Circuit delay prediction method, device, terminal equipment and readable storage medium

By using graph neural network to predict the delay of circuits in IC design, combined with EDA tools and feature vector assignment, the timing uncertainty problem caused by insufficient wiring information is solved, faster and more accurate delay prediction is achieved, and the design process is optimized .

CN113973061BActive Publication Date: 2024-11-15SHENZHEN HONGXIN MICRO NANO TECH CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202111254166.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-10-27
Publication Date
2024-11-15
Estimated Expiration
2041-10-27

AI Technical Summary

Technical Problem

In the layout stage of IC design, due to insufficient wiring information, timing estimation uncertainty is caused. Existing methods such as pessimistic prediction lead to over-design or iterative adjustment failure, increasing design turnover time.

Method used

By obtaining the RC network of the circuit, the graph connection matrix and the feature vector matrix are determined, and the graph neural network is used to predict the delay, combined with EDA tools and feature vector assignment, accurate prediction of the wiring structure is achieved.

Benefits of technology

The circuit delay prediction time is shortened, the prediction accuracy and speed is improved, the defects of over-design and iterative adjustment are avoided, and the design process is optimized.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113973061B_ABST
    Figure CN113973061B_ABST
Patent Text Reader

Abstract

The present application discloses a circuit delay prediction method, apparatus, terminal device and readable storage medium. The method comprises: obtaining the i-th RC network corresponding to the i-th net of the circuit, 1≤i≤I, I is the total number of nets in the circuit; determining the i-th graph connection matrix and the i-th eigenvector matrix based on the i-th RC network; and predicting the delay of the i-th RC network using the i-th graph connection matrix and the i-th eigenvector matrix. The graph connection matrix of the present application includes the relationship between each path in the wiring structure to be predicted, and the eigenvector matrix includes the common attributes of each node in the wiring structure to be predicted, and then based on the graph connection matrix and the eigenvector matrix, accurate prediction of the wiring structure to be predicted can be achieved, and the prediction speed is faster.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of semiconductor technology, and in particular to a circuit delay prediction method, device, terminal equipment and readable storage medium. Background Art

[0002] In the IC design process, static timing analysis is required in the layout phase, that is, in the optimization of the layout phase, timing estimation is required to provide guidance before routing, and to adjust the paths that do not meet the timing closure conditions. However, at this stage, accurate timing estimation is difficult because routing has not yet been performed.

[0003] Currently, there are two strategies to address timing uncertainty caused by insufficient routing information. One approach is to take overly pessimistic pre-routing predictions to ensure that timing violations will not occur after routing. This pessimism leads to over-design, wasting resources, area, and optimization time. Another approach is to iterate back to the cell position when routing cannot achieve the required timing power trade-off. However, multiple iterations do not guarantee the success of the iteration and multiple iterations will greatly increase the design turnaround time. Summary of the invention

[0004] In view of the above problems, the present application proposes a circuit delay prediction method, apparatus, terminal device and readable storage medium to shorten the circuit delay prediction time.

[0005] The present application embodiment provides a circuit delay prediction method, the method comprising:

[0006] Get the i-th RC network corresponding to the i-th net of the circuit, 1≤i≤I, I is the total number of nets in the circuit;

[0007] Determining an i-th graph connection matrix and an i-th eigenvector matrix based on the i-th RC network;

[0008] The delay of the i-th RC network is predicted using the i-th graph connection matrix and the i-th eigenvector matrix.

[0009] The circuit delay prediction method described in the embodiment of the present application, wherein determining the i-th graph connection matrix based on the i-th RC network includes:

[0010] Adding labels to each node in the i-th RC network;

[0011] Determine a signal output node from each of the nodes based on the signal flow direction in the i-th RC network;

[0012] Using a node that receives an output signal of the j-th signal output node as a related signal input node of the j-th signal output node, 1≤j≤J, J is the total number of signal output nodes in the i-th RC network;

[0013] The i-th graph connectivity matrix is ​​determined based on the labels of each node and the associated signal input nodes of each signal output node.

