CN109246493B - Optical network-on-chip architecture for multicast broadcast communication perception and communication method - Google Patents
Optical network-on-chip architecture for multicast broadcast communication perception and communication method Download PDFInfo
- Publication number
- CN109246493B CN109246493B CN201810920131.XA CN201810920131A CN109246493B CN 109246493 B CN109246493 B CN 109246493B CN 201810920131 A CN201810920131 A CN 201810920131A CN 109246493 B CN109246493 B CN 109246493B
- Authority
- CN
- China
- Prior art keywords
- communication
- optical
- unit
- resonators
- resonator
- Prior art date
- Legal status (The legal status 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 status listed.)
- Active
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04Q—SELECTING
- H04Q11/00—Selecting arrangements for multiplex systems
- H04Q11/0001—Selecting arrangements for multiplex systems using optical switching
- H04Q11/0005—Switch and router aspects
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04Q—SELECTING
- H04Q11/00—Selecting arrangements for multiplex systems
- H04Q11/0001—Selecting arrangements for multiplex systems using optical switching
- H04Q11/0062—Network aspects
- H04Q11/0067—Provisions for optical access or distribution networks, e.g. Gigabit Ethernet Passive Optical Network (GE-PON), ATM-based Passive Optical Network (A-PON), PON-Ring
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04Q—SELECTING
- H04Q11/00—Selecting arrangements for multiplex systems
- H04Q11/0001—Selecting arrangements for multiplex systems using optical switching
- H04Q11/0062—Network aspects
- H04Q11/0071—Provisions for the electrical-optical layer interface
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04Q—SELECTING
- H04Q11/00—Selecting arrangements for multiplex systems
- H04Q11/0001—Selecting arrangements for multiplex systems using optical switching
- H04Q11/0005—Switch and router aspects
- H04Q2011/0007—Construction
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04Q—SELECTING
- H04Q11/00—Selecting arrangements for multiplex systems
- H04Q11/0001—Selecting arrangements for multiplex systems using optical switching
- H04Q11/0005—Switch and router aspects
- H04Q2011/0037—Operation
- H04Q2011/0047—Broadcast; Multicast
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04Q—SELECTING
- H04Q11/00—Selecting arrangements for multiplex systems
- H04Q11/0001—Selecting arrangements for multiplex systems using optical switching
- H04Q11/0062—Network aspects
- H04Q2011/0079—Operation or maintenance aspects
Abstract
The invention provides an optical network-on-chip architecture for multicast broadcast communication perception and a communication method, which are used for solving the technical problems of high power consumption and poor expandability of the optical network-on-chip architecture in the prior art and the technical problems of high time delay and high power consumption of the communication method. In the communication method, the global control unit processes the communication request of the processor module, and sends the obtained control information to the multi-wavelength laser source and the first active micro-ring resonator for network communication configuration.
Description
Technical Field
The invention belongs to the technical field of communication, and particularly relates to an optical network-on-chip architecture for multicast broadcast communication perception and a communication method, which can be used for multicast broadcast communication between processor modules in a computing system chip.
Background
With the exponential increase of the number of processor cores integrated on a single chip, high-performance communication between the processor cores becomes a key for improving the performance of a many-core computing system on a chip. Compared with the traditional on-chip communication architecture, the on-chip network based on the computer network design idea can effectively improve the communication performance among the processor cores and becomes a preferred scheme of a many-core computing system on a chip. With the continuous expansion of the scale of the network on chip and the development of high-performance computing application, the current network on chip based on electrical interconnection faces the limitation of the bandwidth, time delay, power consumption and other factors of communication among processor cores. The optical on-chip network based on optical interconnection can effectively break through the performance limit of the traditional electrical interconnection on-chip network, provide higher bandwidth density, smaller communication time delay and lower system power consumption, and effectively improve the communication performance between processor cores. Therefore, the network on optical chip has become a key technology of many-core computing systems on chip in the communication technology field.
Meanwhile, the wide application of a cache coherence protocol in the network on chip and the occurrence of computation-intensive and data-intensive applications and the like enable the multicast broadcast communication proportion in the inter-processor core communication to be increased continuously, and the highest proportion reaches 52.4%. Meanwhile, researches indicate that 5% of multicast broadcast communication can cause the saturation point of network load to be reduced by 42% -74%, and the average time delay is increased by 7-22 times. Therefore, designing an optical network-on-chip architecture and a communication method for effectively supporting multicast broadcast communication has become a key for improving the performance of the on-chip multi-core computing system.
At present, how to reduce the wavelength resource overhead and the optical insertion loss in the network and improve the power consumption, expandability and parallelism of the network by reasonable layout design of optical devices such as optical waveguides, micro-ring resonators and the like is a main research content of the design of an optical network-on-chip architecture supporting multicast broadcast communication at present. Based on the optical power separation technology of the micro-ring resonator, the coupling distance between the micro-ring resonator and the power waveguide is set to realize N equal division of optical power (currently, the maximum power can be divided into 8 equal divisions), so that wavelength resources are fully utilized, and the optical power separation technology can be effectively applied to optical network architecture design on an optical chip. Meanwhile, how to improve the time delay, bandwidth and power consumption performance of multicast broadcast communication is a main research content for supporting the design of a multicast broadcast communication method at present.
