CN115276820A - Method for setting power gradient of on-chip optical interconnection light source with mapping assistance - Google Patents
Method for setting power gradient of on-chip optical interconnection light source with mapping assistance Download PDFInfo
- Publication number
- CN115276820A CN115276820A CN202210909758.1A CN202210909758A CN115276820A CN 115276820 A CN115276820 A CN 115276820A CN 202210909758 A CN202210909758 A CN 202210909758A CN 115276820 A CN115276820 A CN 115276820A
- Authority
- CN
- China
- Prior art keywords
- light source
- source power
- node
- insertion loss
- power
- 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.)
- Granted
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B10/00—Transmission systems employing electromagnetic waves other than radio-waves, e.g. infrared, visible or ultraviolet light, or employing corpuscular radiation, e.g. quantum communication
- H04B10/50—Transmitters
- H04B10/564—Power control
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B10/00—Transmission systems employing electromagnetic waves other than radio-waves, e.g. infrared, visible or ultraviolet light, or employing corpuscular radiation, e.g. quantum communication
- H04B10/07—Arrangements for monitoring or testing transmission systems; Arrangements for fault measurement of transmission systems
- H04B10/075—Arrangements for monitoring or testing transmission systems; Arrangements for fault measurement of transmission systems using an in-service signal
- H04B10/079—Arrangements for monitoring or testing transmission systems; Arrangements for fault measurement of transmission systems using an in-service signal using measurements of the data signal
- H04B10/0795—Performance monitoring; Measurement of transmission parameters
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B10/00—Transmission systems employing electromagnetic waves other than radio-waves, e.g. infrared, visible or ultraviolet light, or employing corpuscular radiation, e.g. quantum communication
- H04B10/07—Arrangements for monitoring or testing transmission systems; Arrangements for fault measurement of transmission systems
- H04B10/075—Arrangements for monitoring or testing transmission systems; Arrangements for fault measurement of transmission systems using an in-service signal
- H04B10/079—Arrangements for monitoring or testing transmission systems; Arrangements for fault measurement of transmission systems using an in-service signal using measurements of the data signal
- H04B10/0795—Performance monitoring; Measurement of transmission parameters
- H04B10/07953—Monitoring or measuring OSNR, BER or Q
-
- 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
-
- 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
- H04Q2011/0083—Testing; Monitoring
-
- 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
- H04Q2011/0084—Quality of service aspects
Abstract
The invention discloses an application mapping-assisted method for setting power gradient of an on-chip optical interconnection light source, which comprises the following steps: acquiring an application to be mapped and a corresponding network topology; establishing an optical signal-to-noise ratio model according to a network topology, a selected optical router and a routing algorithm, and obtaining the maximum insertion loss under the global worst condition; obtaining an optimal mapping scheme of application and network topology by using an optical signal-to-noise ratio model and an intelligent algorithm; obtaining minimum insertion loss by using an optimal mapping scheme, and dividing a plurality of intervals between minimum light source power and maximum light source power by adopting a non-uniform distribution mode to form a plurality of light source power gradient levels; and setting the light source power of each communication pair source node according to the light source power gradient level. According to the invention, the IP core is automatically mapped to the ONoC node to optimize the OSNR under the worst condition, and the gradient setting is carried out on the light source power on the basis, so that the ONoC energy consumption is optimized, and the on-chip communication power efficiency and reliability are improved.
Description
Technical Field
The invention belongs to the technical field of on-chip optical networks, and particularly relates to a power gradient setting method of an on-chip optical interconnection light source with mapping assistance.
Background
With the development of multi-core systems, conventional electrical networks on chip have encountered serious performance bottlenecks in terms of bandwidth, latency, and the like. The advent of on-Chip Optical networks-on-Chip (ONoC) provides a promising solution for on-Chip communication, but reliability challenges (e.g., insertion loss, crosstalk) that it suffers limit its scalability, and thus optimization of Optical Signal-to-Noise Ratio (OSNR) in on-Chip Optical interconnects is necessary. Meanwhile, the energy efficiency of the ONoC has great dependence on the power of the light source, the static power consumed by the light source accounts for a large proportion of the total power, and excessive power causes extra waste for inter-task communication, so that reasonable setting of the power of the light source is also necessary, and the light source can be utilized to the maximum extent while ensuring normal communication of optical interconnection on the chip.
Some existing work has been to reduce the power consumed by the light sources by adjusting the network bandwidth, mainly by using temporal and spatial variations in the traffic in the network, in order to optimize the network power consumption. The existing work mainly has the following defects: when communicating in an on-chip optical interconnect, even under low network load conditions, these methods still need to provide large optical source power for worst-case insertion loss to compensate for the insertion loss, if the insertion loss encountered from the source node to the destination node is small.
Disclosure of Invention
In order to solve the above problems in the prior art, the present invention provides a power gradient setting method for an on-chip optical interconnection light source with mapping assistance. The technical problem to be solved by the invention is realized by the following technical scheme:
the invention provides an application mapping-assisted method for setting power gradient of an on-chip optical interconnection light source, which comprises the following steps:
acquiring an application to be mapped and a corresponding network topology;
establishing an optical signal-to-noise ratio model according to a network topology, a selected optical router and a routing algorithm, and obtaining the maximum insertion loss under the global worst condition;
obtaining an optimal mapping scheme of application and network topology by utilizing the optical signal-to-noise ratio model and an intelligent algorithm;
obtaining the minimum insertion loss by using the optimal mapping scheme, and dividing a plurality of intervals between the minimum light source power and the maximum light source power by adopting a non-uniform distribution mode to form a plurality of light source power gradient levels;
and setting the light source power of each communication pair source node according to the light source power gradient grade.