[0014] The circuit delay prediction method described in the embodiment of the present application determines the i-th eigenvector matrix based on the i-th RC network, including:

[0015] Determine the slew rate of each pin node in the i-th RC network using an EDA tool;

[0016] Determine a pin node associated with the kth non-pin node, and assign a slew rate of the kth non-pin node based on the slew rate of the associated pin node, wherein the pin node associated with the kth non-pin node is a pin node belonging to the same path as the kth non-pin node, 1≤k≤K, and K is the total number of non-pin nodes in the i-th RC network;

[0017] Determine the capacitance value parameters of each node based on the relevant capacitance of each node, the relevant capacitance of the pin node includes the pin capacitance and the connection capacitance, and the relevant capacitance of the non-pin node includes the connection capacitance;

[0018] Determining resistance value parameters of each node based on the connection resistance of each node;

[0019] An i-th eigenvector matrix is ​​determined based on the slew rate of each node, the capacitance value parameter of each node, and the resistance value parameter of each node.

[0020] The circuit delay prediction method described in the embodiment of the present application, wherein the slew rate of the kth non-pin node is assigned based on the slew rate of the related pin node, comprises:

[0021] The slew rate of the kth non-pin node is assigned using the following rule:

[0022] The slew rate of the kth non-pin node is greater than or equal to the slew rate of the first related pin node and less than or equal to the slew rate of the second related pin node;

[0023] The closer the distance between the kth non-pin node and the first related pin node is, the closer the slew rate of the kth non-pin node is to the slew rate of the first related pin node.

[0024] The circuit delay prediction method described in the embodiment of the present application, wherein the delay of the i-th RC network is predicted by using the i-th graph connection matrix and the i-th eigenvector matrix, comprises:

[0025] Inputting the i-th graph connection matrix and the i-th eigenvector matrix into a pre-trained graph neural network;

[0026] The graph neural network is used to predict the time delay of the i-th RC network.

[0027] The circuit delay prediction method described in the embodiment of the present application, wherein the delay of the i-th RC network is predicted by using the graph neural network, comprises:

[0028] Aggregating the eigenvector of each node and the eigenvectors of adjacent nodes based on the i-th graph connectivity matrix and the i-th eigenvector matrix to obtain respective first aggregated eigenvectors;

[0029] Encoding the first aggregated feature vectors using a first fully connected layer and a first activation function to obtain first expanded-dimensional feature vectors;

[0030] Aggregating the first expanded-dimensional feature vector of each node and the first expanded-dimensional feature vectors of adjacent nodes to obtain respective second aggregated feature vectors;

[0031] Encoding each of the second aggregated feature vectors using a second fully connected layer and a second activation function to obtain each second dimension-expanded feature vector;

[0032] Regression processing is performed on each of the second expanded-dimensional feature vectors to predict the time delay of the i-th RC network.

[0033] The circuit delay prediction method described in the embodiment of the present application further includes:

[0034] Pre-acquire multiple RC networks and determine the graph connection matrix and eigenvector matrix of each RC network;

[0035] Determine the actual time delay of each RC network and use the actual time delay of each RC network as a label for each graph connection matrix and each eigenvector matrix;

[0036] The graph neural network is trained using the graph connection matrices and feature vector matrices with labels.

[0037] The present application also provides a circuit delay prediction device, the device comprising:

[0038] An acquisition module, used for acquiring an i-th RC network corresponding to an i-th net of a circuit, 1≤i≤I, where I is the total number of nets in the circuit;

[0039] A determination module, configured to determine an i-th graph connection matrix and an i-th eigenvector matrix based on the i-th RC network;

[0040] A prediction module is used to predict the delay of the i-th RC network using the i-th graph connection matrix and the i-th eigenvector matrix.

[0041] An embodiment of the present application further provides a terminal device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program runs on the processor, the circuit delay prediction method described in the embodiment of the present application is provided.

[0042] The embodiment of the present application further provides a readable storage medium storing a computer program. When the computer program is run on a processor, the circuit delay prediction method described in the embodiment of the present application is executed.

[0043] The present application obtains the i-th RC network corresponding to the i-th net of the circuit, 1≤i≤I, I is the total number of nets in the circuit; determines the i-th graph connection matrix and the i-th eigenvector matrix based on the i-th RC network; and predicts the delay of the i-th RC network using the i-th graph connection matrix and the i-th eigenvector matrix. The graph connection matrix of the present application includes the relationship between each path in the wiring structure to be predicted, and the eigenvector matrix includes the common attributes of each node in the wiring structure to be predicted. Based on the graph connection matrix and the eigenvector matrix, accurate prediction of the wiring structure to be predicted can be achieved, and the prediction speed is faster. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solution of the present invention, the following is a brief introduction to the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope of protection of the present invention. In each of the drawings, similar components are numbered similarly.