The existing optical network architecture supporting multicast broadcast communication is, in view of the present public data:
the patent application with the application publication number of CN103442311A and the name of 'network on optical chip system supporting multicast communication' discloses a network on optical chip system supporting multicast communication. The implementation mode of the network on optical chip system is as follows: optical waveguides and multi-wavelength micro-ring resonators in the first-stage optical switching network and the second-stage optical switching network are reasonably arranged, each communication node can be ensured to simultaneously send optical signals to all other communication nodes, and the function of multicast broadcast communication is achieved. The optical network-on-chip system has the defects that the number of micro-ring resonators and optical waveguides used in two stages of optical switching units under the same scale is large, the introduced optical insertion loss is high, the problem of high power consumption of the network is caused, and in addition, the problem of poor expandability is caused by large area overhead of the layout design of the optical waveguides in the two stages of networks.
A paper entitled "Extending the Performance and Energy-Efficiency of Shared Memory with Nanophotonic Technology" was published by random Morris et al, university of Ohio, in 2014, IEEE Transactions on Parallel and Distributed Systems, journal 2. The paper discloses an optical network-on-chip architecture supporting multicast broadcast communication based on a tree form and a communication method. The proposed optical on-chip network architecture realizes multicast broadcast communication in the whole tree network by adopting a one-to-two power splitter and active micro-ring resonance combination in each tree branch. The optical network-on-chip architecture has the disadvantage that the optical power splitter adopted can cause higher optical insertion loss, thereby causing larger power consumption of the laser source. Meanwhile, in the proposed communication method, due to the fact that optical token arbitration is used, extra photoelectric conversion is introduced into control configuration in the communication process, and communication delay is high; in addition, the adoption of the wavelength allocation based on the communication nodes in the communication method causes more optical signals generated by the laser source in each communication, and increases the power consumption overhead of the laser source.
Feiyang Liu et al, the university of Western electronic technology, published a paper named "Dynamic Ring-Based Multicast with wavelet Reuse for Optical Network on Chips" in an IEEE 10th International Symposium on Embedded Multi/Man-core Systems-on-chip (MCSOC) International conference 2016, and disclosed an Optical Network architecture and a communication method for supporting Multicast broadcast communication Based on rings. The network architecture on the optical chip is composed of a processor core layer, a control layer and a transmission layer, wherein the control layer receives a multicast request from the processor core layer and distributes a multicast ring in the control layer as a routing path through a multicast ring distributor, and the transmission layer realizes multicast broadcast communication transmission according to the routing path. The optical network-on-chip architecture and the communication method have the defects that extra time delay is introduced in the control configuration process of multi-level communication, the optical insertion loss in the transmission process is increased, and the time delay and the power consumption of multicast broadcast communication are larger.
Disclosure of Invention
The invention aims to overcome the defects in the prior art, and provides an optical network-on-chip architecture for multicast broadcast communication perception, which is used for solving the technical problems of high power consumption and poor expandability of the optical network-on-chip architecture in the prior art and the technical problems of high time delay and high power consumption of a communication method.
In order to achieve the purpose, the invention adopts the technical scheme that:
an optical network-on-chip architecture for multicast broadcast communication awareness, comprising a multi-wavelength laser source 1, a communication sub-network 2, a global control unit 3 and n processor modules 4, n ∈ [8,256] and taking a multiple of 8, wherein:
the multi-wavelength laser source 1 is used for providing optical signals to the communication subnet 2;
the communication sub-network 2 comprises an optical waveguide 21, n coupling ports 22 and n first active micro-ring resonators 23; the optical waveguide 21 is used for realizing optical signal transmission between the processor modules 4; the first active microring resonator 23 is configured to couple the optical signal in the optical waveguide 21 into the corresponding coupling port 22; the coupling port 22 is used for transmitting the optical signal in the optical waveguide 21 to the corresponding processor module 4;
the global control unit 3 comprises a buffer queue 31, an arbitration unit 32 and an analysis unit 33; the buffer queue 31 is used for receiving and storing the communication request of the processor module 4; the arbitration unit 32 is configured to arbitrate the communication requests in the buffer queue 31; the analysis unit 33 is configured to analyze the arbitrated communication request, and send the analyzed control information to the multi-wavelength laser source 1 and the first active micro-ring resonator 23;
the processor module 4 is used for receiving, processing, sending and storing data, and modulating and demodulating optical signals;
the n processor modules 4 form k clusters arranged in an x by y array, k being n/8,
each cluster comprises an array consisting of 2 x 4 processor modules 4, each processor module 4 is connected with a coupling port 22, and each coupling port 22 is coupled and connected with a first active microring resonator 23; the optical waveguide 21 extends downwards after being connected with the 1 st row processor module 4 in the first row Cluster along the output direction of the multi-wavelength laser source 1, is reversely connected with the 2 nd row processor module in the first row Cluster until being connected with the processor modules 4 in the rest rows of clusters according to the method, then extends upwards to sequentially pass through the middles of the 1 st row active micro-ring resonators 23 in the first row Cluster and the 2 nd row active micro-ring resonators 23 in the first row Cluster, extends downwards to reversely pass through the middles of the 1 st row active micro-ring resonators in the second row Cluster and the 2 nd row active micro-ring resonators in the second row Cluster until passing through all the first active micro-ring resonators 23 in the Cluster k to form a snake-shaped optical waveguide structure;
the coupling distance between the first active microring resonator 23 in the same one of the k clusters and the optical waveguide 21 is set according to the principle of a power splitting system; the resonators of the first active microring resonator 23 in the same cluster are the same in size, the resonators of the first active microring resonator 23 in different clusters are different in size, and the resonance wavelength of each resonator is determined by its own size.
