• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    This optical arbitration device (12), between conflicting requests for access to a shared resource transmitted by a plurality of N processing nodes (14, 16, 18, 20) of a network-on-a-chip system (10) , comprises at least one main optical arbitration bus (22, 24), at least one optical source (26, 28) for transmitting a first optical signal in the main optical arbitration bus, and a succession of N optical cells arbitration means (30, 32, 34, 36) coupled to the main optical arbitration bus, each of these arbitration optical cells being associated with a processing node and having means for selecting the processing node with which it is associated by diverting the first optical signal. The optical source (26, 28) is arranged to emit a second optical signal propagating in a direction opposite to the first optical signal along the main optical arbitration bus (22, 24). In addition, the selection means of each arbitration optical cell (30, 32, 34, 36) are adapted to select the processing node (14, 16, 18, 20) with which it is associated by diverting the first and second optical signals.
    公开号:FR3014563A1
    申请号:FR1362301
    申请日:2013-12-09
    公开日:2015-06-12
    发明作者:Mounir Zid；Yvain Thonnart
    申请人:Commissariat a lEnergie Atomique CEA；Commissariat a lEnergie Atomique et aux Energies Alternatives CEA；
    IPC主号:
    专利说明:

    [0001] The present invention relates to an optical arbitration device between conflicting requests for access to a shared resource, these conflicting requests being issued by a plurality of N processing nodes of a network-on-a-chip system. It also relates to a corresponding method.
    [0002] It applies in the field of photonics. Photonics is a very promising new technology platform for designing high-performance systems that are increasingly demanding bandwidth. For example, the new multi-processor systems are capable of providing enormous computational power requiring data transfer rates greater than 100 terabytes per second. These data transfer rates may be necessary during data exchanges between processing cores, especially inside a multi-core processor. Optical interconnects are nowadays the only technology capable of transmitting such flows. They can be achieved by optical interconnections thanks to the technique of Wavelength Division Multiplexing (WDM). This technique makes it possible to propagate several optical signals of different wavelengths simultaneously in the same optical waveguide, without any risk of collision between the different signals. Wavelength division multiplexing is an economical solution for increasing the bandwidth capacity of the waveguide. More concretely, photonics is a promising technology for the design of chip-based network systems, since it makes it possible, in addition to increasing the bandwidth, to overcome the problems associated with the high density of electrical interconnections in this type. electronic systems by providing space-saving interconnections. The use of optical interconnects instead of electrical interconnections in integrated circuits has two other advantages. On the one hand, the transmission of an optical signal through a waveguide serves to shorten the transmission delays and thus reduce the attenuation of the transmitted signal. On the other hand, the attenuation of the optical signal propagating in the waveguide is independent of the data transfer rate, which can potentially induce energy savings and performance gains of the circuits thus designed.
    [0003] In order to use optical interconnects in an integrated circuit, the use of a number of optical devices in the integrated circuit is required. This is the case of the following devices: at least one optical source, in particular a laser source, capable of generating and transmitting an optical signal intended to propagate in the optical interconnections, at least one optical modulator, for example implemented in the form of a ring micro-resonator for modulating the optical signal, at least one optical bus, in particular a silicon waveguide, through which the modulated optical signal is transmitted, at least one optoelectronic converter, in particular a photodiode, enabling detecting the optical signal and converting it into an electrical signal upon arrival at destination (ie upon arrival at the input of a processing node).
    [0004] Numerous advances in the field of integrating optical functions into microelectronic chips have recently taken place, opening up innovative perspectives for improving the performance of integrated circuits. Thus, today, more and more optical or optoelectronic devices such as ring micro-resonators, germanium detectors, silicon on insulator waveguides (SOI) , etc., are designed to be able to replace some electrical functions in some microelectronic circuits. These advances made it possible to use optical interconnects for routing data between the processing nodes of a network-on-a-chip system. In practice, an optical switching matrix providing the communication between N processing nodes of a network-on-chip system may consist of an optical bus, in particular a waveguide, and an optical source, in particular external to the system-on-chip, intended to generate the optical signal propagating in this optical bus. This optical signal may comprise a set of N optical signals multiplexed according to the WDM technique described above, each of different wavelength. This set of multiplexed optical signals is also known to those skilled in the art under the name of optical frequency comb. A MWSR (Multiple-Writer Single-Reader) communication scheme may also be used between the N processing nodes, allowing simultaneous access to the optical bus of the N processing nodes for writing. of data to be transmitted and the individual access to this bus of one of these N processing nodes for reading this data. To set up this communication scheme, each of the N processing nodes is sensitive to reading at a single wavelength of the optical frequency comb, this wavelength being characteristic of the optical signals addressed to this node. On the other hand, each processing node is designed to be able to write modulate an optical signal according to the N-1 wavelengths different from that which characterizes it. A processing node can thus modulate and transmit a message to any one of the N-1 other processing nodes that are read-sensitive to one of these N-1 wavelengths. Thus, each of the N processing nodes controls N-1 optical modulators, each optical modulator being coupled to one of the N-1 wavelengths of the optical frequency comb. An optical modulator may in particular consist of an optical ring micro-resonator used as a switch in order to modulate the optical signal coupled to its resonant frequency. Similarly, each of the N processing nodes controls an optical filter for reading any optical signal addressed to it. This optical filter may also consist of an optical ring micro-resonator for extracting from the frequency comb only the optical signal of wavelength characteristic of the node in question. Furthermore, a particular configuration of the optical bus back and forth can first of all allow the transmission of the optical signals in writing in a first direction of the bus, before allowing the reading of optical signals transmitted in the bus in a second direction. return from the bus.
    [0005] This communication scheme allows each of the N processing nodes to write and / or read data on the optical bus at the same time as the others, these data being addressed to or from other processing nodes. Nevertheless, several processing nodes may want to modulate signals of the same wavelength at the same time, which may cause the transmission of several write signals to the same destination node and at the same time may cause collisions between the data and induce errors in the information transmitted. The involved nodes will then have to modulate and transmit their data again, thereby decreasing system performance. In order to avoid collisions between the data, a more efficient solution than retransmission of the data is the arbitration between the conflicting requests for write access to the same resource (processing node or other) by several processing nodes, for the selection of one of these processing nodes considered as a priority. But this arbitration is usually done at the cost of an additional transmission delay induced by the decision of the arbitration device that implements it and an increase in the complexity of the integrated circuit. In addition, existing purely electronic arbitration devices are not suitable for new high performance chip array systems including optical devices. Indeed, these arbitration devices often use unfair algorithms, according to which all nodes do not have the same probability of access to shared resources, causing high latencies in the transmission of data and inefficient use of their network. when the number of nodes wishing to access the same shared resource is high. These arbitration devices therefore limit the bandwidth performance of systems designed with optical interconnections.
    [0006] In order to improve the efficiency of the communication systems, in particular MWSR type, using optical interconnections, fast arbitration devices based on optical technology are proposed. The invention thus applies more particularly to an optical arbitration device between conflicting requests for access to a shared resource transmitted by a plurality of N processing nodes of a network-on-a-chip system, for the selection of a processing node among the plurality of N processing nodes wishing to access the shared resource, comprising: at least one main optical arbitration bus, at least one optical source for transmitting an optical signal in said at least one main optical bus arbitration, and a succession of N arbitration optical cells coupled to said at least one main optical arbitration bus, each of said arbitration optical cells being associated with a processing node among the plurality of N processing nodes and each having means for selecting the processing node with which it is associated by diverting the optical signal emitted by the source. Dalk and Towles, entitled "Principles and Practices of Interconnection Networks", published by Morgan Kaufmann in 2004, proposes on pages 252 to 255 an electronic device of series arbitration (of the English "daisy-chain arbiter") . This arbitration device consists of a set of arbitration cells connected to each other in series, forming a chain of arbitration cells successively arranged along this series connection. Each arbitration cell aims to control access to a shared resource for a processing node assigned to it. The selection of one of these nodes is articulated according to a predefined priority scheme: each arbitration cell is assigned a fixed priority and relative to its position in the chain. Thus, the first cell in the chain has the highest priority and the last cell in the lowest priority string. Specifically, an authorization signal for access to the shared resource flows along the serial connection in the order of predefined priorities and the first arbitration cell wishing to reserve access to the shared resource for the processing node. with which it is associated changes the value of this authorization signal. It passes for example from the value "1", meaning that the resource is available, to the value "0", meaning that access to the resource is now reserved. To release access to the resource again, it is then sufficient for the arbitration unit concerned to return this authorization signal to "1". By analogy with the technologies available in optics, the teaching of this document makes it possible to design an optical arbitration device in which N arbitration cells associated with N processing nodes are successively arranged along the same main optical bus. arbitration supplied with an optical signal by a source at one of its ends, this supply end defining the predefined order of the priorities. Each arbitration cell then has means for diverting the optical signal emitted by the source so as to reserve access to the shared resource for the processing node with which it is associated. These diversion means may comprise a ring micro-resonator. It will be understood that when an arbitration cell has diverted the optical signal to its advantage, the arbitration cells arranged downstream along the main optical arbitration bus can not reserve access to the shared resource as long as the optical signal is not released again in the optical bus. As a result, the processing node associated with the arbitration cell that has diverted the optical signal can use the shared resource without risk of collision. Thus, advantageously, even if this optical arbitration device is not fair from the point of view of priority management, it allows the rapid resolution of potential conflicts between several processing nodes wishing to access the same shared resource.