[0045] Figure 1 A schematic diagram showing the structure of a circuit network including multiple units proposed in an embodiment of the present application is shown;

[0046] Figure 2 A schematic diagram showing a flow chart of a circuit delay prediction method proposed in an embodiment of the present application is shown;

[0047] Figure 3 A schematic diagram of the structure of an RC network corresponding to a net proposed in an embodiment of the present application is shown;

[0048] Figure 4 A schematic diagram of the structure of a processed RC network proposed in an embodiment of the present application is shown;

[0049] Figure 5 A schematic diagram of a process for training a graph neural network proposed in an embodiment of the present application is shown;

[0050] Figure 6 A schematic diagram of the structure of a circuit delay prediction device proposed in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0051] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0052] The components of the embodiments of the present invention generally described and shown in the drawings herein may be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0053] Hereinafter, the terms "including", "having" and their cognates, which may be used in various embodiments of the present invention, are intended only to indicate specific features, numbers, steps, operations, elements, components or combinations of the foregoing items, and should not be understood as first excluding the existence of one or more other features, numbers, steps, operations, elements, components or combinations of the foregoing items or adding the possibility of one or more features, numbers, steps, operations, elements, components or combinations of the foregoing items.

[0054] Furthermore, the terms “first”, “second”, “third”, etc. are merely used for distinguishing descriptions and are not to be understood as indicating or implying relative importance.

[0055] Unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meanings as those generally understood by those skilled in the art to which the various embodiments of the present invention belong. The terms (such as those defined in generally used dictionaries) will be interpreted as having the same meanings as the contextual meanings in the relevant technical field and will not be interpreted as having idealized meanings or overly formal meanings unless clearly defined in the various embodiments of the present invention.

[0056] It should be noted that the circuit network in the present application includes multiple units, each unit includes multiple pins, and the units are connected to each other through the pins. Figure 1, including four net networks, among which the net network between pin 1 of the first unit and pin 1 of the third unit includes two pin nodes (pin 1 of the first unit and pin 1 of the third unit), excluding non-pin nodes. If the signal flows from pin 1 of the first unit to pin 1 of the third unit, pin 1 of the first unit is a driving pin and pin 1 of the third unit is a target pin; the net network between pin 2 of the first unit and pin 2 of the third unit includes two pin nodes (pin 2 of the first unit and pin 2 of the third unit), excluding non-pin nodes. If the signal flows from pin 2 of the first unit to pin 2 of the third unit, pin 2 of the first unit is a driving pin and pin 2 of the third unit is a target pin; the net network between pin 3 of the first unit and pin 1 of the fourth unit includes eight pin nodes (pin 3 and pin 4 of the first unit, pin 1 and pin 2 of the fourth unit, pin 1 and pin 2 of the second unit, pin 3 and pin 4 of the third unit), six non-pin nodes. Pin nodes (a, b, c, d, e and f), when pin 3 of the first unit and pin 1 of the fourth unit are the main paths and if the signal flows from pin 3 of the first unit to pin 1 of the fourth unit, pin 3 of the first unit is the driving pin, pin 1 of the fourth unit is the target pin, and the remaining pins in the above eight pin nodes are related pins; the net network between pin 3 of the second unit and pin 3 of the fourth unit includes two pin nodes (pin 3 of the second unit and pin 3 of the third unit), excluding non-pin Node, if the signal flows from pin 3 of the second unit to pin 3 of the fourth unit, pin 3 of the second unit is the driving pin, and pin 3 of the fourth unit is the target pin; the net network between pin 4 of the second unit and pin 4 of the fourth unit includes two pin nodes (pin 4 of the second unit and pin 4 of the third unit), and does not include non-pin nodes. If the signal flows from pin 4 of the second unit to pin 4 of the fourth unit, pin 4 of the second unit is the driving pin, and pin 4 of the fourth unit is the target pin.

[0057] This application determines the graph connection matrix based on the ideal RC (Resistance and Cap) network structure of each net between each cell in the wiring structure to be predicted, and determines the eigenvector matrix based on the common properties of each node in the RC (Resistance and Cap) network structure (for example, capacitance, resistance, slew rate), and uses the graph neural network to predict the network delay after the wiring structure to be predicted is wired according to the graph connection matrix and the eigenvector matrix. Then, when the predicted network delay meets the preset conditions, the wiring structure to be predicted can be used as the wiring of the circuit. When the predicted network delay does not meet the preset conditions, the wiring structure to be predicted is replanned and adjusted, and the new wiring structure to be predicted is re-predicted until the predicted network delay meets the wiring position of the preset conditions.

[0058] It can be understood that the graph connection matrix of the present application includes the relationship between each path in the wiring structure to be predicted, and the eigenvector matrix includes the common attributes of each node in the wiring structure to be predicted. Based on the graph connection matrix and the eigenvector matrix, accurate prediction of the wiring structure to be predicted can be achieved, and the prediction speed is faster.

[0059] The present application will be further explained below through specific embodiments.

[0060] Example 1

[0061] See also Figure 2 In one embodiment of the present application, a circuit delay prediction method is provided, the method comprising the following steps:

[0062] Step S100, obtaining the i-th RC network corresponding to the i-th net of the circuit, 1≤i≤I, I is the total number of nets in the circuit.

[0063] For example, see Figure 3, an RC network corresponding to a net is given, wherein the path between a driving pin node (Drivingpin) and a target pin node (target pin) can be used as a main path, context pin 0, contextpin 1, context pin 2, context pin 3, and context pin 4 are related pin nodes, node 0, node 1, and node 2 are non-pin nodes, and each node includes a capacitance attribute, a resistance attribute, and a slew rate attribute. The capacitance value parameter corresponding to the capacitance attribute of the pin node is equal to the sum of the pin capacitance and the wire capacitance, that is, C=Cwire+Cpin, Cwire is the wire capacitance, and Cpin is the pin capacitance. The resistance value parameter corresponding to the resistance attribute of the pin node is equal to the value obtained by dividing the sum of all wire resistance values ​​related to the pin node by two (for example, the resistance value parameter R of the driving pin node (Driving pin) Driving =R0 / 2), the slew rate corresponding to the slew rate attribute of the pin node can be determined using EDA tools, the capacitance value parameter corresponding to the capacitance attribute of the non-pin node is equal to the wire capacitance, that is, C=Cwire, and the resistance value parameter corresponding to the resistance attribute of the non-pin node is equal to the sum of all wire resistance values ​​related to the non-pin node divided by two (for example, the resistance value parameter R of the non-pin node (node ​​0) node0 =(R0+R1+R2+R3) / 2), the slew rate corresponding to the slew rate attribute of the non-pin node can be assigned according to the slew rate of the pin node.

[0064] Step S200: determining an i-th graph connection matrix and an i-th eigenvector matrix based on the i-th RC network.

[0065] In this embodiment, the i-th graph connection matrix may be determined based on the i-th RC network.

[0066] For example, see Figure 4 , add labels to each node in the i-th RC network. Each node can be labeled in a certain labeling order, as long as the labeling order is always consistent. In this embodiment, labeling starts from the driving pin node (Driving pin), and the labels of each node are recorded as 0, 1, 2, 3, 4, 5, 6, 7, 8, and 9 respectively.

[0067] Furthermore, since the signal in each RC network flows from the driving pin node to each related pin node and the target pin node, the signal output node can be determined from all nodes based on the signal flow direction in the i-th RC network. Figure 4There are 4 signal output nodes in total, which are the nodes corresponding to label 0, label 2, label 4 and label 5.

[0068] Furthermore, the node receiving the output signal of the j-th signal output node is used as the related signal input node of the j-th signal output node, 1≤j≤J, J is the total number of signal output nodes in the i-th RC network. Figure 4 Taking the example shown, J=4, when j=1, the first signal output node is the node corresponding to label 0, and the node receiving the node output signal corresponding to label 0 is the node corresponding to label 2, when j=2, the second signal output node is the node corresponding to label 2, and the nodes receiving the node output signal corresponding to label 2 are the nodes corresponding to labels 1 and 3, when j=3, the third signal output node is the node corresponding to label 5, and the nodes receiving the node output signal corresponding to label 5 are the nodes corresponding to labels 4 and 6, when j=4, the fourth signal output node is the node corresponding to label 4, and the nodes receiving the node output signal corresponding to label 4 are the nodes corresponding to labels 8 and 9.

[0069] Further, the i-th graph connection matrix is ​​determined based on the labels of each node and the related signal input nodes of each signal output node. Exemplarily, Figure 4 The corresponding graph connection matrix can be expressed as follows:

[0070] S_indeX: [0, 2, 2, 2, 5, 5, 5, 4, 4]

[0071] T_indeX: [2, 1, 3, 5, 4, 6, 7, 8, 9]

[0072] In this embodiment, the ith eigenvector matrix may be determined based on the ith RC network, and the ith eigenvector matrix is ​​mainly composed of the slew rate of each node, the capacitance value parameter of each node, and the resistance value parameter of each node.