The above optical network on chip architecture for sensing multicast broadcast communication, the processor module 4 includes 4 processor cores 41, 4 private primary caches 42, a shared secondary cache 43, a demodulation unit 44, and a modulation unit 45, where:
the processor core 41 is configured to communicate with the private first-level cache 42, and process data according to an instruction;
the private first-level cache 42 is used for storing data required by the processor core 41 and communicating with the processor core 41 and the shared second-level cache 43;
the shared second-level cache 43 is used for storing data required by the private first-level cache 42; while communicating with private level one cache 42, modulation unit 44 and demodulation unit 45;
the modulation unit 44 is configured to modulate the optical signal in the optical waveguide 21 according to the electrical signal sent by the shared secondary buffer 43;
the demodulation unit 45 is configured to demodulate the optical signal in the coupled port 22 into an electrical signal received by the shared second-level buffer.
The above optical network-on-chip architecture for sensing multicast broadcast communication, the modulation unit 44 includes a serialization unit 441, a driving circuit 442, and k second active microring resonators 443, where:
the serialization unit 441 is configured to serialize the multiple paths of low-speed electrical signals output by the shared second-level cache 43 into a path of high-speed electrical signals;
the driving circuit 442 is configured to control an operating state of the second active micro-ring resonator 443 according to the high-speed electrical signal generated by the serialization unit 441;
the sizes of the resonators in the k second active microring resonators 443 are different, and the resonance wavelength of each resonator is determined by its own size, and is used for modulating an optical signal corresponding to the resonance wavelength in the optical waveguide 21.
The above optical network-on-chip architecture for sensing multicast broadcast communication, the demodulation unit 45 includes a deserialization unit 451, a detection circuit 452, and a passive micro-ring resonator 453, where:
the deserializing unit 451 is configured to deserialize the one-path high-speed electrical signal sent by the detection circuit 452, and transmit the obtained multiple paths of low-speed electrical signals to the shared secondary cache 43;
the detection circuit 452 is configured to transmit a path of high-speed electrical signal generated according to the demodulation information of the passive micro-ring resonator 453 to the deserializing unit 451;
the passive microring resonator 453 is configured to transmit demodulation information obtained by demodulating the optical signal in the coupling port 22 to the detection circuit 452.
In the optical network-on-chip architecture for sensing multicast broadcast communication, the resonators in the passive microring resonators 453 in the same cluster have the same size, the resonators in the passive microring resonators 453 in different clusters have different sizes, and the resonance wavelength of each resonator is determined by the size of the resonator, and is used for demodulating the optical signal corresponding to the resonance wavelength in the coupling port 22.
A communication method of network architecture on optical chip for sensing multicast broadcast communication comprises the following steps:
(1) the processor module as a source node sends a communication request to the global control unit:
(1a) the processor core sends a request to the private first-level cache;
(1b) the private first-level cache judges whether the data requested by the processor core is stored, if so, the data is sent to the processor core, the communication is finished, otherwise, the private first-level cache sends a request to the shared second-level cache, and the step (1c) is executed;
(1c) the shared second-level cache judges whether the requested data are stored, if so, the data are sent to the processor core through the private first-level cache, the communication is finished, and otherwise, a communication request is sent to the global control unit;
(2) the global control unit processes the communication request:
(2a) the buffer queue receives and stores the communication request;
(2b) judging whether the communication request is positioned at the head of the cache queue by the arbitration unit, if so, sending the communication request to the analysis unit, and executing the step (2d), otherwise, executing the step (2 c);
(2c) the arbitration unit advances the position of the communication request in the buffer queue by 1 and executes the step (2 b);
(2d) the analysis unit analyzes the communication request, sends control information obtained by analysis to the multi-wavelength laser source and the first active micro-ring resonator, and simultaneously sends a permission response to the processor module serving as a source node;
(3) the multi-wavelength laser source generates an optical signal:
the multi-wavelength laser source generates an optical signal corresponding to the wavelength required by the communication according to the control information and inputs the optical signal into the optical waveguide of the communication subnet;
(4) the processor module as a source node modulates an optical signal:
(4a) after the shared secondary cache obtains the permission response, sending a communication electric signal to a sequencer in the modulation unit;
(4b) a sequencer in the modulation unit serializes the communication electric signals and sends the obtained high-speed electric signals to a driving circuit;
(4c) the driving circuit controls the working state of the second active micro-ring device through a high-speed electric signal to realize the modulation of the optical signal in the optical waveguide and obtain the optical signal carrying communication information;
(5) the first active micro-ring resonator is coupled with the modulated optical signal;
the first active resonator with the same resonance wavelength as the wavelength required by the communication contained in the control information is set to be in an open state, and the optical signal carrying the communication information in the optical waveguide is coupled into the coupling port and then transmitted to the processor module connected with the coupling port;
(6) the processor module as the destination node processes the optical signal:
(6a) the passive micro-ring resonator in the demodulation unit demodulates the optical signal light loaded with communication information and input by the coupling port, and transmits the obtained demodulation information to the detection circuit;
(6b) the detection circuit integrates the demodulation information and sends the obtained high-speed electric signal to the deserializing unit;
(6c) the deserializing unit deserializes the high-speed electric signal and sends the obtained low-speed electric signal to a shared secondary cache;
(6d) and the shared second-level cache sends the low-speed electric signal to a corresponding processor core through the private first-level cache, and the communication is finished.