    [0007] In addition, it is easily extensible, being able to adapt to a possible addition of processing nodes in the system. It is enough to add, along the main optical arbitration bus, as many arbitration cells as nodes added in the system, without any other necessary modification. On the other hand, it is also understood that when a first arbitration cell has diverted the optical signal to its advantage, the arbitration cells arranged upstream along the main optical arbitration bus can quite also divert it. Thus, if a second arbitration cell, located upstream of the first, in turn diverts the optical signal, it automatically interrupts access to the shared resource reserved by the first. This causes interruptions in the transmission of data. In order to avoid such untimely interruptions, the article by Vantrease et al, titled "Corona: system implications of emerging nanophotonic technology", published in ISCA Proceedings of the 35th Annual International Symposium on Computer Architecture, 2008, pages 153-164, proposes a dynamic optical arbitration device based on a so-called "token capture" mechanism. In this article, a plurality of N processing nodes compete for access to a plurality of shared resources. In order to avoid conflicts between the nodes for access to these resources, an optical arbitration device is set up. This arbitration device comprises a main waveguide and a ring waveguide. The main waveguide is powered by an optical source providing a comb of optical frequencies propagating in this waveguide, each wavelength of the comb being associated with one of the shared resources. The ring waveguide manages the "token capture" mechanism for allocating the different shared resources to the different processing nodes. N arbitration cells each associated with one of the plurality of N processing nodes are disposed between the ring waveguide and the main waveguide. Each arbitration cell has means for diverting the optical signal propagating in the main waveguide so as to reserve the access of the processing node with which it is associated with one of the shared resources. The diversion means may comprise several pairs of optical filters, for example micro-ring resonators. Each pair of filters comprises, on the one hand, a first optical filter capable of injecting into the ring waveguide one of the wavelengths of the frequency comb propagating in the main waveguide, generating thus a corresponding token. On the other hand, each pair has a second optical filter capable of diverting to its advantage one of the chips propagating in the ring waveguide. Therefore, the processing node associated with the arbitration cell which diverts a token can use the shared resource associated with the wavelength of the diverted token without risk of collision or interruption. Indeed, when an arbitration cell diverted to its advantage one of the chips, none of the other arbitration cells arranged along the ring waveguide no longer has access to this token: it does not In particular, there is no concept of upstream or downstream disposition in this ring waveguide. As a result, the other arbitration cells can not reserve access to the associated shared resource until this token is injected again into the ring waveguide by the first optical filter of the cell that has diverted the token. This arbitration device thus makes it possible to overcome the aforementioned problems of collisions between data and interruptions of transfers related to priorities. Nevertheless, this device is not completely fair, because not all nodes have the same probability of access to shared resources. Indeed, the way in which the tokens propagate in the ring waveguide implies that the neighboring nodes of a node that releases the token, in the direction of propagation of the chips, have priority for the reservation of this token.
    [0008] It may thus be desirable to provide an optical arbitration device that makes it possible to overcome at least some of the aforementioned problems and constraints. The invention thus relates to an optical arbitration device between conflicting requests for access to a shared resource transmitted by a plurality of N processing nodes of a network-on-chip system for the selection of a processing node. among the plurality of N processing nodes wishing to access the shared resource, comprising: at least one main optical arbitration bus, at least one optical source for transmitting a first optical signal in said at least one main optical arbitration bus , and a succession of N arbitration optical cells coupled to said at least one main optical arbitration bus, each of these arbitration optical cells being associated with a processing node among the plurality of N processing nodes and each having means for selecting the processing node with which it is associated by diverting the first optical signal, wherein: said at least one optical source is configured to transmit a second optical signal propagating in a direction opposite to the first optical signal with respect to said succession of N arbitration optical cells along said at least one main arbitration bus, and the selection means of each optical cell Arbitration are designed to select the processing node with which it is associated by diverting the first and second optical signals. Thus, thanks to the invention, the optical arbitration device allows a fast and fair conflict resolution. Indeed, the selection of the processing node being effected according to the detection of two optical signals propagating in opposite directions in the main optical arbitration bus, the access probability of each processing node to the shared resources is independent of the disposition of its arbitration cell along this bus.
    [0009] In addition, this arbitration device is easily extensible, being able to adapt at low cost to a possible increment of the number of nodes in the circuit. Indeed, the addition of a new processing node to the network-on-chip system involves only the addition to the arbitration device of a new arbitration optical cell associated with this new processing node.
    [0010] Optionally, the selection means of each arbitration cell comprise: at least one optical filter for the diversion of the first and second optical signals, at least one secondary optical arbitration bus coupled to said at least one optical filter and intended for the propagation of the first and second diverted optical signals, at least one optoelectronic converter connected to one of the ends of said at least one secondary optical arbitration bus for the conversion of the first and second optical signals diverted into at least one electrical signal of acquittal. Optionally also, an optical arbitration device according to the invention may comprise: two optical sources designed to generate respectively the first and second optical signals, a first main optical arbitration bus, connected to one of the two sources optical and coupled to the N arbitration cells, in which is intended to propagate the first optical signal in a first direction with respect to said succession of N arbitration optical cells, a second main optical arbitration bus, connected to the other of the two optical sources and coupled to the N arbitration cells, wherein is intended to propagate the second optical signal in a second direction, opposite to the first direction, with respect to said succession of N arbitration optical cells.
    [0011] Optionally also, an optical arbitration device according to the invention may comprise: a single optical source, an optical signal splitter connected at the output of the single optical source, designed to split an optical signal generated by the single optical source into two optical signals of the same wavelength propagating respectively in two output branches of the divider, a first main optical arbitration bus, connected to one of the two output branches of the divider and coupled to the N arbitration cells, in which is propagated one of the two optical signals from the divider as the first optical signal in a first direction with respect to said succession of N arbitration optical cells, and a second main optical arbitration bus, connected to the other of the two output branches of the divider and coupled to the N arbitration cells, wherein is intended to propagate the other of the two optical signals from u divider as a second optical signal in a second direction, opposite to the first direction, with respect to said succession of N arbitration optical cells. Also optionally, the selection means comprise: two optical filters for the respective diversion of the first and second optical signals, two secondary optical arbitration buses each coupled to one of the two optical filters and each being intended for the propagation of one of the first and second diverted optical signals, two optoelectronic converters each connected to one end of one of the two secondary optical arbitration buses for the conversion of the first and second diverted optical signals into first and second electrical acknowledgment signals. Optionally also, an optical arbitration device according to the invention may comprise: two optical sources designed to generate respectively the first and second optical signals, these being of different wavelengths, a single main optical bus of arbitration, connected to the two optical sources respectively at its two ends and coupled to the N arbitration cells, in which are intended to propagate the first optical signal in a first direction with respect to said succession of N arbitration optical cells and the second optical signal in a second direction, opposite the first direction, with respect to said succession of N arbitration optical cells.
    [0012] Also optionally, the selection means comprise: two optical filters for the respective diversion of the first and second optical signals; a single secondary optical arbitration bus coupled to the two optical filters and intended for the propagation of the first and second signals; diverted optics, - two optoelectronic converters respectively connected to both ends of the secondary optical arbitration bus for the conversion of the first and second optical signals diverted into first and second electrical acknowledgment signals.