In an embodiment of the present invention, establishing an optical signal-to-noise ratio model according to a network topology and a selected optical router and routing algorithm, and obtaining a maximum insertion loss under a global worst case includes:
obtaining insertion loss and crosstalk generated when any pair of nodes in N nodes in the network topology communicates according to the optical router and a routing algorithm to form an OSNR matrix W, wherein the size of the matrix is NxN, and an element W (i, j) in the OSNR matrix W represents a node tiTo node tjOptical signal to noise ratio during communication
The maximum insertion loss under the global worst case is obtained, namely the maximum insertion loss generated when all nodes in the network topology communicate two by two.
In one embodiment of the invention, the intelligent algorithm is a resilient network algorithm.
In an embodiment of the present invention, obtaining an optimal mapping scheme of an application and a network topology by using the osnr model and an intelligent algorithm includes:
constructing an elastic network and training the elastic network by using a row vector in an OSNR matrix W corresponding to each node to obtain the trained elastic network;
and obtaining an optimal mapping scheme of the application and the network topology by utilizing the trained elastic network and the application communication relation matrix R.
In an embodiment of the present invention, constructing an elastic network and training the elastic network by using a row vector in an OSNR matrix W corresponding to each node, to obtain a trained elastic network, includes:
constructing an elastic network, and training the elastic network by using a row vector in an OSNR matrix W corresponding to each node to obtain the trained elastic network, wherein an energy function of the elastic network is as follows:
wherein α, β and K are scale parameters, xiFor the row vector, y, corresponding to the ith nodejIs the weight vector of the jth neuron, yj +1 is the weight vector of the jth +1 neuron, N is the number of nodes, h is the number of neurons,
the weight vector change updating of the neuron adopts a gradient descent method, and for the jth neuron, the variable quantity of the weight vector is as follows:
the update formula of the weight vector is as follows:
yj=yj+Δyj,
wherein, ω isijThe proximity of the node and the neuron is represented, and when K =0, the weight of the neuron is not changed any more, and the training is ended.
In an embodiment of the present invention, obtaining an optimal mapping scheme of an application and a network topology by using a trained elastic network and an application communication relation matrix R includes:
after training is finished, keeping the weight vector of the neuron in the elastic network unchanged, and acquiring an application communication relation matrix R;
in the application communication relation matrix R, euclidean distances between a row vector corresponding to each IP core and all neurons are calculated, the neuron which is the minimum Euclidean distance between each IP core and the neuron is obtained, neuron index numbers corresponding to the IP cores and the nodes are respectively sorted from small to large, one-to-one mapping is completed, and an optimal mapping scheme of application and network topology is obtained.
In an embodiment of the present invention, obtaining a minimum insertion loss by using the optimal mapping scheme, and dividing a plurality of intervals between the minimum light source power and the maximum light source power by using a non-uniform distribution manner to form a plurality of light source power gradient levels, includes:
and obtaining the insertion loss corresponding to each communication pair in the optimal mapping scheme by utilizing the optical signal-to-noise ratio model, and selecting the minimum insertion loss as the minimum insertion loss. Acquiring minimum light source power and maximum light source power by using the minimum insertion loss and the maximum insertion loss;
and dividing a plurality of intervals between the minimum light source power and the maximum light source power in a non-uniform distribution mode to form a plurality of light source power gradient levels.
In an embodiment of the present invention, dividing a plurality of intervals between the minimum light source power and the maximum light source power in a non-uniform distribution manner to form a plurality of light source power gradient levels includes:
at minimum light source powerAnd maximum light source powerObtaining the power of the midpoint light sourceAt the minimum light source powerPower of light source at midpointR/3-r/2 light source power gradient grades are uniformly arranged between the two points, and the light source power is at the midpointAnd the maximum light source powerAnd 0-2 light source power gradient levels are uniformly arranged between the two, wherein r represents the number of communication pairs.
In one embodiment of the present invention, setting the light source power of each communication pair source node according to the light source power gradient level comprises:
for each communication pair in the optimal mapping scheme, according toCalculating the light source power required by the source node, wherein S is the sensitivity of the photoelectric detector of the destination node,represents a node tsTo node tdInsertion loss in communication; and checking the interval of the light source power in the light source power gradient grade, and setting the maximum value in the current interval as the final light source power corresponding to the current source node.
The invention further provides an electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor realizes the steps of the method for setting the power gradient of the on-chip optical interconnection light source with mapping assistance according to any one of the embodiments when calling the computer program in the memory.
Compared with the prior art, the invention has the beneficial effects that:
1. the invention provides an on-chip optical interconnection light source power gradient setting method with mapping assistance, which optimizes the optical signal to noise ratio (OSNR) under the worst condition by automatically mapping an IP core to an ONoC node and performs gradient setting on the light source power on the basis, thereby achieving the purposes of optimizing ONoC energy consumption and improving on-chip communication power efficiency and reliability.
2. The invention optimizes the OSNR through mapping, and improves the reliability and the expandability of on-chip optical interconnection.
3. The invention sets reasonable light source power for each communication pair after optimizing OSNR, reduces the static power consumed by the light source and improves the power utilization rate of on-chip optical interconnection.
The present invention will be described in further detail with reference to the accompanying drawings and examples.