Compared with the prior art, the invention has the following advantages:
firstly, the invention adopts the mode that the processor modules are divided into clusters which are arranged in an array, each cluster comprises an array consisting of 2 multiplied by 4 processor modules, each processor module is coupled and connected with one first active micro-ring resonator through a coupling port, the clusters which are arranged in the array are connected through a snakelike optical waveguide structure, and the coupling distance distribution between the first active micro-ring resonator and the optical waveguide is set by the principle of a power separation system, so that the use number of the optical waveguide and the micro-ring resonator in the network can be effectively reduced, the optical insertion loss in the network is effectively reduced, and the power consumption problem of the network is solved; meanwhile, regular scale expansion is carried out by adopting an array arrangement mode taking clusters as units, the whole area overhead of the network and the use number of the optical waveguides and the micro-ring resonators can be effectively reduced, and the expandability of the network is further effectively improved.
Secondly, the mode of sending control information to the multi-wavelength laser source by the global control unit is adopted, so that the multi-wavelength laser source only generates optical signals corresponding to the wavelengths required by the communication at each time of communication, the overall power overhead of the laser source is effectively reduced, and the power consumption of the multicast broadcast communication is further effectively reduced.
Thirdly, because the invention adopts the electricity-based global control unit to arbitrate the network communication request and configures the multicast broadcast communication path by controlling the working state of the active micro-ring resonator, the photoelectric conversion times and the configuration time overhead in the communication process are effectively reduced, and the time delay of the multicast broadcast communication is further effectively reduced.
Drawings
Fig. 1 is a schematic diagram of an optical network-on-chip architecture for multicast broadcast sensing according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of communication subnets in an embodiment of the present invention;
FIG. 3 is a schematic diagram of a global control unit according to the present invention;
FIG. 4 is a schematic diagram of a processor module of the present invention;
FIG. 5 is a schematic diagram of a modulation unit according to an embodiment of the present invention;
FIG. 6 is a schematic diagram of a demodulation unit according to an embodiment of the present invention;
FIG. 7 is a flow chart of the communication method of the present invention
Detailed Description
The invention is described in further detail below with reference to the figures and the specific embodiments.
With reference to fig. 1, comprising a multi-wavelength laser source 1, a communication sub-network 2, a global control unit 3 and 32 processor modules 4, wherein:
the multi-wavelength laser source 1 is configured to provide an optical signal corresponding to a wavelength required by each communication to the communication subnet 2 according to the control information sent by the global control unit 3;
referring to fig. 2, the communication sub-network 2 comprises an optical waveguide 21, 32 coupling ports 22 and 32 first active micro-ring resonators 23; the optical waveguide 21 is used for realizing optical signal transmission among the 32 processor modules 4; the 32 first active microring resonators 23 are used for coupling the optical signals in the optical waveguide 21 into the corresponding coupling ports 22; the 32 coupling ports 22 for transmitting the optical signals in the optical waveguide 21 to the respective processor modules 4;
referring to fig. 3, the global control unit 3 includes a buffer queue 31, an arbitration unit 32, and an analysis unit 33; the buffer queue 31 is used for receiving and storing the communication request of the processor module 4; the arbitration unit 32 is configured to arbitrate the communication requests in the buffer queue 31; the analysis unit 33 is configured to analyze the arbitrated communication request, and send the analyzed control information to the multi-wavelength laser source 1 and the first active micro-ring resonator 23;
referring to fig. 4, the processor module 4 is configured to receive, process, transmit, and store data, and simultaneously modulate and demodulate an optical signal; the processor module 4 includes 4 processor cores 41, 4 private primary caches 42, a shared secondary cache 43, a demodulation unit 44, and a modulation unit 45, where: the processor core 41 is configured to communicate with the private first-level cache 42, and process data according to an instruction; the private first-level cache 42 is used for storing data required by the processor core 41 and communicating with the processor core 41 and the shared second-level cache 43; the shared second-level cache 43 is used for storing data required by the private first-level cache 42; while communicating with private level one cache 42, modulation unit 44 and demodulation unit 45;
the 32 processor modules 4 form 4 clusters arranged in a 2 × 2 array, each cluster comprises an array formed by 2 × 4 processor modules 4, each processor module 4 is connected with one coupling port 22, and each coupling port 22 is coupled and connected with one first active microring resonator 23; the optical waveguide 21 is connected with the 1 st row processor module 4 in the first row Cluster along the output direction of the multi-wavelength laser source 1, then extends downwards, is reversely connected with the 2 nd row processor module 4 in the first row Cluster until the connection with the processor modules 4 in the other row clusters is completed according to the method, then extends upwards to sequentially pass through the middles of the 1 st row and the 2 nd row first active micro-ring resonators 23 in the first row Cluster, extends downwards to sequentially pass through the middles of the 1 st row and the 2 nd row active micro-ring resonators in the second row Cluster in a reverse direction until all the first active micro-ring resonators 23 in the Cluster 4 are passed, and a snake-shaped optical waveguide structure is formed;
the coupling distance between the first active microring resonator 23 in the same one of the 4 clusters and the optical waveguide 21 is set according to the principle of a power splitting system. The resonators of the first active microring resonator 23 in the same cluster are the same in size, the resonators of the first active microring resonator 23 in different clusters are different in size, and the resonance wavelength of each resonator is determined by its own size. The power separation system realizes N equal division of optical power in the power waveguide by setting the coupling distance between the micro-ring resonators and the power waveguide, namely 1/N of the coupling optical power of each micro-ring resonator on the power waveguide. At present, the highest available technology can realize 8 equal divisions of optical power in a power waveguide, the invention adopts the technology to set the coupling distance between 8 active micro-ring resonators in the same cluster and the optical waveguide, and each active micro-ring resonator can couple 1/8 of optical signal power corresponding to the resonance wavelength in the optical waveguide. The layout mode of the active micro-ring resonators can effectively reduce the number of micro-ring resonators used in a network architecture, so that the insertion loss of the network is reduced.
