being coupled in a line without a backplane; and at least one processor disposed on each said board, said processors being logically coupled in a two- dimensional array.
2. A connector system, comprising means for coupling a plurality of substantially identical circuit boards in a line; and means for forming logical connections between said circuit boards as in a two-dimensional array.
3. A connector system, comprising a plurality of circuit boards, each said circuit board having at least one connector port, each said port capable of coupling to a corresponding port on an adjacent circuit board by means of a coupling element; wherein a plurality of said ports are disposed in a logical helix of connection paths, whereby data from a plurality of input ports is circularly shifted to a plurality of output ports in a plurality of circuit boards, so that the data is ultimately coupled from a connector port on one circuit board to a corresponding port on a nonadjacent circuit board.
4. A connector system, comprising a plurality of circuit boards coupled in a line; and means for coupling a set of components disposed on said circuit boards with east-west connections, north- south connections, and row and column connections.
5. A connector system as in claim 4, comprising at least one of the following:
SUBSTITUTE SHEET east-west connections which are made directly between adjacent circuit boards; east-west connections which are made by means of a logical helix coupling between nonadjacent circuit boards; north-south connections which are made by means of a logical helix coupling between circuit boards which are one row apart; row connections which are made by means of a plurality of logical helix couplings; and column connections which are made directly and by reference to the row connections.
6. A connector system as in claim 4, wherein each said circuit board comprises at least one processor; and wherein the connector system couples said circuit boards in a square array, so as to form a square array of interconnecting processors.
7. A multiprocessor system comprising at least 4096 processors coupled in a 64 x 64 array in a diameter of no more than 13 cm and a volume of no more than 1000 cm3.
SUBSTITUTESHEET
PCT/US1992/006345 1991-07-31 1992-07-31 Multiboard multiprocessor connector system WO1993003442A1 (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US73821291A 1991-07-31 1991-07-31
US738,212 1991-07-31

Publications (1)

Publication Number Publication Date
WO1993003442A1 true WO1993003442A1 (en) 1993-02-18

Family

ID=24967042

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US1992/006345 WO1993003442A1 (en) 1991-07-31 1992-07-31 Multiboard multiprocessor connector system

Country Status (1)

Country Link
WO (1) WO1993003442A1 (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
DE4408543A1 (en) * 1994-03-14 1995-09-21 Siemens Nixdorf Inf Syst Addressing through permutation

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4709327A (en) * 1983-05-31 1987-11-24 Hillis W Daniel Parallel processor/memory circuit

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4709327A (en) * 1983-05-31 1987-11-24 Hillis W Daniel Parallel processor/memory circuit

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
IEEE MICRO, published October 1986, J.P. HAYES et al., "A Microprocessor-Based Hypercube Supercomputer", see pages 6-17. *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
DE4408543A1 (en) * 1994-03-14 1995-09-21 Siemens Nixdorf Inf Syst Addressing through permutation
DE4408543C2 (en) * 1994-03-14 1998-04-09 Siemens Nixdorf Inf Syst Addressing through permutation

Similar Documents

Publication Publication Date Title
JP3992148B2 (en) Electronic circuit boards for building large and scalable processor systems
US6976064B2 (en) Apparatus and methods for connecting modules using remote switching
US6504841B1 (en) Three-dimensional interconnection geometries for multi-stage switching networks using flexible ribbon cable connection between multiple planes
US5836785A (en) Apparatus and method to uniquely identify similarly connected electrical devices
US5793998A (en) Method and apparatus for interconnection of multiple modules
JPH0833873B2 (en) Method and apparatus for interconnecting processors in a hyperdimensional array
US5715391A (en) Modular and infinitely extendable three dimensional torus packaging scheme for parallel processing
EP0631678B1 (en) Low-neighborhood three-dimensional interconnect
JP3090593B2 (en) Scalable parallel processing network and node interconnection method
JPH06334368A (en) Switchover mid-plane for connecting large number of signals with each other, and apparatus therefor
US20070124565A1 (en) Reconfigurable processing array having hierarchical communication network
US8131903B2 (en) Multi-channel memory connection system and method
US5754777A (en) Apparatus and method for distributed arbitration of shared resources
US7633767B2 (en) Memory modules including SIMM and DIMM-type connection structures and related memory systems
WO1993003442A1 (en) Multiboard multiprocessor connector system
US20060080484A1 (en) System having a module adapted to be included in the system in place of a processor
KR20050059227A (en) Single shelf router
US7187674B2 (en) Method and apparatus for using adaptive switches for providing connections to point-to-point interconnection fabrics
US5497471A (en) High performance computer system which minimizes latency between many high performance processors and a large amount of shared memory
JPH04288659A (en) Processor array
US10185691B2 (en) Two-dimensional torus topology
WO1985003403A1 (en) Data processor arrays and a method of producing them
US5161980A (en) Electrical interconnection of circuit devices
JPH04113445A (en) Parallel computer
WO2007056737A2 (en) Reconfigurable processing array having hierarchical communication network

Legal Events

Date Code Title Description
AK Designated states

Kind code of ref document: A1

Designated state(s): JP

AL Designated countries for regional patents

Kind code of ref document: A1

Designated state(s): AT BE CH DE DK ES FR GB GR IT LU MC NL SE

DFPE Request for preliminary examination filed prior to expiration of 19th month from priority date (pct application filed before 20040101)
122 Ep: pct application non-entry in european phase