Drawings
Fig. 1 is a flowchart of a method for setting a power gradient of an on-chip optical interconnection light source by using mapping assistance according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a framework for setting a power gradient of an on-chip optical interconnection light source using mapping assistance according to an embodiment of the present invention;
fig. 3 is a diagram of on-chip communication path loss provided by an embodiment of the present invention;
FIG. 4 is a schematic diagram of an elastic network composed of neurons according to an embodiment of the present invention;
FIG. 5 is a schematic diagram of a power gradient of a light source according to an embodiment of the present invention;
fig. 6 is a schematic structural diagram of a Cygnus router according to an embodiment of the present invention;
fig. 7 is a mapping scheme for optimizing OSNR for MWD;
fig. 8 is a schematic diagram of MWD applied source node light source power gradient level setting obtained by the method of the embodiment of the invention.
Detailed Description
To further illustrate the technical means and effects of the present invention adopted to achieve the predetermined object, a power gradient setting method for an on-chip optical interconnection light source using mapping assistance according to the present invention is described in detail below with reference to the accompanying drawings and the detailed description.
The foregoing and other technical contents, features and effects of the present invention will be more clearly understood from the following detailed description of the embodiments taken in conjunction with the accompanying drawings. The technical means and effects of the present invention adopted to achieve the predetermined purpose can be more deeply and specifically understood through the description of the specific embodiments, however, the attached drawings are provided for reference and description only and are not used for limiting the technical scheme of the present invention.
It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Furthermore, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that an article or device that comprises a list of elements does not include only those elements but may include other elements not expressly listed. Without further limitation, an element defined by the phrase "comprising one of 8230, and" comprising 8230does not exclude the presence of additional like elements in an article or device comprising the element.
The embodiment provides a mapping-assisted method for setting power gradient of an on-chip optical interconnection light source, referring to fig. 1 and 2, the method includes:
s1: and acquiring the application to be mapped and the corresponding network topology.
In the on-chip optical network, the applied IP cores need to be mapped to the nodes of the network topology one-to-one, the mapped nodes of the network topology communicate according to the communication relationship between the IP cores, and different nodes have different insertion loss and crosstalk when communicating. The position of the IP core in the network is reasonably determined through mapping, the OSNR can be optimized, and the power saving of the light source can be further realized on the basis.
S2: and establishing an OSNR model according to the network topology, the selected optical router and a routing algorithm, and obtaining the maximum insertion loss under the global worst condition.
For the mapping problem, the application is a directed graph G1= (C, E), representing communication relationship between IP cores, where C represents a set of IP cores, each vertex CiE represents an IP core, E represents a set of directed edges, and each directed edge EijE denotes IP core ciTo cjCommunication of (e)ij≠eji(ii) a Likewise, an undirected graph G2= (T, L) may represent network topology, where T represents a set of nodes, each vertex Tje.T represents a node, L represents the link set connecting the nodes, LijE L represents a connection node tiAnd tjPhysical link of lij=lji. The mapping function f can be expressed as formula (1), i.e., one IP core is mapped to one node, and one node is mapped by one IP core.
In this embodiment, a reasonable mapping scheme is solved with the purpose of optimizing the OSNR under the worst case condition, and the light source power is set for the source node corresponding to each communication pair in the network topology based on the reasonable mapping scheme. It should be noted that, if the application is an IP core c1To IP core c2There is a communication relationship, after mapping, IP core c1Mapping to a network topology node t3IP core c2Mapping to a network topology node t4Then the mapped corresponding communication is node t3To node t4This is called a communication pair, i.e. a communication pair refers to two network topology nodes having a communication relationship.
Specifically, the insertion loss and crosstalk generated when any pair of nodes in N nodes in the network topology communicates can be obtained according to the selected optical router and routing algorithm, and an OSNR matrix W with the size of N × N is formed, wherein W (i, j) is a node tiTo node tjOptical signal to noise ratio during communication
In ONoC, the communication process between nodes is as follows: wavelength lambda provided by a light source in a source node1,λ2,...,λnGuided to the chip in the optical fiber and coupled to the waveguide, the resonance wavelengths of the n micro ring resonators (MR) respectively correspond to the n wavelengths, and data can be modulated to the corresponding wavesLong optical signals. The modulated optical signals are transmitted along the waveguide to reach the destination node, and the n MRs are designed to resonate at the same wavelength as the optical signals so as to filter the optical signals. The filtered optical signal is subjected to photoelectric conversion through a photoelectric detector, amplified through a trans-impedance amplifier, and finally the data modulated on the wavelength is extracted to complete communication. Some losses are generated in the whole communication process, including: coupling loss L of light sourcecouplerLoss at Through end (pass end) in MRLoss at Drop end (download end)Modulator loss LMDetector loss LPDWaveguide cross loss LCAnd waveguide bending loss LbAnd the like. As shown in fig. 3, these losses together constitute insertion loss, which causes the light source power to decrease continuously during transmission.
Communication source node tsWith destination node tdInsertion loss therebetweenIs as in equation (2), where θ, m, p, q are the numbers of the through port, the drop port, the waveguide crossing and the waveguide bending respectively passing through the optical router during the communication.
Destination node t of communicationdReceived signal power ofIt is calculated as formula (3), where PinIs a source node tsOptical signal input power of (1):
source node tsAnd destination node tdThe crosstalk calculation for a communication subject to other communications is shown in equation (4), where,the noise power generated for the current communication for the ith communication pair, and r is the number of communication pairs.
Source node tsAnd destination node tdIs/are as followsAnd (5) calculating the OSNR generated when any pair of nodes in the N nodes in the network topology communicate to form an OSNR matrix W.