Referring to fig. 5, the modulation unit 44 is configured to modulate the optical signal in the optical waveguide 21 according to the electrical signal sent by the shared secondary buffer 43; the modulation unit 44 includes a serialization unit 441, a driving circuit 442, and k second active microring resonators 443, wherein: the serialization unit 441 is configured to serialize the multiple paths of low-speed electrical signals output by the shared second-level cache 43 into a path of high-speed electrical signals; the driving circuit 442 is configured to control an operating state of the second active micro-ring resonator 443 according to the high-speed electrical signal generated by the serialization unit 441; the sizes of the resonators in the k second active microring resonators 443 are different, and the resonance wavelength of each resonator is determined by its own size, and is used for modulating an optical signal corresponding to the resonance wavelength in the optical waveguide 21.
Referring to fig. 6, the demodulation unit 45 is configured to demodulate an optical signal in the coupled port 22 into an electrical signal received by the shared second-level buffer; the demodulation unit 45 includes a deserializing unit 451, a detection circuit 452, and a passive micro-ring resonator 453, in which: the deserializing unit 451 is configured to deserialize the one-path high-speed electrical signal sent by the detection circuit 452, and transmit the obtained multiple paths of low-speed electrical signals to the shared secondary cache 43; the detection circuit 452 is configured to integrate the demodulation information of the passive micro-ring resonator 453 into a high-speed electrical signal and transmit the high-speed electrical signal to the deserializing unit 451; the passive microring resonator 453 is configured to transmit demodulation information obtained by demodulating the optical signal in the coupling port 22 to the detection circuit 452. The resonators in the passive microring resonators 453 in the same cluster have the same size, the resonators in the passive microring resonators 453 in different clusters have different sizes, and the resonance wavelength of each resonator is determined by the size of the resonator, and is used for demodulating the optical signal corresponding to the resonance wavelength in the coupling port 22.
The structure of example 2 is the same as that of example 1, with the following parameters adjusted:
the 8 processor modules 4 form 1 cluster arranged in a 1 × 1 array, and the cluster contains an array consisting of 2 × 4 processor modules 4.
The structure of example 3 is the same as that of example 1, with the following parameters adjusted:
the 256 processor modules 4 form 32 clusters arranged in a 16 × 16 array, and each cluster contains an array of 2 × 4 processor modules 4.
Referring to fig. 7, a communication method of an optical network-on-chip architecture for multicast broadcast communication awareness includes the following steps:
step 1) the processor module as the source node sends a communication request to the global control unit:
step 1a) a processor core sends a request to a private first-level cache;
step 1b), the private first-level cache judges whether data requested by a processor core is stored, if so, the data is sent to the processor core, the communication is finished, otherwise, the private first-level cache sends a request to the shared second-level cache, and step 1c) is executed;
step 1c), the shared secondary cache judges whether the requested data is stored, if so, the data is sent to the processor core through the private primary cache, the communication is finished, otherwise, a communication request is sent to the global control unit;
step 2), the global control unit processes the communication request:
step 2a) the buffer queue receives and stores the communication requests, and the communication requests are sequentially stored in the buffer queue according to the sending time sequence, namely the storage positions of the communication requests in the buffer queue are arranged according to the arrival time sequence;
step 2b) the arbitration unit judges whether the communication request is positioned at the head of the cache queue, if so, the communication request is sent to the analysis unit, and step 2d) is executed, otherwise, step 2c) is executed;
step 2c) the arbitration unit advances the position of the communication request in the buffer queue by 1 and executes step 2 b); when the arbitration unit finds that the current request is not positioned at the head of the cache queue, the arbitration unit firstly sends out the communication request positioned at the head and then advances the positions of the communication requests in the cache queue to 1;
step 2d), the analysis unit analyzes the communication request, sends control information obtained by analysis to the multi-wavelength laser source and the first active micro-ring resonator, and sends permission response to the processor module serving as the source node; the control information sent to the laser source mainly comprises wavelength information required by the communication, and the control information sent to the first active micro-ring resonator comprises working state information of the first active micro-ring controller;
step 3), generating an optical signal by a multi-wavelength laser source:
the multi-wavelength laser source generates an optical signal corresponding to the wavelength required by the communication according to the control information and inputs the optical signal into the optical waveguide of the communication subnet; according to the difference of the target nodes of multicast broadcast communication and the difference of the wavelength required by each communication, the multi-wavelength laser source only generates the optical signal corresponding to the wavelength required by each communication according to the control information of the global control unit in the communication method, so that the overall power overhead of the laser source can be effectively reduced.