[0073] For example, the slew rate of each pin node in the i-th RC network can be determined by using an EDA tool. Figure 4 As shown in the figure, for example, the nodes corresponding to label 0, label 1, label 3, label 6, label 7, label 8 and label 9 are pin nodes, so the slew rate of the nodes corresponding to label 0, label 1, label 3, label 6, label 7, label 8 and label 9 can be determined by using EDA tools. However, the EDA tool cannot determine the slew rate of non-pin nodes, so in this embodiment, the slew rate of non-pin nodes can be determined by the following method:

[0074] A pin node related to the kth non-pin node is determined, and a slew rate of the kth non-pin node is assigned based on the slew rate of the related pin node. It should be noted that, in this embodiment, the pin node related to the kth non-pin node is a pin node belonging to the same path as the kth non-pin node, 1≤k≤K, and K is the total number of non-pin nodes in the i-th RC network.

[0075] It should be noted that if the Elmore Delay of the main path from the driving pin node to the target pin node is D dri , the slew rate is S dri , the delay (Elmore Delay) of the node corresponding to label 2 is D 0 , the slew rate is S 0 , the delay (ElmoreDelay) of the node corresponding to label 5 is D 1 , the slew rate is S 1 , the slew rate of the target pin is S tar ,but:

[0076] S 0 =S dri +(S tar -S dri )(D dri –D 1 ) / D dri

[0077] S 1 =S dri +(S tar -S dri )(D dri –D 0 ) / D dri

[0078] Among them, S 0 and S 1 Unknown, the other parameters are known, if S dri =10, S tar =20, D dri =1ns, D 0 =0.5ns, D 1 = 0.2ns, then the slew rate of the node corresponding to label 2 is S 0 =10+(20–10)(1–0.5) / 1=15, the slew rate S of the node corresponding to label 5 1 =10+(20–10)(1–0.2) / 1=18.

[0079] Therefore, when assigning a slew rate to a non-pin node, the slew rate to the non-pin node may be assigned based on the following rules:

[0080] The slew rate of the kth non-pin node is greater than or equal to the slew rate of the first related pin node and less than or equal to the slew rate of the second related pin node; the closer the distance between the kth non-pin node and the first related pin node is, the closer the slew rate of the kth non-pin node is to the slew rate of the first related pin node.

[0081] by Figure 4 For example, K=3, the first non-pin node is the node corresponding to label 2, the second non-pin node is the node corresponding to label 5, and the third non-pin node is the node corresponding to label 4. When determining the slew rate of the node corresponding to label 5 and the node corresponding to label 2, the pin nodes (driving pin to target pin) on the main path (the path between the driving pin node and the target pin node) are first used as the related pin nodes of the node corresponding to label 5 and the node corresponding to label 2. If the slew rate of the node corresponding to label 0 (driving pin) is 10, the node corresponding to label 7 (target pin) is pin)) is 20, then the slew rate of the node corresponding to label 2 and the node corresponding to label 5 is greater than or equal to 10 and less than or equal to 20. Because the node corresponding to label 2 is closer to the node corresponding to label 0, the slew rate of the node corresponding to label 2 is closer to 10. Compared with the slew rate of the node corresponding to label 2, the slew rate of the node corresponding to label 5 is farther away from 10 and closer to 20. Therefore, the slew rate of the node corresponding to label 2 can be assigned 15, and the slew rate of the node corresponding to label 5 can be assigned 18.

[0082] Furthermore, the capacitance value parameters of each node can be determined based on the relevant capacitance of each node. The relevant capacitance of the pin node includes the pin capacitance and the connection capacitance. The capacitance value parameter of the pin node is equal to the sum of the pin capacitance and the connection capacitance. The relevant capacitance of the non-pin node includes the connection capacitance. The capacitance value parameter of the non-pin node is equal to the connection capacitance.

[0083] Furthermore, the resistance value parameter of each node is determined based on the connection resistance of each node, and the resistance value parameter of each node is equal to the sum of all connection resistances corresponding to the node divided by 2.

[0084] For example, Figure 4 For example, the i-th eigenvector matrix is:

[0085]

[0086] Among them, the dimension of the i-th eigenvector matrix is ​​3*10, 3 corresponds to the total number of node attributes, 10 corresponds to the total number of nodes, each column of the matrix is ​​a eigenvector of a node, and the matrix places the eigenvectors in the order of node labels from small to large from left to right. It can be understood that the placement order of the eigenvectors can also be other orders, for example, from large to small, as long as the placement order remains unique. The number of rows and columns is variable, the number of columns changes with the number of nodes, and the number of rows is related to the total number of pre-selected node attributes.

[0087] Step S300: predicting the delay of the i-th RC network using the i-th graph connection matrix and the i-th eigenvector matrix.

[0088] The delay of the i-th RC network can be predicted using any trained model, for example, Artificial Neural Network (ANN) or Random Forest (RF).

[0089] In this embodiment, taking into account the characteristics of the RC network, the delay is predicted using a graph neural network to ensure the accuracy of the prediction. The i-th graph connection matrix and the i-th eigenvector matrix can be input into a pre-trained graph neural network; the graph neural network is used to predict the delay of the i-th RC network. The use of a graph neural network to predict the delay of the i-th RC network mainly includes three steps, namely the first embedding, the second embedding, and regression. Each embedding includes two steps of aggregation and encoding, and regression obtains the final predicted delay.

[0090] Exemplarily, the eigenvector of each node and the eigenvector of the adjacent nodes can be aggregated based on the i-th graph connection matrix and the i-th eigenvector matrix to obtain each first aggregated eigenvector, that is, the aggregation process in the first embedding. The corresponding calculation formula is:

[0091]

[0092] Among them, Agg1 j is the first aggregated feature vector of the jth node, is the feature vector of the jth node, is the feature vector of the upstream node, su(v) is the feature vector set of the downstream node, That is, the downstream node feature vector of the jth node, w pr and w su is the training weight determined by pre-training. Figure 4 Taking the node corresponding to label 2 as an example, when calculating the first aggregated feature vector of the node corresponding to label 2, That is the feature vector of the node corresponding to label 2, That is, it is the feature vector of the node corresponding to label 0, and su(v) is the feature vector set of the nodes corresponding to label 1, label 3, and label 5. If the currently calculated node has no upstream node or downstream node, the corresponding item in the above formula is 0.

[0093] Furthermore, the first fully connected layer and the first activation function are used to encode each first aggregated feature vector to obtain each first expanded dimensional feature vector, that is, the encoding process in the first embedding, which uses the first fully connected layer and the first activation function to encode each 3*1 first aggregated feature vector to obtain the corresponding 8*1 first expanded dimensional feature vector. It should be noted that the first expanded dimensional feature vector can also be other dimensions, which can be greater than 3 dimensions in principle.

[0094] Furthermore, the first expanded dimension feature vector of each node and the first expanded dimension feature vector of the adjacent nodes are aggregated to obtain each second aggregated feature vector, that is, the aggregation process in the second embedding, and the corresponding calculation formula is:

[0095]

[0096] Among them, Agg2 j is the second aggregated feature vector of the jth node, is the first expanded feature vector of the jth node, is the first expanded dimension feature vector of the upstream node, su(v) is the first expanded dimension feature vector set of the downstream node, That is, it is the first expanded dimension feature vector of the downstream node of the jth node.

[0097] Furthermore, the second fully connected layer and the second activation function are used to encode the respective second aggregated feature vectors to obtain respective second expanded dimensional feature vectors, that is, the encoding process in the second embedding, which uses the second fully connected layer and the second activation function to encode each 8*1 second aggregated feature vector to obtain the corresponding 16*1 second expanded dimensional feature vector. It should be noted that the second expanded dimensional feature vector can also be of other dimensions, which in principle can be larger than the dimension of the first aggregated feature vector.

[0098] Further, regression processing is performed on each of the second dimension-expanded feature vectors to predict the delay of the i-th RC network, wherein the formula corresponding to the regression processing is: Agg3 j is the regression result, and w is the weight determined by pre-training the graph neural network.

[0099] Finally, the regression results are processed by the fully connected layer of the graph neural network to obtain a one-dimensional vector, and the final predicted delay is obtained through the one-dimensional vector.

[0100] In this embodiment, the graph connection matrix includes the relationship between each path in the wiring structure to be predicted, and the feature vector matrix includes the common attributes of each node in the wiring structure to be predicted. The graph neural network can accurately predict the wiring structure to be predicted based on the graph connection matrix and the feature vector matrix, and the prediction speed is faster.

[0101] Example 2

[0102] See also Figure 5 In this embodiment, the circuit delay prediction method proposed further includes the following steps:

[0103] Step S10, pre-acquire multiple RC networks, and determine the graph connection matrix and eigenvector matrix of each RC network.

[0104] The graph connection matrix and eigenvector matrix of each RC network may be determined by the method in the above embodiment.

[0105] Step S20, determining the actual time delay of each RC network, and using the actual time delay of each RC network as a label for each graph connection matrix and each eigenvector matrix.

[0106] Step S30, training the graph neural network using the graph connection matrices and feature vector matrices with labels.

[0107] This embodiment determines the optimal values ​​of relevant weights in each formula in the graph neural network by pre-training the graph neural network, thereby ensuring the accuracy of the graph neural network in predicting circuit network delays.

[0108] Example 3

[0109] See also Figure 6 In an embodiment of the present application, a circuit delay prediction device 10 is further proposed, including: an acquisition module 11, a determination module 12 and a prediction module 13.

[0110] An acquisition module 11 is used to obtain the i-th RC network corresponding to the i-th net of the circuit, 1≤i≤I, and I is the total number of nets in the circuit; a determination module 12 is used to determine the i-th graph connection matrix and the i-th eigenvector matrix based on the i-th RC network; a prediction module 13 is used to predict the delay of the i-th RC network using the i-th graph connection matrix and the i-th eigenvector matrix.

[0111] The circuit delay prediction device 10 disclosed in this embodiment is used to execute the circuit delay prediction method described in the above embodiment through the coordinated use of the acquisition module 11, the determination module 12 and the prediction module 13. The implementation scheme and beneficial effects involved in the above embodiment are also applicable to this embodiment and will not be repeated here.

[0112] Example 4

[0113] An embodiment of the present application further proposes a terminal device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program runs on the processor, the circuit delay prediction method described in the above embodiment of the present application is executed.

[0114] Example 5

[0115] An embodiment of the present application further provides a readable storage medium storing a computer program, which executes the circuit delay prediction method described in the above embodiment of the present application when running on a processor.

[0116] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can also be implemented in other ways. The device embodiments described above are merely schematic. For example, the flowcharts and structure diagrams in the accompanying drawings show the possible architecture, functions and operations of the devices, methods and computer program products according to multiple embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in an alternative implementation, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the structure diagram and / or the flow chart, and the combination of boxes in the structure diagram and / or the flow chart, can be implemented with a dedicated hardware-based system that performs a specified function or action, or can be implemented with a combination of dedicated hardware and computer instructions.

[0117] In addition, the functional modules or units in the various embodiments of the present invention may be integrated together to form an independent part, or each module may exist independently, or two or more modules may be integrated to form an independent part.

[0118] If the functions are implemented in the form of software function modules and sold or used as independent products, they can be stored in a readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a smart phone, a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned readable storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0119] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A circuit delay prediction method, characterized in that: The method comprises: Get the i-th RC network corresponding to the i-th net of the circuit, 1≤i≤I, I is the total number of nets in the circuit; Determining an i-th graph connection matrix and an i-th eigenvector matrix based on the i-th RC network; Predicting the delay of the i-th RC network using the i-th graph connection matrix and the i-th eigenvector matrix; Wherein, determining the i-th graph connection matrix based on the i-th RC network includes: Adding labels to each node in the i-th RC network; Determine a signal output node from each of the nodes based on the signal flow direction in the i-th RC network; Using a node that receives an output signal of the j-th signal output node as a related signal input node of the j-th signal output node, 1≤j≤J, J is the total number of signal output nodes in the i-th RC network; Determine an i-th graph connection matrix based on the labels of each node and the associated signal input nodes of each signal output node; Wherein, determining the i-th eigenvector matrix based on the i-th RC network includes: Determine the slew rate of each pin node in the i-th RC network using an EDA tool; Determine a pin node associated with the kth non-pin node, and assign a slew rate of the kth non-pin node based on the slew rate of the associated pin node, wherein the pin node associated with the kth non-pin node is a pin node belonging to the same path as the kth non-pin node, 1≤k≤K, and K is the total number of non-pin nodes in the i-th RC network; Determine the capacitance value parameters of each node based on the relevant capacitance of each node, the relevant capacitance of the pin node includes the pin capacitance and the connection capacitance, and the relevant capacitance of the non-pin node includes the connection capacitance; Determining resistance value parameters of each node based on the connection resistance of each node; Determine an i-th eigenvector matrix based on the slew rate of each node, the capacitance value parameter of each node, and the resistance value parameter of each node; The step of predicting the delay of the i-th RC network by using the i-th graph connection matrix and the i-th eigenvector matrix comprises: Pre-acquire multiple RC networks and determine the graph connection matrix and eigenvector matrix of each RC network; Determine the actual time delay of each RC network and use the actual time delay of each RC network as a label for each graph connection matrix and each eigenvector matrix; Use each labeled graph connection matrix and feature vector matrix to train the graph neural network; Inputting the i-th graph connection matrix and the i-th eigenvector matrix into a pre-trained graph neural network; Predicting the delay of the i-th RC network using the graph neural network; The step of using the graph neural network to predict the delay of the i-th RC network includes: Aggregating the eigenvector of each node and the eigenvectors of adjacent nodes based on the i-th graph connectivity matrix and the i-th eigenvector matrix to obtain respective first aggregated eigenvectors; Encoding the first aggregated feature vectors using a first fully connected layer and a first activation function to obtain first expanded-dimensional feature vectors; Aggregating the first expanded-dimensional feature vector of each node and the first expanded-dimensional feature vectors of adjacent nodes to obtain respective second aggregated feature vectors; Encoding each of the second aggregated feature vectors using a second fully connected layer and a second activation function to obtain each second dimension-expanded feature vector; Regression processing is performed on each of the second expanded-dimensional feature vectors to predict the time delay of the i-th RC network.