In a specific application, the worst-case OSNR is calculated as formula (6), i.e. after mapping the specific application to the ONoC, the minimum OSNR of the nodes having communication relationship in the ONoC when they communicate with each other is also the target of mapping optimization.
Wherein, cs、cdIs an IP core, C is an IP core set, esdE denotes IP core csTo cdE denotes a set of directed edges, f (c)s)、f(cd) Respectively represent csAnd cdThe network topology nodes mapped by the formula (1), T represents a node set,represents node f (c)s) To f (c)d) OSNR when communicating.
It should be noted that, in the practical situation of this embodiment, because the mapping needs to be solved, the location where the IP core having the communication relationship is mapped to the node cannot be determined, and the mapping scheme needs to be evaluated according to the mapped communication OSNR. The OSNR matrix W of the entire ONoC is obtained in advance and the maximum insertion loss in the global worst case is obtained.
Specifically, the global worst case maximum insertion loss is calculated as equation (7), i.e., the maximum insertion loss that all nodes in ONoC generate when communicating two-by-two without any mapping applied, where,represents a node tsTo node tdInsertion loss in communication.
S3: and obtaining an optimal mapping scheme of the application and the network topology by utilizing the OSNR model and the intelligent algorithm.
It should be noted that the intelligent algorithm may be any algorithm capable of solving the optimization problem quickly, such as a Simulated Annealing (SA) algorithm, a Particle Swarm Optimization (PSO) algorithm, an Artificial Bee Colony (ABC) algorithm, an artificial fish swarm optimization (AFSA) algorithm, a Continuous Hopfield Neural Network (CHNN) algorithm, and the like. The intelligent algorithm selected by the present embodiment is a resilient network algorithm.
Referring to fig. 4, fig. 4 is a schematic diagram of an elastic network composed of neurons according to an embodiment of the present invention. The Elastic Network (EN) is a one-dimensional annular network composed of neurons, the number of the neurons is three times that of IP cores, the neurons are connected end to end, and the elastic network has good geometric characteristics. The elastic network represents the target in the form of an energy function, the minimum of which corresponds to the optimal solution of the problem. Its idea of optimizing OSNR: in the OSNR matrix W, an original elastic network is subjected to iterative training by using row vectors in the matrix W corresponding to each node, one row vector is used each time, the weight vectors of neurons in the elastic network are updated according to the row vectors, the row vectors are sequentially and continuously used for updating the weight vectors of the neurons, the neurons are close to the nodes, and finally each node corresponds to one neuron in the elastic network. The number of neurons is 3 times or more the number of nodes.
Specifically, when elastic network training is performed by using a row vector in the OSNR matrix W corresponding to each node, an energy function is defined as:
where α, β and K are scale parameters, where α, β are constants, K is equivalent to the temperature in the simulated annealing algorithm, decreases with training, and xiIs the row vector, y, corresponding to the ith nodejIs the weight vector of the jth neuron, yj+1Is the weight vector of the j +1 th neuron, N is the number of nodes, and h is the number of neurons. When the energy function H is minimal, each node corresponds to a neuron.
The weight vector change of the neuron is updated by gradient descent method, and for the jth neuron, the change amount of the weight vector is delta yjAs in equation (9). Wherein ω isijLike equation (10), the weight vector is updated like equation (11).
yj=yj+Δyj j∈{1,...,h} (11)
Wherein, ω isijRepresenting the proximity of nodes and neurons. When K =0, if neuron j finally converges to node i, ω can be obtainedij=1,xi-yj=0, when Δ y in equation (9)jIs 0(ii) a If neuron j does not converge to node i, ωij=0, Δ y in equation (9)jAlso 0. That is, when K =0, the weight of the neuron is not changed any more, and the training is ended.
After the training is finished, the elastic network is kept unchanged, namely the weight vector of the neuron is unchanged, and an application communication relation matrix R is obtained. As described above, the application is a directed graph G1= (C, E) in directed graph G1In (e), if there is a directed edge eijThen R (i, j) is 1, otherwise 0, thereby forming the application communication relationship matrix R. And in the application communication relation matrix R, calculating Euclidean distances between the row vector corresponding to each IP core and all the neurons, wherein each IP core is required to be the neuron with the minimum Euclidean distance to the neuron. And sorting the index numbers of the neurons corresponding to the IP cores and the nodes from small to large respectively to complete one-to-one mapping so as to obtain an optimal mapping scheme of application and network topology.
S4: and obtaining the minimum insertion loss by using the optimal mapping scheme, and dividing a plurality of intervals between the minimum light source power and the maximum light source power by adopting a non-uniform distribution mode to form a plurality of light source power gradient levels.
The power of the light source depends on many technical parameters, such as insertion loss and photodetector sensitivity. In on-chip communication, in order to correctly convert an optical signal into an electrical signal at a destination node, the received optical signal needs to have a minimum power higher than the sensitivity of the photodetector, i.e., the optical source power should cancel the insertion loss and ensure the minimum power of the destination node. In the case of single wavelength communication for each communication pair, the power requirement of the light source satisfies equation (12):
wherein the content of the first and second substances,for communication to source node light source power, S is the photodetector sensitivity of the destination node, and is generally set to-20 dBm.
The conventional light source power allocation to the source node per communication is an even allocation, i.e. according to the global worst case maximum insertion loss ILmaxThe light source power is set, and the light source power of each communication to the source node at this time is equation (13).