Step 4) the processor module as the source node modulates the optical signal:
step 4a), after the shared secondary cache obtains the permission response, sending a communication electric signal to a sequencer in the modulation unit, wherein the electric signal is a multi-path low-speed electric signal;
step 4b) a sequencer in the modulation unit sequences the multiple paths of low-speed electric signals for communication and sends the obtained path of high-speed electric signals to a driving circuit;
step 4c), the driving circuit controls the working state of the second active micro-ring device through a high-speed electric signal to realize the modulation of the optical signal in the optical waveguide and obtain the optical signal carrying communication information;
step 5), coupling the modulated optical signal by the first active micro-ring resonator;
the first active resonator with the same resonant wavelength as the wavelength required by the communication contained in the control information is set to be in an open state, so that the accurate configuration of a multicast broadcast communication path in a network is realized, the extra communication time delay caused by multi-level communication configuration and optical token arbitration in the prior art is reduced, and an optical signal carrying the communication information in the optical waveguide is coupled into the coupling port and then transmitted to the processor module connected with the coupling port;
step 6) processing the optical signal as a processor module of the destination node:
step 6a) the passive micro-ring resonator in the demodulation unit demodulates the optical signal light loaded with communication information and input by the coupling port, and transmits the obtained demodulation information to the detection circuit;
step 6b) the detection circuit integrates the demodulation information and sends the obtained high-speed electric signal to a deserializing unit;
step 6c), the deserializing unit deserializes the high-speed electric signal and sends the obtained multi-path low-speed electric signal to a shared secondary cache;
and 6d) the shared secondary cache sends the low-speed electric signal to a corresponding processor core through the private primary cache, and the communication is finished.
Claims (6)
1. An optical network-on-chip system for multicast broadcast communication awareness, comprising a multi-wavelength laser source (1), a communication sub-network device (2), a global control unit (3) and n processor modules (4), n being [8,256] and taking a multiple of 8, wherein:
the multi-wavelength laser source (1) is used for providing optical signals to the communication subnet device (2);
the communication sub-network device (2) comprising an optical waveguide (21), n coupling ports (22) and n first active micro-ring resonators (23); the optical waveguide (21) is used for realizing optical signal transmission between the processor modules (4); -said first active microring resonator (23) for coupling an optical signal in an optical waveguide (21) into a respective coupling port (22); -said coupling port (22) for transmitting the optical signals in the optical waveguide (21) to the respective processor module (4);
the global control unit (3) comprises a buffer queue (31), an arbitration unit (32) and an analysis unit (33); the buffer queue (31) is used for receiving and storing the communication request of the processor module (4); the arbitration unit (32) is used for arbitrating the communication requests in the buffer queue (31); the analysis unit (33) is used for analyzing the arbitrated communication request and sending the analyzed control information to the multi-wavelength laser source (1) and the first active micro-ring resonator (23);
the processor module (4) is used for receiving, processing, sending and storing data, and modulating and demodulating optical signals;
the method is characterized in that:
the n processor modules (4) form k clusters arranged in an x y array, k being n/8,
each cluster comprises an array consisting of 2 x 4 processor modules (4), each processor module (4) is connected with a coupling port (22), and each coupling port (22) is coupled with a first active micro-ring resonator (23); the optical waveguide (21) is connected with the 1 st row of processor modules (4) in the first row of clusters along the output direction of the multi-wavelength laser source (1), then extends downwards, is reversely connected with the 2 nd row of processor modules (4) in the first row of clusters until the connection with the processor modules (4) in the other rows of clusters is completed according to the method, then extends upwards to sequentially pass through the middle of the 1 st row of first active micro-ring resonators (23) in the first row of clusters and the 2 nd row of first active micro-ring resonators (23), extends downwards to sequentially pass through the middle of the 1 st row of active micro-ring resonators in the second row of clusters and the 2 nd row of active micro-ring resonators in the second row of clusters in a reverse direction until all the first active micro-ring resonators (23) in a Cluster k;
a first active microring resonator (23) in the same one of the k clusters, the coupling distance between which and the optical waveguide (21) is set according to the power splitting system principle; the resonators of the first active microring resonators (23) in the same cluster are the same in size, the resonators of the first active microring resonators (23) in different clusters are different in size, and the resonance wavelength of each resonator is determined by the size of the resonator.