    [0013] Also optionally, the selection means comprise: two optical filters for the diversion of the first and second optical signals; a single secondary optical arbitration bus coupled to the two optical filters and intended for the propagation of the first and second optical signals. diverted, - a single optoelectronic converter connected to at least one of the two ends of the secondary optical arbitration bus for the conversion of the first and second diverted optical signals into a single electrical acknowledgment signal.
    [0014] The invention also relates to an optical arbitration method between conflicting requests for access to a shared resource transmitted by a plurality of N processing nodes of a network-on-chip system, for the selection of a node of one of the plurality of N processing nodes wishing to access the shared resource, comprising the steps of: transmitting a first optical signal by at least one optical source in at least one main optical arbitration bus to which N optical cells are coupled successive arbitration arrays, each of these optical arbitration cells being associated with a processing node among the plurality of N processing nodes, selection of a processing node by diversion, by the optical arbitration cell associated with this node processing, the first optical signal, characterized in that: said at least one optical source emits a second optical signal propagating in one direction opposite to the first optical signal with respect to said succession of N arbitration optical cells along said at least one main optical arbitration bus, and the selection of the processing node is carried out by diversion, by the associated arbitration cell at this processing node, first and second optical signals. Optionally, the selection of the processing node comprises the following steps: transmission, by each processing node wishing to access the shared resource, of at least one request signal to the arbitration optical cell associated with it, attempt diverting, by each optical arbitration cell that has received said at least one request signal, first and second optical signals propagating in opposite directions along said at least one optical main arbitration transfer bus, by an optoelectronic converter the arbitration optical cell having succeeded in diverting the first optical signal from the first optical signal to a first electrical acknowledgment signal and transmitting this first electrical acknowledgment signal to the processing node associated with this optical cell; arbitration, conversion, by an optoelectronic converter of the optical arbitration cell which managed to divert the second optical signal, from the second optical signal to a second electrical acknowledgment signal, and transmission of this second electrical acknowledgment signal to the processing node associated with this optical arbitration cell, one of the processing nodes wishing to access the shared resource being selected if it receives the two electrical acknowledgment signals in response to its request. Optionally also, if one of the processing nodes wishing to access the shared resource is selected, it maintains the transmission of said at least one request signal to the arbitration optical cell associated with it during an access time. to the shared resource and stops this transmission as soon as it releases the shared resource. Also optionally, any processing node wishing to access the shared resource but receiving at most one acknowledgment signal in response to its request, stops the transmission of said at least one request signal during a programmable waiting time. . Optionally also, the waiting time is calculated by each processing node according to a round Robin arbitration scheme.
    [0015] The invention will be better understood with the aid of the description which follows, given solely by way of example and with reference to the appended drawings, in which: FIG. 1 schematically represents the general structure of a network system of on-chip processing nodes comprising an optical arbitration device according to a first embodiment of the invention, FIG. 2 illustrates in detail an optical arbitration cell of the optical arbitration device of FIG. 1, FIG. schematically the possible general structure of one of the processing nodes of the network-on-chip system of FIG. 1, FIG. 4 illustrates the successive steps of an optical arbitration method according to an embodiment of the invention, by example implemented by the optical arbitration device of FIG. 1, FIG. 5 illustrates the successive steps of a method of controlling a processing node of the network system on the basis of FIG. 1 of FIG. 1, implemented during the execution of the optical arbitration method of FIG. 4, FIGS. 6 and 7 schematically represent the general structure of a network-on-chip system comprising an optical arbitration device according to second and third embodiments of the invention, FIG. 8 illustrates in detail an optical arbitration cell of the optical arbitration device of FIG. 7, FIG. 9 schematically represents the general structure of a network system. on chip comprising an optical arbitration device according to a fourth embodiment of the invention, Figures 10 and 11 illustrate in detail two possible variants of the optical arbitration cell of Figure 8, Figure 12 illustrates the successive steps of a method of calibrating an electrical signal detector for acquiring the on-chip network system of FIG. 9, and FIG. 13 illustrates in detail the electrical diagram of a calibration element of an acknowledgment electrical signal detector of the network-on-chip system of FIG. 9.
    [0016] FIG. 1 illustrates the general structure of a network-on-a-chip system 10 comprising an optical arbitration device 12 according to a first embodiment of the invention. In general, this optical arbitration device 12 is capable of selecting a processing node from among a plurality of N processing nodes of the network-on-chip system 10, each of these processing nodes wishing to potentially access a shared resource . More precisely, FIG. 1 comprises: four processing nodes 14, 16, 18 and 20 ordered from left to right in the plane of the figure and wishing to access the shared resource, and the optical arbitration device 12. For reasons of clarity, the shared resource as well as the connections between the processing nodes and this shared resource are not shown in FIG. 1. The optical arbitration device 12 comprises first and second main optical arbitration buses 22 and 24, first and second optical sources 26 and 28, each for transmitting an optical signal propagating respectively through one of the two main optical arbitration buses 22 and 24 and a succession of four arbitration optical cells 30, 32, 34 and 36 arranged from left to right in the plane of the figure along the first and second main optical arbitration buses 22 and 24. Each of the arbitration optical cells 30, 32, 34 and 36 is coupled to the two main optical arbitration buses 22 and 24. In addition, each of the arbitration optical cells 30, 32, 34 and 36 is respectively associated with one of the four processing nodes 14, 16, 18 and 20 and is intended to control access to the shared resource of the processing node associated with it. Each of the main optical arbitration buses 22 and 24 is supplied with an optical signal at one of its ends by one of the optical sources 26 and 28 respectively. In this embodiment, the optical signals emitted by the optical sources 26 and 28 may be indifferently of the same wavelength or of different wavelengths. Each of these optical signals propagates in one of the two main optical arbitration buses 22 and 24 in a direction opposite to the other with respect to the succession of arbitration optical cells 30, 32, 34 and 36 along of the two main optical arbitration buses 22 and 24. Thus, the first optical source 26, disposed at the left end, in the plane of FIG. 1, of the first main optical arbitration bus 22, emits a first optical signal propagating from left to right through this first main optical arbitration bus 22. The second optical source 28, disposed at the right end, in the plane of FIG. 1, of the second main optical arbitration bus 24, transmits a second optical signal propagating from right to left through this second main optical arbitration bus 24. Each of the arbitration optical cells 30, 32, 34 and 36 has means for selecting the processing node 14, 16, 18 or 20 with which it is associated by diverting the first and second optical signals mentioned above. FIG. 2 illustrates in detail the selection means of any of the optical arbitration cells 30, 32, 34 and 36 of the optical arbitration device 12 of FIG. 1 previously described, these selection means comprising means for diverting the first and second optical signals propagating respectively in the first and second main optical arbitration buses 22 and 24 to which the arbitration optical cell 30, 32, 34 or 36 is coupled. In this figure, the general reference 38 identifies any arbitration optical cells 30, 32, 34 or 36 previously mentioned. The means for diverting the selection means of the optical arbitration cell 38 comprise two optical filters 40 and 42 respectively coupled to the first and second main optical bus arbitration 22 and 24 for the diversion of the first and second optical signals propagating in these two main optical arbitration buses 22 and 24. The processing node 14, 16, 18 or 20 associated with the arbitration optical cell 38, when it wishes to access the shared resource, thus transmits two request signals 44 and 46 to the optical arbitration cell 38, more specifically addressed respectively to the two optical filters 40 and 42. Upon reception of these two request signals, the two optical filters 40 and 42 are activated and attempt to divert the first and second respectively. optical signals propagating in the two main optical arbitration buses 22 and 24. Each optical filter 40 or 42 may in particular be made using a micro-ring resonator. Note that, alternatively, for certain applications, the two request signals 44 and 46 may refer to the same and the same request signal sent to the arbitration optical cell 38. The selection means of the optical cell of FIG. arbitration 38 furthermore comprise two secondary optical arbitration buses 48 and 50 respectively coupled to the two optical filters 40 and 42 and intended for the propagation of one of the first and second diverted optical signals. Two optoelectronic converters 52 and 54, for example photodiodes, are respectively connected to one end of the two secondary optical arbitration buses 48 and 50 for the conversion of the first and second diverted optical signals into first 56 and second 58 signals. Acknowledgment electricals returned optionally to the processing node 14, 16, 18 or 20 associated with the optical arbitration cell 38. The selection of a processing node among several processing nodes wishing to access the shared resource s' articulates according to a predefined priority scheme. It will now be demonstrated that, thanks to an arbitration device according to the invention, such as for example that described above, all the processing nodes 14, 16, 18 and 20 have the same average probability of access to the resource shared and therefore that an optical arbitration device according to the invention is fair from the point of view of the management of priorities.
    [0017] In the embodiment illustrated in FIG. 1, each arbitration optical cell 30, 32, 34 or 36 is assigned two priorities relative to its position with respect to the directions of propagation of the first and second optical signals in the two optical buses. 22 and 24. Thus, the arbitration optical cell 30 receiving first the first optical signal transmitted by the optical source 26 in the first main optical arbitration bus 22 has the highest priority, for example "4", this priority gradually decreasing from "3" to "1" for the arbitration optical cells 32, 34 and 36 arranged successively downstream in the direction of propagation of the first optical signal. This same arbitration optical cell 30 receives last the second optical signal emitted by the optical source 28 in the second main optical arbitration bus 24 and in this case has the lowest priority, for example "1", this priority gradually increasing from "2" to "4" for the arbitration optical cells 32, 34 and 36 disposed successively upstream in the direction of propagation of the second optical signal.
    [0018] Thus, by averaging these priorities two by two since the criterion for selecting a processing node is based on the diversion of the two optical signals propagating in opposite directions in the two main optical arbitration buses 22 and 24, as a result, the predefined average priority of each optical arbitration cell is the same as for the others and, in this case, equal to 5/2.
    [0019] By generalizing to N arbitration optical cells, the predefined mean priority is the same for all arbitration optical cells and equal to (N + 1) / 2. By way of example, FIG. 3 diagrammatically represents a possible implementation of one of the processing nodes of FIG. 1. Thus, the general reference 60 identifies any one of the processing nodes 14, 16, 18 and 20 previously mentioned. This processing node 60 may comprise several data processing modules 62, 64, 66 and 68, in particular processor cores, dedicated chips, memories, and so on. The data processing modules 62, 64, 66 and 68 are connected by a computer bus 70 enabling data exchanges between them. The processing node may further comprise a control module 72 also connected to the computer bus 70 for data exchanges between each data processing module 62, 64, 66 and 68 and the control module 72. This control module 72 allows the communication of each of the processing modules 62, 64, 66 and 68 with the optical arbitration device 12.
    [0020] In this exemplary implementation, the control module 72 is part of the processing node 60, but in other embodiments it can be placed outside the processing node 60. Also, in this implementation example four data processing modules 62, 64, 66 and 68 are integrated in the processing node 60, but in other variants this number of processing modules may be higher or lower. The communication between the processing node 60 and the optical arbitration device 12 is carried out by means of four electrical connections placed between the control module 72 and the optical arbitration cell 38 associated with the processing node 60. when the processing node 60 wishes to access the shared resource, it transmits, by means of the control module 72 and through two of the four aforementioned electrical connections, the two request signals 44 and 46 to the optical cell of arbitration 38 associated with it. As soon as the optical arbitration cell 38 succeeds in obtaining access to the shared resource by diverting the first and second optical signals passing through the two main arbitration optical buses 22 and 24, it transmits to the processing node 60 the two electrical acknowledgment signals 56 and 58. FIG. 4 illustrates the successive steps of an optical arbitration method that can be implemented with the aid of an optical arbitration device according to the invention, in particular with respect to The optical arbitration device 12 illustrated in FIG. 1. This optical arbitration method makes it possible to select one of several processing nodes wishing to access the shared resource in a potentially conflicting manner. In a first step 200 of putting into operation the optical arbitration device 12, the first optical source 26 generates and continuously transmits the first optical signal in the first main optical arbitration bus 22. This first optical signal is propagated in a first direction, from left to right in FIG. 1, so as to first pass at the level of the optical arbitration cell 30, then at the level of the arbitration optical cell 32, then at the level of the optical arbitration cell 34, then at the optical arbitration cell 36. During this same step 200, the second optical source 28 continuously generates and transmits the second optical signal in the second main optical arbitration bus 24. This second optical signal is propagated in a second direction opposite to the first, from the right to the left in FIG. 1, so as to first pass at the level of the arbitration optical cell 36, then at the level of FIG. the cell arbitration optic 34, then at the optical arbitration cell 32, then at the optical arbitration cell 30. In a next step 202, each processing node 14, 16, 18 or 20 wishing accessing the shared resource simultaneously transmits the two request signals 44 and 46 to the arbitration optical cell 30, 32, 34 or 36 associated therewith. During a step 204, each arbitration optical cell 30, 32, 34 or 36 having received the request signals 44 and 46 attempts to divert the first and second optical signals propagating in opposite directions respectively along the two buses. main optical arbitration 22 and 24. In a subsequent step 206, the optoelectronic converter of the arbitration optical cell has succeeded in diverting the first optical signal, converts this first optical signal into a first electrical acknowledgment signal 56 , and transmits this first electrical acknowledgment signal 56 to the processing node associated with this optical arbitration cell. During a step 208, performed before, during or after step 206, but after step 204, the optoelectronic converter of the arbitration optical cell having succeeded in diverting the second optical signal, converts this second optical signal into a second electrical acknowledgment signal 58, and transmits this second electrical acknowledgment signal 58 to the processing node associated with this optical arbitration cell. In a last step 210, one of the processing nodes wishing to access the shared resource is selected if it receives the two acknowledgment electrical signals 56 and 58 in response to its request.
    [0021] Thus, an optical arbitration cell must divert to its benefit the two optical signals flowing in opposite directions in the two main optical arbitration buses to give access to the shared resource to the processing node with which it is associated. Nevertheless, when an arbitration optical cell diverts only one of the two optical signals to its advantage, it does not obtain the right of access to the shared resource but prevents the rest of the arbitration optical cells from gaining access to the shared resource. to this optical signal and therefore to access the shared resource as the optical signal it has diverted is not released again in the corresponding main optical arbitration bus.
    [0022] FIG. 5 thus illustrates the successive steps of a control method implemented during the execution of the optical arbitration method detailed above, in order to avoid any blocking situation. This method is executed by the control module 72 of each processing node, represented in FIG. 3 by the general reference 60, when the latter wishes to access the shared resource. In a first step 212, the control module 72 transmits the two request signals 44 and 46 to the arbitration optical cell 38 associated with the processing node 60. In a following step 214, the control module 72 verifies the expected receipt of the acknowledgment electrical signals 56 and 58 from the arbitration optical cell 38 in response to its request. If the control module 72 receives at most one electrical acknowledgment signal in response to its request, it concludes that its request is not acknowledged. In this case, during a subsequent step 216, the control module 72 stops the transmission of the two request signals 44 and 46 to the associated arbitration optical cell 38 during a waiting time T. This time should be waiting time T is different for each control module, as this makes it possible to avoid the renewal of conflicts between optical arbitration cells for the diversion of the first and second optical signals. Indeed, the diversion attempts by each of the arbitration optical cells will take place at different times if the arbitration optical cells receive the request signals at different times. Various methods of calculating this waiting time T exist and will be presented later. After this waiting time T, the process returns to step 212 in which the control module 72 again transmits the two request signals 44 and 46. The successive steps 212, 214 and 216 previously detailed are repeated until the processing node 60 is selected for access to the shared resource. If, in step 214, the control module 72 receives the two acknowledgment electrical signals 56 and 58, it concludes that the corresponding processing node 60 has been selected. In this case, the step 214 is followed by a step 218, during which the control module 72 maintains the transmission of the two request signals 44 and 46 to the associated arbitration optical cell 38 during the entire period of time. access of the selected processing node 60 to the shared resource. This forces the arbitration cell 38 to keep the first and second optical signals diverted and thus reserves the exclusive use of the shared resource at the selected processing node 60. This step 218 for maintaining the transmission of the request signal is advantageous because it makes it possible to avoid any interruption in data transmission that may be caused by other optical arbitration cells wishing to reserve access to the shared resource. In a last step 220, when the selected processing node 60 releases the shared resource, the control module 72 stops the transmission of the two request signals 44 and 46 to the optical arbitration cell 38 thus allowing the access to the shared resource to the other processing nodes. As mentioned above, there are several methods for determining the waiting time T. This can be defined statically or dynamically. A static waiting time is a waiting time assigned to each control module 72 during its design and stored in a specific register of the control module. In general, the priority of a processing node decreases statistically when the value of the static waiting time of its control module increases. Thus, the determination of a different static waiting time for each processing node generates a statistical inequality in the effective access priorities, although the management of the predefined priorities according to the invention is a priori equitable. As a variant, each processing node 60 can calculate dynamically, that is to say, while it is executing its processes, the waiting time T of its control module 72. The value of the waiting time T dynamically calculated is updated in the specific register of the control module 72 using a JTAG bus. The advantage of this variant is that the dynamic calculation of the waiting times T can be designed in a statistically fair manner for all the processing nodes. The control module 72 itself can calculate the waiting time T using a calculation function. This calculation function is determined during the design of the network-on-a-chip system, as it must be designed to minimize the latencies of the network-on-chip system and to ensure equity between the processing nodes when sharing resources. . This function can be simply a random function. It can also respond to an arbitration scheme well known to those skilled in the art under the name of Round Robin which makes it possible to assign different waiting times to the different processing nodes cyclically and without giving them particular priorities. Other more complex functions taking into account many parameters such as the number of consecutive failures for access to the shared resource, the temperature of the processing node, the size of the data to be transmitted, etc., can also be used for the calculating the waiting time T.
    [0023] FIG. 6 illustrates the general structure of a network-on-chip system comprising an optical arbitration device 74 according to a second embodiment of the invention. This optical arbitration device 74 differs from the previous one in that it comprises a single optical source 76 connected at its output to an optical signal splitter 78, designed to split the optical signal generated by the single optical source 76 into two signals. optical same wavelength propagating respectively in two output branches of the divider. A first main arbitration optical bus 80 is connected to one of the two output branches of the divider 78 and is coupled to the four arbitration cells 30, 32, 34 and 36. One of the two optical signals from the divider is intended to propagate as a first optical signal in this first main optical arbitration bus 80 in a first direction with respect to said succession of the four arbitration optical cells 30, 32, 34 and 36. A second main optical bus arbitration 82, connected to the other of the two output branches of the divider 78 is also coupled to the four arbitration optical cells 30, 32, 34 and 36. The other of the two optical signals from the divider is intended to propagate in this second main optical arbitration bus 82 as a second optical signal in a second direction, opposite to the first direction, with respect to said succession of four arbitration optical cells 30, 32, 34 and 36.
    [0024] The arbitration cells of this second embodiment of the invention are identical to those of the first embodiment described above with reference to FIGS. 1 and 2. The processing nodes 14, 16, 18 and 20 are likewise unchanged. The operating principle of this second embodiment is also identical to that described above with reference to FIGS. 3 to 5. This second embodiment has the advantage of being less expensive than the previous one since only one optical source 76 is necessary. . The general structure of a network on chip system comprising an optical arbitration device 84 according to a third embodiment of the invention is shown schematically in FIG. 7. The processing nodes 14, 16, 18 and 20 are unchanged. . On the other hand, the optical arbitration device 84 differs from the two previous ones in that it comprises two optical sources 86 and 88 respectively generating the first and second optical signals and transmitting them in a single main optical arbitration bus 90 in the sense inverses. In this case, the two optical signals are necessarily of different wavelengths. The single main arbitration optical bus 90 is connected to the two optical sources 86 and 88 respectively at its two ends and is coupled to four arbitration cells 92, 94, 96 and 98 different from the arbitration optical cells 30, 32 , 34 and 36 illustrated in Figures 1, 2 and 6. The first optical signal propagates in this main optical arbitration bus 90 in a first direction with respect to said succession of four arbitration optical cells 92, 94, 96 and 98 and the second optical signal in a second direction, opposite to the first.
    [0025] FIG. 8 illustrates in detail a possible implementation of any of the arbitration optical cells 92, 94, 96 and 98 of the optical arbitration device 84 of FIG. 7. In this figure, the general reference 100 identifies the any arbitration cells 92, 94, 96 and 98. The means for diverting the selection means of the arbitration optical cell 100 illustrated in FIG. 8 comprise two optical filters 102 and 104, coupled to the single bus. main optical arbitration 90 and able to receive the two request signals 44 and 46 defined above. As soon as the optical filters 102 and 104 receive these two request signals 44 and 46, they are activated and attempt to respectively divert the first and second optical signals, of different wavelengths, propagating in opposite directions in the single main optical arbitration bus 90. For this purpose they are respectively adjusted to the different wavelengths of the two optical signals. In other words, they are not tuned together, not resonating for the same frequencies. The selection means of the arbitration optical cell 100 further comprise a single secondary optical arbitration bus 106 coupled to the two optical filters 102 and 104 and intended for the propagation of the first and second optical signals diverted. Two optoelectronic converters 108 and 110, for example photodiodes, respectively connected to the two ends of the secondary optical arbitration bus 106 allow the conversion of the first and second optical signals diverted into the first and second electrical acknowledgment signals 56 and 58 defined above. . Advantageously, this third embodiment makes it possible to spare a main optical arbitration bus, as well as a secondary optical arbitration optical arbitration optical bus, thus presenting a more compact and less costly optical arbitration device. those of the previous embodiments. The operating principle of this third embodiment is also identical to that described above with reference to FIGS. 3 to 5. FIG. 9 illustrates the general structure of a network-on-chip system comprising an optical arbitration device 112 according to a fourth embodiment of the invention. This optical arbitration device 112 comprises, like the arbitration device 12 of FIG. 1, the two main optical arbitration buses 22, 24 and the two optical sources 26, 28 for propagation of the first and second optical signals identical. to that of the first embodiment. However, it differs in the use of four arbitration optical cells 114, 116, 118 and 120 different from the arbitration optical cells 30, 32, 34 and 36 illustrated in FIGS. 1, 2 and 6. The four arbitration optical cells 114, 116, 118 and 120 are also different from the arbitration optical cells 92, 94, 96 and 98 illustrated in FIGS. 7 and 8. The optical arbitration cell 114 firstly comprises, in FIG. as means for diverting the first and second optical signals, two optical filters 122 and 124 respectively coupled to the two main optical bus arbitration 22, 24 and able to receive the two request signals 44 and 46 defined above.
    [0026] As soon as the optical filters 122 and 124 receive these two request signals 44 and 46, they are activated and attempt to respectively divert the first and second optical signals propagating in opposite directions in the two main optical arbitration buses 22 and 24. The optical arbitration cell 114 further comprises a single secondary optical arbitration bus 126 coupled to the two optical filters 122 and 124 and intended for propagation of the first and second diverted optical signals. This single secondary optical arbitration bus 126 is configured in a round-trip direction, that is to say in a U-shaped manner, so as to impose a single direction of propagation of the two optical signals diverted along this secondary optical bus. arbitration 126 to only one of its ends provided with a single optoelectronic converter 128. The optical filter 124 coupled to the second main arbitration bus 24 is thus for example connected to a branch of the secondary optical arbitration bus 126 , while the optical filter 122 coupled to the first arbitration main bus 22 is connected to a return branch of the secondary optical arbitration bus 126.
    [0027] The single optoelectronic converter 128, for example a photodiode, enables the conversion of the first and second diverted optical signals into a single electrical acknowledgment signal supplied at the input of a detector 130 of this electrical acknowledgment signal, this detector 130 connecting the arbitration optical cell 114 at the processing node 14.
    [0028] It will be noted that, in order to avoid any inadvertent leakage, to the first optical main arbitration bus 22, of the second optical signal diverted into the secondary optical arbitration bus 126, it is necessary in this fourth embodiment also that the two optical signals they are of different wavelengths and the corresponding optical filters 122 and 124 are respectively set at these different wavelengths. Indeed, if this were not the case, the second optical signal diverted into any arbitration optical cells 114, 116, 118 and 120 could be partially diverted a second time by the optical filter 122 to the first main optical arbitration bus 22. In addition, if the wavelengths were identical, the two optical signals diverted into the secondary optical arbitration bus 126 could interfere with each other, such constructive or destructive interference then being able to detract from their detection. The optical arbitration cells 116, 118 and 120 are identical to the arbitration optical cell 114 and will therefore not be detailed. They are respectively connected to the processing nodes 16, 18 and 20 by detectors 132, 134 and 136 identical to the detector 130. The processing nodes 14, 16, 18 and 20 are unchanged with respect to the previous embodiments if it is not possible is that they are designed to receive a single electrical acknowledgment signal representative of the two optical signals diverted by their optical arbitration cell.
    [0029] Advantageously, this fourth embodiment makes it possible to spare a secondary optical arbitration bus and an optoelectronic converter by an arbitration optical cell, thus presenting an optical arbitration device that is more compact and less expensive than that of the first embodiment.
    [0030] On the other hand, this fourth embodiment is not compatible with the second embodiment illustrated in FIG. 6 since the two optical signals must be of different wavelengths. It is also not possible to modify the third embodiment illustrated in FIG. 7 by replacing the arbitration optical cells 92, 94, 96 and 98 by the arbitration optical cells 114, 116, 118 and 120 because that the third embodiment has only one main optical arbitration bus 90. But to exploit the same idea as that of the fourth embodiment, it is possible to modify the third embodiment illustrated in FIG. replacing the arbitration optical cells 92, 94, 96 and 98 not by the arbitration optical cells 114, 116, 118 and 120 but by instances of a new hook-shaped optical arbitration cell 138 illustrated on FIG. FIG. 10. This optical arbitration cell 138 comprises, like the arbitration optical cell 100, the two optical filters 102 and 104 coupled to the single main optical arbitration bus 90 and not tuned together, as well as the one of two Optoelectronic lifters 108 and 110, identified by reference 140 in this figure. The other of the two optoelectronic converters as well as the secondary optical arbitration bus 106 are replaced by a single arbitrary secondary optical arbitration bus 142 in the form of a hook. The main branch of this secondary optical arbitration bus 142 is coupled to the optical filter 102 while the hook return branch of this secondary optical arbitration bus 142 is coupled to the optical filter 104. Thus, the hook shape of this The secondary optical arbitration bus 142 performs the same function as the U-shape of the secondary optical arbitration bus 126 of FIG. 9, namely to impose a single direction of propagation of the two optical signals diverted along this secondary optical bus. arbitration 142 by the two optical filters 102, 104 to only one of its ends, that of the main branch, provided with the single optoelectronic converter 140. The output of the optoelectronic converter 140 then provides a single electrical acknowledgment signal S treated by a detector 144 identical to the detectors 130, 132, 134 and 136 for the provision of a CT acknowledgment bit signal. The acknowledgment bit CT takes a first value, for example "1", if the arbitration optical cell 138 has succeeded in diverting the two optical signals, and a value "0" otherwise.
    [0031] Another variant of the arbitration optical cells 114, 116, 118 and 120, compatible with the third embodiment and equivalent to the solution illustrated in FIG. 10, is illustrated in FIG. 11. According to this alternative variant, one of any arbitrary optical cells coupled to the processing nodes 14, 16, 18 and 20 comprise two optical filters 148 and 150 not tuned together, coupled to a single main optical arbitration bus, for example the bus 90 of the FIG. 7, and able to receive the two request signals 44 and 46 defined previously. As soon as the optical filters 148 and 150 receive these two request signals 44 and 46, they are activated and attempt to respectively divert the first and second optical signals propagating in opposite directions in the main optical arbitration bus 90. arbitration optics 146 further comprises a single secondary optical arbitration bus 152 coupled to the two optical filters 148 and 150 and for propagating the first and second diverted optical signals. This single secondary optical arbitration bus 152 is configured onion-shaped, i.e., so that its two ends meet in one. Thus, even if the two diverted optical signals propagate in opposite directions in the secondary optical arbitration bus 152, they finally meet, at this single end, at the input of a single optoelectronic converter 154. The output of this optoelectronic converter 154, for example a photodiode, then provides the single aforementioned electrical acknowledgment signal S processed by the detector 144 defined above. It will be noted that the reception of a single electrical request signal S, instead of the two signals 56 and 58 received in the first three embodiments, does not change the principle of the control method illustrated in FIG. to replace, in step 214, the verification of the expected reception of the signals 56 and 58 by checking the value, "0" or "1", of the acknowledgment bit CT. CT = 0 leads to step 216 while CT = 1 leads to step 218. The operation of detector 144, like that of detectors 130, 132, 134 and 136, will now be detailed. Indeed, in order to be able to operate, this detector 144 must be able to compare at any instant the value of the electric acknowledgment signal S with a predefined threshold value Sref making it possible to decide whether the two optical signals have been diverted or not. It must also be calibrated for the determination of this threshold value Sref.
    [0032] With regard to the comparison of the electric acknowledgment signal S with the predefined threshold value Sref, it suffices to carry it out by means of a comparator, for example an operational amplifier operating in a saturated state. As far as calibration is concerned, it should be done in such a way that the threshold value found is relevant, ie neither too high for the detector to be too selective, nor too low for the detector to not too much passing. In particular, a relevant theoretical threshold value can be given by the following equation: Sref = max (Ei, E2) + max (E01, E02), where E1 is the signal received at the output of the optoelectronic converter 140 or 154 (or 128 with respect to one of the detectors 130, 132, 134 and 136) when only the first optical filter diverting the first optical signal is activated (i.e., set to the carrier frequency of the first optical signal), E2 is the signal received at the output of the optoelectronic converter 140 or 154 (or 128 with respect to one of the detectors 130, 132, 134 and 136) when only the second optical filter diverting the second optical signal is activated, E01 is the signal residual electric current corresponding to the output of the first optical filter when it is deactivated (that is to say not set on the carrier frequency of the first optical signal) and E02 is the residual electrical signal corresponding to the output of the second optical filter rsince it is disabled. This relevant theoretical threshold value assumes that the diverted optical signals do not interfere, that is to say that their powers add up to form the signal S, which is the case of the embodiments illustrated in FIGS. 11 since the two optical signals do not have the same wavelength. In practice, the values of E01 and E02 are not easy to know. Consequently, assuming that min (Ei, E2) "max (E01, E02), it is possible to propose a calculation of Sref which depends only on E1 and E2, for example according to the following equation: min (Ei, E2) Sref = max (Ei, E2) + '2 Thus, a method of calibrating the detector 144 associated with the arbitration optical cell 138 or 146 (or one of the detectors 130, 132, 134, 136 associated with the arbitration optical cells 114, 116, 118, 120) can be simply implemented, as illustrated in FIG. 12. It can be executed when the device is switched on or initialised (repeated or not). corresponding optical arbitration, while it is not yet operational but while the first and second optical signals flow well in the (case of arbitration optical cell 138 or 146) or the (cases of the optical cell arbitration 114, 116, 118 or 120) main optical buses / arbitration. It may in particular be implemented during the execution of step 200 described above. In a first step 300 of this calibration method, the first optical filter is activated while the second filter is deactivated, for the diversion of the first optical signal and the value El is measured by the electrical acknowledgment signal S. When in a second step 302 of this calibration method, the second optical filter is activated while the first filter is deactivated, for the diversion of the second optical signal and the value E2 is measured by the electrical acknowledgment signal S. In a last step 304 of this calibration method, the value of Sref is calculated using the two values E1 and E2 measured previously. The previously proposed calculation for the determination of Sref as a function of E1 and E2 involves a complex electronic implementation because of the min and max functions. It can then be simplified by supposing that El and E2 have very close values. In this case, it is indeed possible to propose a simplified Sref calculation, for example according to the following equation: Sref = G. (Ei + E2), where G is a constant value gain of between 0.5 and 1 As G is close to 1, detector 144 is selective. If E1 and E2 are voltages, the electronic implementation of such a calculation can be simply performed using an operational amplifier 160 used in a non-inverter adder assembly such as that illustrated in FIG. mounting, two inputs mounted in parallel on the non-inverting input of the operational amplifier 160 are fed respectively and sequentially by the voltage values E1 and E2. These two inputs respectively have resistors R1 and R2. The inverting input of the operational amplifier 160, resistor R3, is connected to ground. Finally, the output of the operational amplifier 160 provides the electrical request signal Sref and is connected to ground by a resistance feedback loop R4. As a result, the following relation is known per se: R4 (R2.Ei + R1.E2 Sref = (1+ -R3) R1 + R2) - By judiciously choosing R1 = R2 and R3 = 2.R4, this gives G = 0.75, the relevant value of G when E1 and E2 are indeed close.
    [0033] It clearly appears that an optical arbitration device such as one of those described above makes it possible to perform fast arbitration, low noise, low losses and insensitive to electromagnetic waves. In addition: it easily integrates a token blocking function, it is designed independently of the data packet sizes to be transmitted, it has a round trip time (weak "Round Trip Time"), it is robust against the loss of token, it allows a fair access of all the processing nodes to a shared resource, including in case of collisions in access requests, it uses at least the waveguide resources, it allows great flexibility of implementation, and it is easily expandable. Note also that the invention is not limited to the embodiments described above. It will be apparent to those skilled in the art that various modifications can be made to the embodiments described above, in the light of the teaching that has just been disclosed. In the following claims, the terms used are not to be construed as limiting the claims to the embodiments set forth in this specification, but should be interpreted to include all the equivalents that the claims are intended to cover because of their formulation and whose prediction is within the reach of the person skilled in the art by applying his general knowledge to the implementation of the teaching which has just been disclosed to him.
    权利要求:
    Claims (13)
    [0001]
    REVENDICATIONS1. An optical arbitration device (12; 74; 84; 112) between contention requests for access to a shared resource transmitted by a plurality of N process nodes (14,16,18,20, 60) of a system ( 10) for selecting one of the plurality of N processing nodes desiring access to the shared resource, comprising: at least one main optical arbitration bus (22, 24; 80, 82); 90), at least one optical source (26, 28; 76; 86,88) for transmitting a first optical signal into said at least one main optical arbitration bus and a succession of N arbitration optical cells ( 30, 32, 34, 36, 38, 92, 94, 96, 98, 100, 114, 116, 118, 120, 138, 146) coupled to said at least one main optical arbitration bus, each of said optical cells being arbitration being associated with a processing node among the plurality of N processing nodes and each having means (40, 42, 48, 50, 52, 54; 102, 104, 106, 108, 110; 122, 124, 126, 128; 102, 104, 140, 142; 148, 150, 152, 154) for selecting the processing node with which it is associated by diverting the first optical signal, characterized in that: said at least one optical source (26, 28; 76; 86, 88) is designed to transmitting a second optical signal propagating in a direction opposite to the first optical signal with respect to said succession of N arbitration optical cells (30, 32, 34, 36, 38, 92, 94, 96, 98, 100; 116, 118, 120; 138; 146) along said at least one main optical arbitration bus, and the selection means (40, 42, 48, 50, 52, 54; 102, 104, 106, 108, 110; 122, 124, 126, 128; 102, 104, 140, 142, 148, 150, 152, 154) of each arbitration optical cell are adapted to select the processing node with which it is associated by diverting the first and second optical signals.
    [0002]
    An optical arbitration device (12; 74; 84; 112) according to claim 1, wherein the selection means (40,42,48,50,52,54; 102,104,106,108,110; 122 124, 126, 128, 102, 104, 140, 142; 148, 150, 152, 154) of each arbitration cell comprise: at least one optical filter (40, 42; 102, 104; 122, 124; 150) for diverting the first and second optical signals, at least one secondary optical arbitration bus (48, 50; 106; 126; 142; 152) coupled to said at least one optical filter and for propagating the first and second diverted optical signals, at least one optoelectronic converter (52, 54; 108, 110; 128; 140; 154) connected to one end of said at least one secondary optical arbitration bus for conversion of the first and second optical signals diverted into at least one electrical acknowledgment signal.
    [0003]
    An optical arbitration device (12; 112) according to claim 1 or 2, comprising: two optical sources (26,28) adapted to generate the first and second optical signals, respectively, a first main optical arbitration bus (22); ), connected to one (26) of the two optical sources and coupled to the N arbitration cells (30, 32, 34, 36, 38; 114, 116, 118, 120), in which is intended to propagate the first optical signal in a first direction with respect to said succession of N arbitration optical cells, a second main optical arbitration bus (24), connected to the other (28) of the two optical sources and coupled to the N cells; arbitration (30, 32, 34, 36, 38; 114, 116, 118, 120), wherein is propagated the second optical signal in a second direction, opposite the first direction, with respect to said succession of N arbitration optical cells.
    [0004]
    An optical arbitration device (74) according to claim 1 or 2, comprising: a single optical source (76), an optical signal splitter (78) connected at the output of the single optical source, adapted to split an optical signal generated by the single optical source into two optical signals of the same wavelength propagating respectively in two output branches of the divider, a first main optical arbitration bus (80), connected to one of the two output branches of the divider (78) and coupled to the N arbitration cells (30, 32, 34, 36, 38), in which is propagated one of the two optical signals from the divider as the first optical signal in a first sense with respect to said succession of N arbitration optical cells, and a second main optical arbitration bus (82), connected to the other of the two output branches of the divider (78) and coupled to the N arbitration cells (30, 32, 34, 36, 38), in which is intended to propagating the other of the two optical signals from the divider as a second optical signal in a second direction, opposite to the first direction, with respect to said succession of N arbitration optical cells.
    [0005]
    An optical arbitration device (12; 74) according to claim 3 or 4, wherein the selection means (40, 42, 48, 50, 52, 54) comprises: two optical filters (40, 42) for the respectively diverting the first and second optical signals, two secondary optical arbitration buses (48, 50) each coupled to one of the two optical filters and each being intended for the propagation of one of the first and second optical signals diverted , two optoelectronic converters (52, 54) each connected to one end of one of the two secondary optical arbitration buses for the conversion of the first and second optical signals diverted into first and second electrical acknowledgment signals ( 56, 58).
    [0006]
    An optical arbitration device (84) according to claim 1 or 2, comprising: two optical sources (86, 88) adapted to respectively generate the first and second optical signals, which are of different wavelengths; single main optical arbitration bus (90), connected to the two optical sources respectively at its two ends and coupled to the N arbitration cells (92, 94, 96, 98, 100), in which are intended to propagate the first an optical signal in a first direction with respect to said succession of N optical cells of arbitration and the second optical signal in a second direction, opposite to the first direction, with respect to said succession of N arbitration optical cells.
    [0007]
    7. An optical arbitration device (84) according to claim 6, wherein the selection means (102, 104, 106, 108, 110) comprise: two optical filters (102, 104) for the respective diversion of the first and second optical signals, a single secondary optical arbitration bus (106) coupled to the two optical filters and for propagating the first and second diverted optical signals, two optoelectronic converters (108, 110) respectively connected to both ends of the secondary optical bus; arbitration for converting the first and second optical signals diverted into first and second electrical acknowledgment signals (56,58).
    [0008]
    An optical arbitration device (112) according to claim 3 or 6, wherein the selection means (122, 124, 126, 128; 102, 104, 140, 142; 148, 150, 152, 154) comprises: two optical filters (122, 124, 102, 104, 148, 150) for the diversion of the first and second optical signals, a single secondary optical arbitration bus (126; 142; 152) coupled to the two optical filters and for propagating the first and second diverted optical signals, a single optoelectronic converter (128; 140; 154) connected to at least one of the two ends of the secondary optical arbitration bus for converting the first and second optical signals diverted into a single electrical acknowledgment signal (S).
    [0009]
    An optical arbitration method between contention requests for access to a shared resource transmitted by a plurality of N processing nodes (14, 16, 18, 20, 60) of a network-on-a-chip system (10), for selecting a processing node from among the plurality of N process nodes wishing to access the shared resource, comprising the steps of: transmitting (200) a first optical signal by at least one optical source (26, 28; 76, 86, 88) in at least one main optical arbitration bus (22, 24; 80, 82; 90) to which are coupled N successive optical arbitration cells (30, 32, 34, 36, 38; 92, 94 , 96, 98, 100; 114, 116, 118, 120; 138; 146), each of these arbitration optical cells being associated with one of the plurality of N processing nodes, selection (202, 204, 206, 208, 210) of a diversion processing node, by the arbitration optical cell associated with this processing node, the first optical signal, characterized in that said at least one optical source emits (200) a second optical signal propagating in a direction opposite to the first optical signal with respect to said succession of N arbitration optical cells along said at least one optical signal. a main optical arbitration bus, and the selection (202, 204, 206, 208, 210) of the processing node is performed by diverting, by the arbitration cell associated with this processing node, the first and second signals optics.
    [0010]
    An optical arbitration method according to claim 9, wherein the selection (202, 204, 206, 208, 210) of the processing node comprises the steps of: transmission (202), by each processing node wishing to access the shared resource, at least one request signal (44, 46) to the arbitration optical cell associated therewith, attempted diversion (204), by each arbitration optical cell having received said at least one a request, first and second optical signals propagating in opposite directions along said at least one main optical arbitration bus, conversion (206), by an optoelectronic converter of the optical arbitration cell having succeeded in diverting the first optical signal , from the first optical signal to a first electrical acknowledgment signal (56; S), and transmission of this first electrical acknowledgment signal to the processing node associated with this optical arbitration cell, conversion (20; 8), by an optoelectronic converter of the arbitration optical cell having succeeded in diverting the second optical signal from the second optical signal to a second electrical acknowledgment signal (58; S), and transmission of this second electrical acknowledgment signal to the processing node associated with this arbitration optical cell, one of the processing nodes wishing to access the shared resource being selected (210) if it receives both electrical acknowledgment signals (56, 58; S) in response to its request.
    [0011]
    An optical arbitration method according to claim 10, wherein, if one of the processing nodes (14, 16, 18, 20, 60) wishing to access the shared resource is selected, it maintains (218) the transmission said at least one request signal (44, 46) to the arbitration optical cell associated with it during a time of access to the shared resource and ceases (220) this transmission as it releases the shared resource.
    [0012]
    12. An optical arbitration method according to claim 10 or 11, wherein, any processing node (14, 16, 18, 20, 60) wishing to access the shared resource but receiving at most one acknowledgment signal in response to its request, stops (216) transmitting said at least one request signal (44, 46) during a programmable waiting time.
    [0013]
    13. An optical arbitration method according to claim 12, wherein the waiting time is calculated by each processing node (14, 16, 18, 20, 60) according to a Round Robin arbitration scheme.
    类似技术:
    公开号 | 公开日 | 专利标题
    EP2882122B1|2018-01-31|Device and method for optical arbitration in a network
    CA2655948C|2016-03-15|Method of routing virtual links in a frame-switching network with guaranteed determinism
    FR2490434A1|1982-03-19|DEVICE FOR RESOLVING ACCESS CONFLICTS AND ALLOCATION OF A BUS-TYPE LINK INTERCONNECTING A SET OF NON-HIERARCHISTIC PROCESSORS
    WO2003084279A1|2003-10-09|Dynamic method of inserting data at the nodes of an optical transmission network
    FR2598575A1|1987-11-13|LOCAL AREA NETWORK CONTROL DEVICE
    EP1592159B1|2007-04-25|Optical transmission network with tree structure
    EP2070382A1|2009-06-17|Method for managing connections in an optical access network, and related platform, exchange, network and computer software product
    FR2978315A1|2013-01-25|INFORMATION TRANSMISSION NETWORK AND CORRESPONDING NETWORK NODE
    EP2874334B1|2016-06-08|Optical network and data processing system comprising such an optical network
    EP3545631B1|2020-10-07|Determination of regenerative relay nodes included in a transmission line of an optical network
    EP1871058B1|2009-12-30|System and method for the management of messages transported in an interconnection network
    EP2783485B1|2015-12-30|Data transmission network and programmable network node
    FR2963999A1|2012-02-24|METHOD FOR SWITCHING AN OPTICAL DATA STREAM, COMPUTER PROGRAM PRODUCT, STORAGE MEDIUM AND CORRESPONDING NODE.
    EP1841151A1|2007-10-03|Message switching system
    EP2200204B1|2011-11-16|Signilisation for establishing a connection in an optical network
    EP0379436A1|1990-07-25|Apparatus and process for request arbitration and resolution of conflicts linked with the access to memories with independent banks for computing machines
    EP1326473A1|2003-07-09|Optical switching device and control method therefor
    FR2832561A1|2003-05-23|DISTRIBUTED ARCHITECTURE CONTROL SYSTEM FOR STATIC POWER CONVERTERS
    FR2805105A1|2001-08-17|METHOD FOR ESTABLISHING A TWO-WAY OPTICAL COMMUNICATION BETWEEN A CENTER EQUIPMENT AND A REMOTE EQUIPMENT
    EP2073566B1|2016-11-16|Method for establishing a recovery connection in an optical network
    FR2978850A1|2013-02-08|DEVICE FOR EXCHANGING DATA BETWEEN AT LEAST TWO APPLICATIONS
    EP0214961A1|1987-03-25|Procedure and device for allocating the right of access
    Luo et al.2016|Wavelength spacing optimization to reduce crosstalk in WDM 3D ONoC
    WO2020128231A1|2020-06-25|Method for allocating a point-to-point channel to a user module of an optical communication network, computer program product, storage medium and corresponding device
    FR3007537A1|2014-12-26|RING OPTICAL NETWORK, INTEGRITY ON CHIP | AND ASSOCIATED METHOD
    同族专利:
    公开号 | 公开日
    EP2882122B1|2018-01-31|
    US9369784B2|2016-06-14|
    US20150163570A1|2015-06-11|
    FR3014563B1|2016-02-05|
    EP2882122A1|2015-06-10|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    US6078337A|1994-09-12|2000-06-20|Canon Kabushiki Kaisha|Maintaining consistency of cache memory data by arbitrating use of a connection route by plural nodes|
    DE19745020B4|1996-10-29|2008-01-31|Keymile Ag|Method for controlling data traffic in an ATM network|
    US7720377B2|2006-01-23|2010-05-18|Hewlett-Packard Development Company, L.P.|Compute clusters employing photonic interconnections for transmitting optical signals between compute cluster nodes|
    JP5070342B2|2007-10-23|2012-11-14|ヒューレット−パッカードデベロップメントカンパニーエル．ピー．|All-optical high-speed distributed arbitration in computer system equipment|
    US7532785B1|2007-10-23|2009-05-12|Hewlett-Packard Development Company, L.P.|Photonic interconnects for computer system devices|
    BRPI0821180B1|2008-03-11|2019-01-22|Hewlett Packard Entpr Dev Lp|opto-electronic key and opto-electronic key element|
    CN102090000B|2008-05-09|2015-04-22|惠普开发有限公司|Optically enabled broadcast bus|
    EP2289184A4|2008-06-17|2012-10-31|Hewlett Packard Development Co|Optical broadcast systems and methods|
    US8385740B2|2008-07-18|2013-02-26|Oracle America, Inc.|Arbitration scheme for an optical bus|
    KR101444514B1|2008-10-31|2014-09-24|휴렛-팩커드 디벨롭먼트 컴퍼니, 엘.피.|Prioritized optical arbitration systems and methods|
    US8472808B2|2010-10-06|2013-06-25|Hewlett-Packard Development Company, L.P.|Optical data path systems|CN105763962B|2014-12-18|2019-05-10|华为技术有限公司|Network on mating plate, optical router and the method for transmitting signal|
    US10009135B2|2015-02-06|2018-06-26|The Trustees Of Princeton University|System and method for photonic processing|
    US10203454B2|2016-05-31|2019-02-12|Futurewei Technologies, Inc.|Dense wavelength-division multiplexingnetwork and method|
    US10845550B1|2019-10-18|2020-11-24|The Boeing Company|Input coupler for chip-scale laser receiver device|
    CN111431606A|2020-04-15|2020-07-17|联合华芯电子有限公司|Bus type networking system for bidirectional optical fiber communication|
    法律状态:
    2015-12-15| PLFP| Fee payment|Year of fee payment: 3 |
    2016-12-16| PLFP| Fee payment|Year of fee payment: 4 |
    2017-12-14| PLFP| Fee payment|Year of fee payment: 5 |
    2019-09-27| ST| Notification of lapse|Effective date: 20190906 |
    优先权:
    申请号 | 申请日 | 专利标题
    FR1362301A|FR3014563B1|2013-12-09|2013-12-09|DEVICE AND METHOD FOR OPTICAL ARBITRATION IN A CHIP NETWORK SYSTEM|FR1362301A| FR3014563B1|2013-12-09|2013-12-09|DEVICE AND METHOD FOR OPTICAL ARBITRATION IN A CHIP NETWORK SYSTEM|
    EP14195870.2A| EP2882122B1|2013-12-09|2014-12-02|Device and method for optical arbitration in a network|
    US14/561,372| US9369784B2|2013-12-09|2014-12-05|Optical arbitration device and method in a network-on-chip system|
    [返回顶部]