To avoid power waste, in the OSNRworstAfter optimization, the corresponding insertion loss is obtainedThe source node is then allocated as much light source power as possible for each communication pair based on equation (14)
It is impractical to set the light source power required for each communication pair exactly, and in order to save power as much as possible and achieve reasonable distribution, the present embodiment uses a light source power gradient setting method to set the light source power for the source node of each communication pair. Specifically, S4 of the present embodiment includes:
s4.1: and obtaining the insertion loss corresponding to each communication pair in the optimized one-to-one mapping scheme by using an OSNR model between the communication source node and the destination node, and selecting the minimum insertion loss as the minimum insertion loss. And acquiring minimum light source power and maximum light source power by using the minimum insertion loss and the maximum insertion loss.
Using the above insertion loss model, we can obtain the global worst case maximum insertion loss IL in the network topologymaxGenerally, the source node of each communication pair is set with the optical source power based on this insertion lossResulting in significant power waste.
In the optimized one-to-one optimal mapping scheme, the insertion loss corresponding to each communication pair after mapping can be obtained, and the light source power required by the communication pair corresponding to the minimum insertion loss is solvedThus, the light source power required for all communication pairs after mapping is limitedAndin the meantime.
S4.2: and dividing a plurality of intervals between the minimum light source power and the maximum light source power in a non-uniform distribution mode to form a plurality of light source power gradient levels.
In particular, at minimum light source powerTo the maximum light source powerAnd setting a plurality of light source power gradient levels in the span interval, and setting the light source power of the source node of each communication pair. In fact, in the mapping scheme optimized for OSNR, the light source power required by the source node of few communication pairs is achieved empirically and through extensive calculationsRatio of light source power required by source nodes of most communication pairsSlightly larger. Based on this, the setting of the gradient grade of the light source power adopts a non-uniform distribution mode, in particular, the minimum light source powerAnd maximum light source powerPower of light source with a midpoint At minimum light source powerAnd mid-point light source powerThe power gradient of small intervals is uniformly distributed, and almost all communication pairs require light source power in the interval; at mid-point light source powerTo the maximum light source powerThe power gradient is uniformly distributed in a large interval, and the power gradient level of the light source set in this way can meet the communication requirement and simultaneously avoid the waste of power as much as possible, and a schematic diagram is shown in fig. 5.
Further, it is necessary to make the number of levels of the light source power gradient clear, and the greater the number of levels, the denser the distribution, the finer-grained setting of the light source power of the source node can be made, and the more power can be saved, but this will lead to more complicated laser driver settings. In order to avoid excessive light source power gradient level setting and save power as much as possible, experimental analysis shows that when r communication pairs exist, the minimum light source power is usedAnd power of medium point light sourceThe r/3-r/2 gradient grades are arranged uniformly in the interval, and the power of the light source at the midpoint is properAnd maximum light source powerThe gradient levels of 0-2 are set uniformly, and the communication pairs with the light source power in the interval required by the source node are fewer.
S5: and setting the light source power of each communication pair source node according to the light source power gradient grade.
After the light source power gradient level is set, the level value is stored in a lookup table, and for each communication pair, the communication pair is determined according to the level valueAnd calculating the light source power required by the source node, then comparing the calculated light source power with the grade value in the gradient grade of the light source power, enabling the light source power to fall in a certain section of the gradient grade, and setting the maximum value in the current section as the final light source power corresponding to the current source node. In particular, assuming that the light source power required by a certain communication pair source node is b, for adjacent light source power gradient levels u and v, there is u<b<v, the light source power of the communication to the source node is set to v. For example, assuming that the light source power gradient levels are 1, 2, and 3, the light source power required for a certain communication pair is 2.6, and is set to 3 between the intervals 2 and 3. The setting of a larger grade value can overcome the influence caused by temperature drift and process drift on the chip in actual communication, and the requirement on the reliability of on-chip communication is ensured as much as possible.
Since the on-chip optical device is highly sensitive to temperature drift and process drift, this may result in a high Bit Error Rate (BER), which affects communication. In order to improve communication efficiency and stabilize network performance, the light source power of a source node can be controlled to be dynamically adjusted in a gradient level so as to adapt to the change of an optical device. For a communication pair, when the light source power of the current source node is not enough to complete normal communication, the normal completion of the communication is ensured by increasing a light source power gradient level. When the light source power of the current source node is too large, power is saved by reducing a light source power gradient grade, and when the light source power gradient grade is reduced to 0, the light source is turned off. The magnitude of the source node light source power can be observed and fed back through the BER of the destination node. The dynamic adjustment of the source node light source power is performed by accessing a look-up table.
The embodiment provides an application mapping-assisted on-chip optical interconnection light source power gradient setting method, which is characterized in that an IP core is automatically mapped to an ONoC node to optimize the OSNR under the worst condition, and the power of a light source is subjected to gradient setting on the basis, so that the purposes of optimizing the ONoC energy consumption and improving the on-chip communication power efficiency and reliability are achieved. The embodiment optimizes the OSNR through mapping, and improves the reliability and the expandability of on-chip optical interconnection. After the OSNR is optimized, reasonable light source power is set for each communication pair, the static power consumed by the light source is reduced, and the power utilization rate of on-chip optical interconnection is improved.
Example two
On the basis of the first embodiment, the present embodiment describes in detail the application mapping assisted on-chip optical interconnection light source power gradient setting method proposed by the embodiment of the present invention based on an MWD application and a 3 × 4 mesh network topology.
Specifically, a Cygnus router and an XY routing algorithm are selected by utilizing a MWD application and a 3 x 4 mesh network topology, wherein the structure of the Cygnus router is shown in FIG. 6, loss parameters are shown in Table 1, and the number of waveguide bends, waveguide intersections and the like can be obtained from the structure.
Table 1 loss parameters for routers
Based on the above information, the OSNR model (equation (5)) can be obtained, and then the mapping scheme is obtained by using the elastic network EN intelligent algorithm, as shown in fig. 7.
Obtaining minimum insertion loss and maximum insertion loss by using insertion loss model, and obtaining by using minimum insertion loss and maximum insertion lossThe set light source power gradient levels are shown in fig. 8.
The light source power is set to the source node of each communication pair based on the light source power gradient level shown in fig. 8, as shown in table 2. The total power consumed by all communication pairs was 0.265mW. When all communication is set to the source node light source powerThe total power consumed was 0.372W. Therefore, the light source power gradient setting mode and the light source power of the present embodiment are both set toCompared with 28.8% power saving.
Table 2 communication versus source node light source power gradient setup
Communication pair in mapped network | Gradient value of light source power | Communication pair in mapped network | Gradient value of light source power |
t4->t8 | 0.0207mW | t7->t11 | 0.0220mW |
t4->t2 | 0.0232mW | t11->t9 | 0.0232mW |
t8->t12 | 0.0220mW | t1->t5 | 0.0207mW |
t8->t7 | 0.0220mW | t5->t9 | 0.0220mW |
t3->t2 | 0.0220mW | t5->t6 | 0.0232mW |
t2->t1 | 0.0220mW | t6->t10 | 0.0220mW |
Yet another embodiment of the present invention provides a storage medium having stored therein a computer program for executing the steps of the method for applying mapping assisted on-chip optical interconnect light source power gradient setting described in the above embodiments. Yet another aspect of the present invention provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor, when calling the computer program in the memory, implements the steps of the method for setting power gradient of an on-chip optical interconnection light source with mapping assistance as described in the above embodiment. Specifically, the integrated module implemented in the form of a software functional module may be stored in a computer readable storage medium. The software functional module is stored in a storage medium and includes several instructions to enable an electronic device (which may be a personal computer, a server, or a network device) or a processor (processor) to execute some steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: various media capable of storing program codes, such as a usb disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk.
The foregoing is a more detailed description of the invention in connection with specific preferred embodiments and it is not intended that the invention be limited to these specific details. For those skilled in the art to which the invention pertains, several simple deductions or substitutions can be made without departing from the spirit of the invention, and all shall be considered as belonging to the protection scope of the invention.
Claims (10)
1. A mapping-assisted on-chip optical interconnection light source power gradient setting method is characterized by comprising the following steps:
acquiring an application to be mapped and a corresponding network topology;
establishing an optical signal-to-noise ratio model according to the network topology, the selected optical router and a routing algorithm, and obtaining the maximum insertion loss under the global worst condition;
obtaining an optimal mapping scheme of application and network topology by using the optical signal-to-noise ratio model and an intelligent algorithm;
obtaining the minimum insertion loss by using the optimal mapping scheme, and dividing a plurality of intervals between the minimum light source power and the maximum light source power by adopting a non-uniform distribution mode to form a plurality of light source power gradient levels;
and setting the light source power of each communication pair source node according to the light source power gradient grade.
2. The method for setting the power gradient of the optical interconnection light source on the chip by applying mapping assistance according to claim 1, wherein an optical signal-to-noise ratio model is established according to a network topology, a selected optical router and a routing algorithm, and a global worst-case maximum insertion loss is obtained, and the method comprises the following steps:
obtaining insertion loss and crosstalk generated when any pair of nodes in N nodes in the network topology communicates according to the optical router and a routing algorithm to form an OSNR matrix W, wherein the size of the matrix is NxN, and an element W (i, j) in the OSNR matrix W represents a node tiTo node tjOSNR (optical signal to noise ratio) during communicationti→tj;
The maximum insertion loss under the global worst case is obtained, namely the maximum insertion loss generated when all nodes in the network topology communicate with each other.
3. The method for setting power gradient of optical interconnection light source on chip by applying mapping assistance as claimed in claim 1, wherein the intelligent algorithm is an elastic network algorithm.
4. The method for setting the power gradient of the on-chip optical interconnection light source assisted by application mapping according to claim 1, wherein the obtaining of the optimal mapping scheme of the application and the network topology by using the osnr model and the intelligent algorithm comprises:
constructing an elastic network and training the elastic network by using a row vector in an OSNR matrix W corresponding to each node to obtain the trained elastic network;
and obtaining an optimal mapping scheme of the application and the network topology by utilizing the trained elastic network and the application communication relation matrix R.
5. The method for setting the power gradient of the on-chip optical interconnection light source with the mapping assistance as claimed in claim 4, wherein constructing an elastic network and training the elastic network by using a row vector in an OSNR matrix W corresponding to each node to obtain a trained elastic network comprises:
constructing an elastic network, and training the elastic network by using a row vector in an OSNR matrix W corresponding to each node to obtain the trained elastic network, wherein an energy function of the elastic network is as follows:
where α, β and K are scale parameters, xiFor the row vector corresponding to the i-th node, yjIs the weight vector of the jth neuron, yj+1Is the weight vector of the j +1 th neuron, N is the node number, h is the neuron number,
the weight vector change updating of the neuron adopts a gradient descent method, and for the jth neuron, the variable quantity of the weight vector is as follows:
the update formula of the weight vector is:
yj=yj+Δyj,
wherein, ω isijThe proximity of the node and the neuron is represented, and when K =0, the weight of the neuron is not changed any more, and the training is ended.