2. The circuit delay prediction method according to claim 1, characterized in that: The assigning a slew rate of the kth non-pin node based on the slew rate of the related pin node includes: The slew rate of the kth non-pin node is assigned using the following rule: The slew rate of the kth non-pin node is greater than or equal to the slew rate of the first related pin node and less than or equal to the slew rate of the second related pin node; The closer the distance between the kth non-pin node and the first related pin node is, the closer the slew rate of the kth non-pin node is to the slew rate of the first related pin node.

3. A circuit delay prediction device, characterized in that: The device comprises: An acquisition module, used for acquiring an i-th RC network corresponding to an i-th net of a circuit, 1≤i≤I, where I is the total number of nets in the circuit; A determination module, configured to determine an i-th graph connection matrix and an i-th eigenvector matrix based on the i-th RC network; A prediction module, configured to predict the delay of the i-th RC network using the i-th graph connection matrix and the i-th eigenvector matrix; The determination module is further used to add labels to each node in the i-th RC network; determine a signal output node from each node based on the signal flow direction in the i-th RC network; use a node that receives an output signal of the j-th signal output node as a related signal input node of the j-th signal output node, 1≤j≤J, J is the total number of signal output nodes in the i-th RC network; determine an i-th graph connection matrix based on the labels of each node and the related signal input nodes of each signal output node; The determination module is further used to determine the slew rate of each pin node in the i-th RC network by using an EDA tool; determine the pin node related to the k-th non-pin node, and assign a value to the slew rate of the k-th non-pin node based on the slew rate of the related pin node, the pin node related to the k-th non-pin node is a pin node belonging to the same path as the k-th non-pin node, 1≤k≤K, K is the total number of non-pin nodes in the i-th RC network; determine the capacitance value parameters of each node based on the relevant capacitance of each node, the relevant capacitance of the pin node includes the pin capacitance and the connection capacitance, and the relevant capacitance of the non-pin node includes the connection capacitance; determine the resistance value parameters of each node based on the connection resistance of each node; determine the i-th eigenvector matrix based on the slew rate of each node, the capacitance value parameters of each node and the resistance value parameters of each node; The prediction module is also used to pre-acquire multiple RC networks and determine the graph connection matrix and eigenvector matrix of each RC network; determine the actual delay of each RC network, and use the actual delay of each RC network as a label of each graph connection matrix and each eigenvector matrix; use each labeled graph connection matrix and eigenvector matrix to train the graph neural network; input the i-th graph connection matrix and the i-th eigenvector matrix into the pre-trained graph neural network; use the graph neural network to predict the delay of the i-th RC network; The prediction module is also used to aggregate the feature vectors of each node and the feature vectors of adjacent nodes based on the i-th graph connection matrix and the i-th feature vector matrix to obtain each first aggregated feature vector; encode the each first aggregated feature vector using the first fully connected layer and the first activation function to obtain each first expanded dimension feature vector; aggregate the first expanded dimension feature vector of each node and the first expanded dimension feature vector of the adjacent nodes to obtain each second aggregated feature vector; encode the each second aggregated feature vector using the second fully connected layer and the second activation function to obtain each second expanded dimension feature vector; and perform regression processing on the each second expanded dimension feature vector to predict the delay of the i-th RC network.

4. A terminal device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and when the computer program is run on the processor, the circuit delay prediction method according to any one of claims 1 to 2 is executed.

5. A readable storage medium, characterized in that: A computer program is stored therein, and when the computer program is run on a processor, the circuit delay prediction method according to any one of claims 1 to 2 is executed.