2. The optical network-on-chip system for multicast broadcast communication awareness according to claim 1, wherein the processor module (4) comprises 4 processor cores (41), 4 private level one caches (42), a shared level two cache (43), a demodulation unit (44), and a modulation unit (45), wherein:
the processor core (41) is used for communicating with the private first-level cache (42) and processing data according to instructions;
the private first-level cache (42) is used for storing data required by the processor core (41) and communicating with the processor core (41) and the shared second-level cache (43);
the shared second-level cache (43) is used for storing data required by the private first-level cache (42); simultaneously communicating with a private level one buffer (42), a modulation unit (44) and a demodulation unit (45);
the modulation unit (44) is used for modulating the optical signal in the optical waveguide (21) according to the electric signal sent by the shared secondary buffer (43);
the demodulation unit (45) is used for demodulating the optical signals in the coupling port (22) into the electric signals received by the shared secondary buffer.
3. The optical network-on-chip system for multicast broadcast communication awareness according to claim 2, wherein the modulation unit (44) comprises a serialization unit (441), a driving circuit (442), and k second active micro-ring resonators (443), wherein:
the serialization unit (441) is used for serializing a plurality of paths of low-speed electric signals output by the shared second-level cache (43) into a path of high-speed electric signals;
the driving circuit (442) is used for controlling the working state of the second active micro-ring resonator (443) according to the high-speed electric signal generated by the serialization unit (441);
the sizes of the resonators in the k second active microring resonators (443) are different, and the resonance wavelength of each resonator is determined by the size of the resonator, and is used for modulating the optical signal corresponding to the resonance wavelength in the optical waveguide (21).
4. An optical network-on-chip system for multicast broadcast communication awareness according to claim 2, wherein the demodulation unit (45) comprises a deserialization unit (451), a detection circuit (452), and a passive micro-ring resonator (453), wherein:
the deserializing unit (451) is used for deserializing the high-speed electric signal sent by the detection circuit (452), and transmitting the obtained multiple paths of low-speed electric signals to the shared secondary buffer (43);
the detection circuit (452) is used for integrating demodulation information of the passive micro-ring resonator (453) into a high-speed electric signal and transmitting the high-speed electric signal to the deserializing unit (451);
the passive micro-ring resonator (453) is used for transmitting demodulation information obtained by demodulating the optical signal in the coupling port (22) to the detection circuit (452).
5. The optical network-on-chip system for multicast broadcast communication awareness according to claim 4, wherein the resonators in the passive microring resonators (453) in the same cluster have the same size, the resonators in the passive microring resonators (453) in different clusters have different sizes, and the resonant wavelength of each resonator is determined by its own size and is used for demodulating the optical signal corresponding to the resonant wavelength in the coupling port (22).
6. A communication method of a network-on-chip system for multicast broadcast communication perception is characterized by comprising the following steps:
(1) the processor module as a source node sends a communication request to the global control unit:
(1a) the processor core sends a request to the private first-level cache;
(1b) the private first-level cache judges whether the data requested by the processor core is stored, if so, the data is sent to the processor core, the communication is finished, otherwise, the private first-level cache sends a request to the shared second-level cache, and the step (1c) is executed;
(1c) the shared second-level cache judges whether the requested data are stored, if so, the data are sent to the processor core through the private first-level cache, the communication is finished, and otherwise, a communication request is sent to the global control unit;
(2) the global control unit processes the communication request:
(2a) the buffer queue receives and stores the communication request;
(2b) judging whether the communication request is positioned at the head of the cache queue by the arbitration unit, if so, sending the communication request to the analysis unit, and executing the step (2d), otherwise, executing the step (2 c);
(2c) the arbitration unit advances the position of the communication request in the buffer queue by 1 and executes the step (2 b);
(2d) the analysis unit analyzes the communication request, sends control information obtained by analysis to the multi-wavelength laser source and the first active micro-ring resonator, and simultaneously sends a permission response to the processor module serving as a source node;
(3) the multi-wavelength laser source generates an optical signal:
the multi-wavelength laser source generates an optical signal corresponding to the wavelength required by the communication according to the control information and inputs the optical signal into an optical waveguide of the communication subnet device;
(4) the processor module as a source node modulates an optical signal:
(4a) after the shared secondary cache obtains the permission response, sending a communication electric signal to a sequencer in the modulation unit;
(4b) a sequencer in the modulation unit serializes the communication electric signals and sends the obtained high-speed electric signals to a driving circuit;
(4c) the driving circuit controls the working state of the second active micro-ring device through a high-speed electric signal to realize the modulation of the optical signal in the optical waveguide and obtain the optical signal carrying communication information;
(5) the first active micro-ring resonator is coupled with the modulated optical signal;
the first active resonator with the same resonance wavelength as the wavelength required by the communication contained in the control information is set to be in an open state, and the optical signal carrying the communication information in the optical waveguide is coupled into the coupling port and then transmitted to the processor module connected with the coupling port;