6. The method for setting the power gradient of the optical interconnection light source on the chip assisted by the application mapping according to claim 4, wherein the step of obtaining the optimal mapping scheme of the application and the network topology by using the trained elastic network and the application communication relation matrix R comprises the following steps:
after training is finished, keeping the weight vector of the neuron in the elastic network unchanged, and acquiring an application communication relation matrix R;
in the application communication relation matrix R, euclidean distances between a row vector corresponding to each IP core and all neurons are calculated, the neuron which is the minimum Euclidean distance between each IP core and the neuron is obtained, neuron index numbers corresponding to the IP cores and the nodes are respectively sorted from small to large, one-to-one mapping is completed, and an optimal mapping scheme of application and network topology is obtained.
7. The method as claimed in claim 4, wherein the step of obtaining the minimum insertion loss by using the optimal mapping scheme and dividing a plurality of intervals between the minimum light source power and the maximum light source power in a non-uniform distribution manner to form a plurality of light source power gradient levels comprises:
and obtaining the insertion loss corresponding to each communication pair in the optimal mapping scheme by utilizing the optical signal-to-noise ratio model, and selecting the minimum insertion loss as the minimum insertion loss. Acquiring minimum light source power and maximum light source power by using the minimum insertion loss and the maximum insertion loss;
and dividing a plurality of intervals between the minimum light source power and the maximum light source power in a non-uniform distribution mode to form a plurality of light source power gradient levels.
8. The method for setting power gradient of optical interconnection light source on chip with mapping assistance as claimed in claim 7, wherein dividing a plurality of intervals between the minimum light source power and the maximum light source power in a non-uniform distribution manner to form a plurality of light source power gradient levels comprises:
at minimum light source powerAnd maximum light source powerObtain the power of the midpoint light sourceAt the minimum light source powerPower of light source at midpointR/3-r/2 light source power gradient grades are uniformly arranged between the two points, and the light source power is at the midpointAnd the maximum light source powerAnd 0-2 light source power gradient levels are uniformly arranged between the two, wherein r represents the number of communication pairs.
9. The application-mapping-assisted on-chip optical interconnection optical source power gradient setting method according to claim 7, wherein setting the optical source power of each communication pair source node according to the optical source power gradient level comprises:
for each communication pair in the optimal mapping scheme, according toCalculating the light source power required by the source node, wherein S is the sensitivity of the photoelectric detector of the destination node,represents a node tsTo node tdInsertion loss in communication; and checking the interval of the light source power in the light source power gradient grade, and setting the maximum value in the current interval as the final light source power corresponding to the current source node.
10. An electronic device, comprising a memory in which a computer program is stored and a processor, which when invoked on the computer program in the memory implements the steps of the application mapping assisted on-chip optical interconnect light source power gradient setting method according to any of claims 1 to 9.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210909758.1A CN115276820B (en) | 2022-07-29 | 2022-07-29 | On-chip optical interconnection light source power gradient setting method using mapping assistance |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210909758.1A CN115276820B (en) | 2022-07-29 | 2022-07-29 | On-chip optical interconnection light source power gradient setting method using mapping assistance |
Publications (2)
Publication Number | Publication Date |
---|---|
CN115276820A true CN115276820A (en) | 2022-11-01 |
CN115276820B CN115276820B (en) | 2023-09-01 |
Family
ID=83746718
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210909758.1A Active CN115276820B (en) | 2022-07-29 | 2022-07-29 | On-chip optical interconnection light source power gradient setting method using mapping assistance |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115276820B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117411561A (en) * | 2023-12-05 | 2024-01-16 | 西安电子科技大学 | Optical network-on-chip communication method based on approximate calculation and optical network-on-chip |
Citations (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101548695B1 (en) * | 2014-11-03 | 2015-09-01 | 성균관대학교산학협력단 | Apparatus and method for topology design of hybrid optical networks on chip |
CN105959239A (en) * | 2016-04-22 | 2016-09-21 | 西安电子科技大学 | Communication method of 3D optical on-chip network |
US20180076921A1 (en) * | 2016-09-12 | 2018-03-15 | Eric Koopferstock | Credit based approach to calculating optical paths |
CN108398982A (en) * | 2018-01-30 | 2018-08-14 | 上海电力学院 | A kind of maximum power tracking method of photovoltaic array under local shadow |
CN111049185A (en) * | 2019-12-29 | 2020-04-21 | 国网辽宁省电力有限公司电力科学研究院 | Island division method for power distribution network comprising distributed power supply |
CN111752891A (en) * | 2020-06-05 | 2020-10-09 | 西安电子科技大学 | IP core mapping method for optical network on chip |
KR102346028B1 (en) * | 2020-10-22 | 2022-01-03 | 성균관대학교산학협력단 | Worst-Case OSNR(optical signal-to-noise ratio) Analysis method and appratus for Optical Network-on-Chip |
WO2022060908A1 (en) * | 2020-09-15 | 2022-03-24 | Arizona Board Of Regents On Behalf Of The University Of Arizona | Frequency multiplexed photonic neural networks |
CN114499679A (en) * | 2022-01-25 | 2022-05-13 | 上海交通大学 | Multi-band Raman amplifier design method and system |
CN114598946A (en) * | 2022-01-24 | 2022-06-07 | 西安电子科技大学 | Fuzzy logic-based on-chip optical network self-adaptive routing planning method |
CN114637720A (en) * | 2021-12-13 | 2022-06-17 | 西安电子科技大学 | On-chip optical network mapping method based on growth type annular SOM neural network |
-
2022
- 2022-07-29 CN CN202210909758.1A patent/CN115276820B/en active Active
Patent Citations (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
KR101548695B1 (en) * | 2014-11-03 | 2015-09-01 | 성균관대학교산학협력단 | Apparatus and method for topology design of hybrid optical networks on chip |
CN105959239A (en) * | 2016-04-22 | 2016-09-21 | 西安电子科技大学 | Communication method of 3D optical on-chip network |
US20180076921A1 (en) * | 2016-09-12 | 2018-03-15 | Eric Koopferstock | Credit based approach to calculating optical paths |
CN108398982A (en) * | 2018-01-30 | 2018-08-14 | 上海电力学院 | A kind of maximum power tracking method of photovoltaic array under local shadow |
CN111049185A (en) * | 2019-12-29 | 2020-04-21 | 国网辽宁省电力有限公司电力科学研究院 | Island division method for power distribution network comprising distributed power supply |
CN111752891A (en) * | 2020-06-05 | 2020-10-09 | 西安电子科技大学 | IP core mapping method for optical network on chip |
WO2022060908A1 (en) * | 2020-09-15 | 2022-03-24 | Arizona Board Of Regents On Behalf Of The University Of Arizona | Frequency multiplexed photonic neural networks |
KR102346028B1 (en) * | 2020-10-22 | 2022-01-03 | 성균관대학교산학협력단 | Worst-Case OSNR(optical signal-to-noise ratio) Analysis method and appratus for Optical Network-on-Chip |
CN114637720A (en) * | 2021-12-13 | 2022-06-17 | 西安电子科技大学 | On-chip optical network mapping method based on growth type annular SOM neural network |
CN114598946A (en) * | 2022-01-24 | 2022-06-07 | 西安电子科技大学 | Fuzzy logic-based on-chip optical network self-adaptive routing planning method |
CN114499679A (en) * | 2022-01-25 | 2022-05-13 | 上海交通大学 | Multi-band Raman amplifier design method and system |
Non-Patent Citations (5)
Title |
---|
KE CHEN; HUAXI GU; HUI LI; YAWEN CHEN; HAIBO ZHANG: "TWC-based Approach for Improving Communication Reliability in Optical Network-on-chip", 2013 IEEE INTERNATIONAL CONFERENCE OF IEEE REGION 10 (TENCON 2013) * |
刘凌云,王琨,邓植,张碧霞,顾华玺: "片上网络映射优化问题研究与进展", 计算机应用研究, pages 1929 - 1934 * |
李慧,顾华玺: "多核之间光互连技术的研究", 中国集成电路, pages 50 - 55 * |
李贞妮;李晶皎;金硕巍;: "可靠电压频率岛感知的异步片上网络能耗优化", 信息与控制, no. 06, pages 35 - 40 * |
王爽;刘克;: "马赫-曾德尔结合矩形环谐振腔的超紧凑光滤波器", 光学学报, no. 05, pages 206 - 215 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN117411561A (en) * | 2023-12-05 | 2024-01-16 | 西安电子科技大学 | Optical network-on-chip communication method based on approximate calculation and optical network-on-chip |
Also Published As
Publication number | Publication date |
---|---|
CN115276820B (en) | 2023-09-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
AU2009298151B2 (en) | Systems and methods for automatic creation of agent-based systems | |
CN110225525B (en) | Cognitive radio network-based spectrum sharing method, device and equipment | |
CN115276820B (en) | On-chip optical interconnection light source power gradient setting method using mapping assistance | |
EP3296781B1 (en) | On-chip optical interconnection structure and network | |
CN108769843B (en) | Power optical fiber home-in EPON (Ethernet passive optical network) optimization method and system | |
Wu et al. | Interpretable graph convolutional network for multi-view semi-supervised learning | |
CN106788708A (en) | The OSNR computational methods of OTN networks | |
Anandkumar | Hybrid fuzzy logic and artificial Flora optimization algorithm-based two tier cluster head selection for improving energy efficiency in WSNs | |
CN115037590A (en) | Network virtualization system structure and virtualization method | |
CN111104760B (en) | Electrical performance-oriented reflective surface antenna back frame structure topology optimization method | |
CN114637720B (en) | On-chip optical network mapping method based on growth type annular SOM neural network | |
WO2020168676A1 (en) | Method for constructing network fault handling model, fault handling method and system | |
WO2022083549A1 (en) | Traffic signal conversion method and apparatus, electronic device, and storage medium | |
Sehrawat et al. | NaISEP: Neighborhood aware clustering protocol for WSN assisted IOT network for agricultural application | |
CN113673677B (en) | Method, equipment and medium for realizing nonlinear activation function RELU | |
Chen et al. | Prediction of cloud resources demand based on fuzzy deep neural network | |
Stüdli et al. | On the design of a novel control algorithm and communication structure for discrete-time multi-agent consensus systems | |
WO2020191686A1 (en) | Neural network-based power distribution method and device | |
Pendeza Martinez et al. | Hopfield learning‐based and non‐linear programming methods for resource allocation in OCDMA networks | |
Zhixun et al. | Modeling and Performance Analysis of a Fault-Tolerant 3D Photonic Network-on-Chip Based on Hybrid Photonics–Plasmonics | |
CN112085758B (en) | Edge-end fused terminal context adaptive model segmentation method | |
CN114866459B (en) | Path planning method under multiple constraint conditions | |
Zhang et al. | Exploratioin on Neural Network-Based Wavelength Assignment Method in Silicon Photonic Interconnects on Chip | |
Al-Shamayleh et al. | Performance Evaluation of Optical Transmission Based on Link Estimation by Using Deep Learning Techniques | |
KR20220061835A (en) | Apparatus and method for hardware acceleration |
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 |