(6) the processor module as the destination node processes the optical signal:
(6a) the passive micro-ring resonator in the demodulation unit demodulates the optical signal light loaded with communication information and input by the coupling port, and transmits the obtained demodulation information to the detection circuit;
(6b) the detection circuit integrates the demodulation information and sends the obtained high-speed electric signal to the deserializing unit;
(6c) the deserializing unit deserializes the high-speed electric signal and sends the obtained low-speed electric signal to a shared secondary cache;
(6d) and the shared second-level cache sends the low-speed electric signal to a corresponding processor core through the private first-level cache, and the communication is finished.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810920131.XA CN109246493B (en) | 2018-08-14 | 2018-08-14 | Optical network-on-chip architecture for multicast broadcast communication perception and communication method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810920131.XA CN109246493B (en) | 2018-08-14 | 2018-08-14 | Optical network-on-chip architecture for multicast broadcast communication perception and communication method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109246493A CN109246493A (en) | 2019-01-18 |
| CN109246493B true CN109246493B (en) | 2021-06-08 |
Family
ID=65071246
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810920131.XA Active CN109246493B (en) | 2018-08-14 | 2018-08-14 | Optical network-on-chip architecture for multicast broadcast communication perception and communication method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109246493B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110865969B (en) * | 2019-11-05 | 2022-05-31 | 中国人民解放军国防科技大学 | Method and device for supporting inter-processor interrupt communication of extensible processor |
| CN116546589B (en) * | 2023-06-05 | 2023-11-17 | 重庆邮电大学空间通信研究院 | Hybrid wireless on-chip optical network architecture based on passive reflection super surface |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8407660B2 (en) * | 2007-09-12 | 2013-03-26 | Neal Solomon | Interconnect architecture in three dimensional network on a chip |
| CN103442311A (en) * | 2013-08-31 | 2013-12-11 | 西安电子科技大学 | Optical network-on-chip system supporting multicast communication |
| CN105635861A (en) * | 2015-12-31 | 2016-06-01 | 西安电子科技大学 | Storage interconnection optical network framework based on micro-ring resonator and communication method thereof |
| CN106331909A (en) * | 2016-08-19 | 2017-01-11 | 西安电子科技大学 | Three-dimensional optical network-on-chip based on ring control optical network and communication method thereof |
-
2018
- 2018-08-14 CN CN201810920131.XA patent/CN109246493B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8407660B2 (en) * | 2007-09-12 | 2013-03-26 | Neal Solomon | Interconnect architecture in three dimensional network on a chip |
| CN103442311A (en) * | 2013-08-31 | 2013-12-11 | 西安电子科技大学 | Optical network-on-chip system supporting multicast communication |
| CN105635861A (en) * | 2015-12-31 | 2016-06-01 | 西安电子科技大学 | Storage interconnection optical network framework based on micro-ring resonator and communication method thereof |
| CN106331909A (en) * | 2016-08-19 | 2017-01-11 | 西安电子科技大学 | Three-dimensional optical network-on-chip based on ring control optical network and communication method thereof |
Non-Patent Citations (1)
| Title |
|---|
| 《光片上网络电路交换机制研究》;张博文;《中国优秀硕士学位论文全文数据库信息科技辑》;20170315;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109246493A (en) | 2019-01-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Joshi et al. | Silicon-photonic clos networks for global on-chip communication | |
| Batten et al. | Designing chip-level nanophotonic interconnection networks | |
| Mo et al. | A hierarchical hybrid optical-electronic network-on-chip | |
| Pan et al. | Firefly: Illuminating future network-on-chip with nanophotonics | |
| Werner et al. | Designing low-power, low-latency networks-on-chip by optimally combining electrical and optical links | |
| Hamedani et al. | QuT: A low-power optical network-on-chip | |
| Grani et al. | Flat-topology high-throughput compute node with AWGR-based optical-interconnects | |
| Chittamuru et al. | Swiftnoc: a reconfigurable silicon-photonic network with multicast-enabled channel sharing for multicore architectures | |
| Wu et al. | An inter/intra-chip optical network for manycore processors | |
| US11750531B2 (en) | FPGA-based virtual fabric for data center computing | |
| JP6607966B2 (en) | On-chip optical interconnect structure and network | |
| CN109246493B (en) | Optical network-on-chip architecture for multicast broadcast communication perception and communication method | |
| CN111698583B (en) | On-chip optical switching network architecture and routing algorithm for wavelength-mode hybrid multiplexing | |
| US20150071632A1 (en) | Reconfigurable optical interconnect network | |
| Wang et al. | A highly scalable optical network-on-chip with small network diameter and deadlock freedom | |
| CN107220209B (en) | Three-dimensional optical network-on-chip architecture based on faults, communication method and optical router | |
| CN109257662B (en) | Broadcast optical chip structure and communication method based on micro-ring power divider and grouping | |
| Channegowda et al. | Optical flyways for handling elephant flows to improve big data performance in SDN enabled Datacenters | |
| CN117221212B (en) | Optical network on chip low congestion routing method and related equipment | |
| Lee et al. | A shortest path adaptive routing technique for minimizing path collisions in hybrid optical network-on-chip | |
| Qi et al. | Testudo: A low latency and high-efficient memory-centric network using optical interconnect | |
| Mohamed et al. | Hermes: A hierarchical broadcast-based silicon photonic interconnect for scalable many-core systems | |
| Wang et al. | On the exploration of a low-power photonic network architecture | |
| Yan et al. | A novel non-cluster based architecture of hybrid electro-optical network-on-chip | |
| Li et al. | HPPI: A High-Performance Photonic Interconnect Design for Chiplet-Based DNN Accelerators |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |