Multi-chip integration layout design method
The multi-chip integrated layout design method enhances efficiency and reduces costs by optimizing chip layouts through feature extraction and mining processes, addressing performance issues like airtightness and heat dissipation without manual iterations.
Patent Information
- Application Number
- CN202411184603.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-27
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2044-08-27
AI Technical Summary
The existing multi-chip integrated layout design process is inefficient and costly due to its high complexity and the need for manual iterations to optimize layout schemes, which affects routability and routing parameters.
A multi-chip integrated layout design method that involves determining chip model and layout optimization data, performing feature extraction and mining, and using multi-level feature mining processes to optimize chip layout vectors, ensuring reliable feature representation and minimizing manual intervention.
The method improves efficiency and reduces costs by providing optimized chip layouts that address performance issues such as airtightness, volume, and heat dissipation without manual iterations, ensuring reliable feature representation and optimized layout characteristics.
Smart Images

Figure CN118690721B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of chip design, and in particular to a multi-chip integrated layout design method. Background Art
[0002] In the development of integrated circuits, Electronic Design Automation (EDA) technology has gradually matured and its application scope has gradually expanded. In the physical design stage of integrated circuits, layout and routing are important and time-consuming steps. First, the layout process involves a large number of iterations and optimizations, and the time required will significantly affect the integrated circuit design cycle. Secondly, in the physical design of integrated circuits, there is a close connection between the various steps, and the results of the layout will affect the routability and the parameters such as the running time, congestion and routing rate of the routing process.
[0003] In recent years, in addition to layout algorithms driven by line length and delay, layout algorithms driven by routability have also received attention. Although layout algorithms have made significant progress in the past few decades, fast and efficient layout is still a challenging problem. Especially for the application scenarios of multi-chip integrated layout, the layout design scheme has higher complexity. After the initial scheme is formed, it is generally optimized and adjusted manually to overcome various existing problems. In this way, it is easy to have problems such as low efficiency and high cost of multi-chip integrated layout design. Summary of the invention
[0004] In view of this, the purpose of the present application is to provide a multi-chip integrated layout design method to improve the problems of low efficiency and high cost of multi-chip integrated layout design in the prior art.
[0005] To achieve the above objectives, this application adopts the following technical solutions:
[0006] A multi-chip integrated layout design method, comprising:
[0007] Determine chip model data and chip layout optimization data of the integrated chip to be optimized, wherein the integrated chip to be optimized is an integrated chip in which multiple sub-chips and / or multiple devices are stacked on multiple cavities on the front and back sides of a substrate, the chip model data is formed based on three-dimensional modeling of the integrated chip to be optimized, the chip layout optimization data is formed based on simulation of the chip model data, and the chip layout optimization data is used to reflect performance problems existing in the integrated chip to be optimized, and the performance problems include at least one of problems related to air tightness, problems related to chip volume, problems related to structural stability, and problems related to heat dissipation;
[0008] Performing feature extraction processing on the chip model data and the chip layout optimization data respectively, and outputting a chip layout vector corresponding to the chip model data and a chip layout optimization vector corresponding to the chip layout optimization data;
[0009] Perform multi-level first feature mining processing on the chip layout vector, and output a deep chip layout vector of the integrated chip to be optimized at multiple levels reflected by the chip model data;
[0010] Combined with the deep chip layout vectors at the multiple levels, the chip layout optimization vector is subjected to multi-level second feature mining processing to obtain optimized chip layout vectors at the multiple levels, wherein the optimized chip layout vector is used to reflect the feature information of the integrated chip to be optimized in two dimensions of the chip model data and the chip layout optimization data;
[0011] The optimized chip layout vectors at the multiple levels are subjected to feature restoration processing to form chip model optimization data that integrates the chip layout feature information of the integrated chip to be optimized and the layout optimization feature information of the chip layout optimization data, wherein the chip model optimization data is used to reflect the layout status of the integrated chip after optimization.
[0012] In a preferred embodiment of the present application, in the above-mentioned multi-chip integrated layout design method, the step of performing multi-level first feature mining processing on the chip layout vector and outputting the deep chip layout vector of the integrated chip to be optimized at multiple levels reflected by the chip model data includes:
[0013] Using a plurality of intermediate feature output units in the deep feature mining model formed by the optimization operation, the chip layout vector is subjected to a multi-level first feature mining process, and a deep chip layout vector of the integrated chip to be optimized at multiple levels reflected by the chip model data is output;
[0014] wherein, for the first feature mining process of the first level performed on the chip layout vector, the first intermediate feature output unit among the plurality of intermediate feature output units is used to perform the first feature mining process of the chip layout vector at the first level, and the deep chip layout vector of the integrated chip to be optimized at the first level reflected by the chip model data is output;
[0015] Furthermore, for the first feature mining processing of the ath level performed on the chip layout vector, the ath intermediate feature output unit among the multiple intermediate feature output units is used to perform the first feature mining processing of the ath level on the deep chip layout vector of the bth level, and the deep chip layout vector of the integrated chip to be optimized at the ath level reflected by the chip model data is output, ab=1, and a is greater than 1, and b is greater than or equal to 1.
[0016] In a preferred embodiment of the present application, in the above-mentioned multi-chip integrated layout design method, the intermediate feature output unit is a first gradient optimization unit, and the first gradient optimization unit includes a gradient optimization relationship;
[0017] The step of performing multi-level first feature mining processing on the chip layout vector by using multiple intermediate feature output units in the deep feature mining model formed by the optimization operation, and outputting the deep chip layout vector of the integrated chip to be optimized at multiple levels reflected by the chip model data comprises:
[0018] Using the gradient optimization relationship in each first gradient optimization unit in the plurality of intermediate feature output units, the chip layout vector is subjected to relationship mapping processing to implement multi-level first feature mining processing, thereby outputting the deep chip layout vector of the integrated chip to be optimized at multiple levels reflected by the chip model data;
[0019] Among them, each of the first gradient optimization units includes multiple convolution sub-units. For any one of the first gradient optimization units, after the input vector of the first gradient optimization unit is convolved by the multiple convolution sub-units, the obtained convolution vector and the input vector will be vector aggregated, and the result of the vector aggregation processing will be used as the output vector of the first gradient optimization unit.
[0020] In a preferred embodiment of the present application, in the above-mentioned multi-chip integrated layout design method, the step of combining the deep chip layout vectors at the multiple levels, performing multi-level second feature mining processing on the chip layout optimization vector, and obtaining the optimized chip layout vectors at the multiple levels includes:
[0021] Using the optimized feature mining model formed by the optimization operation, combined with the deep chip layout vectors at the multiple levels, the chip layout optimization vector is subjected to multi-level second feature mining processing to obtain the optimized chip layout vectors at the multiple levels;
[0022] In which, the optimized feature mining model includes multiple second gradient optimization units connected in cascade. For every two adjacent second gradient optimization units, a focused feature mining unit is configured between the two second gradient optimization units, so that the output vector of the previous second gradient optimization unit is processed by the focused feature mining unit and then output to the next second gradient optimization unit for corresponding processing. The number of the second gradient optimization units is the same as the number of the first gradient optimization units in the deep feature mining model.
[0023] In a preferred embodiment of the present application, in the above-mentioned multi-chip integrated layout design method, the step of using the optimized feature mining model formed by the optimization operation, combining the deep chip layout vectors at multiple levels, performing multi-level second feature mining processing on the chip layout optimization vectors, and obtaining the optimized chip layout vectors at multiple levels includes:
[0024] For the chip layout optimization vector, a first second gradient optimization unit among the plurality of second gradient optimization units is used to perform a first-level second feature mining process on the chip layout optimization vector, and a first-level intermediate chip layout optimization vector of the chip layout optimization data after the first-level second feature mining process is output;
[0025] After the chip layout optimization vector is subjected to the second feature mining processing of the cth level, the cth focused feature mining unit in the optimization feature mining model is used to perform vector aggregation processing on the cth level intermediate chip layout optimization vector after the second feature mining processing of the cth level and the deep chip layout vector output by the cth first gradient optimization unit in the deep feature mining model, and the corresponding cth chip layout aggregation vector is output, wherein c is greater than or equal to 1;
[0026] In the process of performing the second feature mining processing of the dth level on the chip layout optimization vector, using the dth second gradient optimization unit among the multiple second gradient optimization units, the chip layout optimization vector and the eth chip layout aggregation vector output by the eth focused feature mining unit are subjected to the second feature mining processing of the dth level, and the intermediate chip layout optimization vector of the dth level after the second feature mining processing of the dth level of the chip layout optimization data is output, wherein d is greater than 1, and de=1;
[0027] After the second feature mining process of the last level is performed, the output intermediate chip layout optimization vectors of multiple levels are used as the optimized chip layout vectors of the multiple levels.
[0028] In a preferred embodiment of the present application, in the above-mentioned multi-chip integrated layout design method, before the step of performing a first-level second feature mining process on the chip layout optimization vector, performing a first-level second feature mining process on the chip layout optimization vector using a first second gradient optimization unit among the plurality of second gradient optimization units, and outputting a first-level intermediate chip layout optimization vector of the chip layout optimization data after the first-level second feature mining process, the multi-chip integrated layout design method further comprises:
[0029] Outputting a random semantic vector corresponding to the optimized feature mining model using a predetermined random vector determination rule;
[0030] Marking the random semantic vector and the chip layout optimization vector as vectors to be processed, and loading them into the optimization feature mining model;
[0031] The step of performing a first-level second feature mining process on the chip layout optimization vector, using a first second gradient optimization unit among the plurality of second gradient optimization units, performing a first-level second feature mining process on the chip layout optimization vector, and outputting a first-level intermediate chip layout optimization vector of the chip layout optimization data after the first-level second feature mining process, comprises:
[0032] With respect to the first-level second feature mining processing of the chip layout optimization vector, the first second gradient optimization unit among the plurality of second gradient optimization units is used to perform the first-level second feature mining processing on the chip layout optimization vector and the random semantic vector, and the first-level intermediate chip layout optimization vector of the chip layout optimization data after the first-level second feature mining processing is output.
[0033] In a preferred embodiment of the present application, in the above-mentioned multi-chip integrated layout design method, the multi-chip integrated layout design method is implemented based on a chip layout optimization network formed through a network optimization operation, the chip layout optimization network includes a first network part and a second network part, the first network part includes a first feature extraction model and an optimized feature mining model, and the second network part includes a second feature extraction model and a deep feature mining model;
[0034] The step of forming the chip layout optimization network through network optimization operation includes:
[0035] Loading the training chip model data in the training chip layout data combination to load into the second network part, performing feature extraction processing on the training chip model data using a second feature extraction model in the second network part, and outputting a training chip layout vector;
[0036] Using the deep feature mining model in the second network part, the training chip layout vector is subjected to multi-level first feature mining processing, and the training deep chip layout vector of the training integrated chip to be optimized at multiple levels reflected by the training chip model data is output;
[0037] Loading the training chip layout optimization data in the training chip layout data combination to load into the first network part, performing feature extraction processing on the training chip layout optimization data using a first feature extraction model in the first network part, and outputting a training chip layout optimization vector;
[0038] Utilizing the optimized feature mining model in the first network part, combined with the training deep chip layout vectors at the multiple levels, the training chip layout optimization vectors are subjected to multi-level second feature mining processing, and the corresponding training optimized chip layout vectors at the multiple levels are output;
[0039] Determining a network optimization error of the chip layout optimization network according to the training depth chip layout vectors at the multiple levels and the training optimization chip layout vectors at the multiple levels;
[0040] According to the network optimization error, the network parameters in the chip layout optimization network are optimized and adjusted, and the chip layout optimization network after the network optimization operation is output.
[0041] In a preferred embodiment of the present application, in the above-mentioned multi-chip integrated layout design method, the step of determining the network optimization error of the chip layout optimization network based on the training depth chip layout vectors at the multiple levels and the training optimization chip layout vectors at the multiple levels includes:
[0042] Determining a first network optimization error of the chip layout optimization network according to the training optimized chip layout vectors at the multiple levels and the training chip model data;
[0043] Determining a second network optimization error of the chip layout optimization network according to the training depth chip layout vectors at the multiple levels and the training optimization chip layout vectors at the multiple levels;
[0044] The first network optimization error and the second network optimization error are integrated to output a network optimization error of the chip layout optimization network.
[0045] In a preferred embodiment of the present application, in the above-mentioned multi-chip integrated layout design method, the step of determining the first network optimization error of the chip layout optimization network based on the training optimized chip layout vectors at the multiple levels and the training chip model data includes:
[0046] Performing feature restoration processing on the training optimized chip layout vectors at the multiple levels to form training chip model optimization data that incorporates the chip layout feature information of the training integrated chip to be optimized and the layout optimization feature information of the training chip layout optimization data;
[0047] The training chip model optimization data and the reference chip model optimization data corresponding to the training chip model data are subjected to difference determination processing, and a first network optimization error of the chip layout optimization network is output, wherein the reference chip model optimization data is used to describe an integrated chip after optimization for performance problems existing in the training integrated chip to be optimized as reflected by the training chip model data.
[0048] In a preferred embodiment of the present application, in the above-mentioned multi-chip integrated layout design method, the step of determining the second network optimization error of the chip layout optimization network according to the training depth chip layout vectors at the multiple levels and the training optimization chip layout vectors at the multiple levels includes:
[0049] Performing feature restoration processing on the training deep chip layout vectors at the multiple levels to form corresponding training chip model restoration data;
[0050] Performing feature restoration processing on the training optimized chip layout vectors at the multiple levels to form training chip model optimization data that incorporates the chip layout feature information of the training integrated chip to be optimized and the layout optimization feature information of the training chip layout optimization data;
[0051] After performing feature extraction processing on the training chip model restoration data and the training chip model optimization data respectively, performing difference determination processing on the output vector, and outputting the second network optimization error of the chip layout optimization network.
[0052] The multi-chip integrated layout design method provided in the present application first determines chip model data and chip layout optimization data; secondly, performs feature extraction processing on the chip model data and the chip layout optimization data respectively, and outputs a chip layout vector and a chip layout optimization vector; then, performs multi-level first feature mining processing on the chip layout vector, and outputs deep chip layout vectors at multiple levels; further, in combination with the deep chip layout vectors at multiple levels, performs multi-level second feature mining processing on the chip layout optimization vector, and obtains optimized chip layout vectors at multiple levels; finally, performs feature restoration processing on the optimized chip layout vectors at multiple levels to form chip model optimization data. Based on the above content, since the chip layout optimization vector will be subjected to multi-level second feature mining processing in combination with the deep chip layout vectors at multiple levels, the optimized chip layout vectors at multiple levels obtained can reflect the feature information of the integrated chip to be optimized in two dimensions of chip model data and chip layout optimization data. That is to say, it can not only characterize the original chip layout characteristics, but also reflect the characteristics of the performance problems that need to be optimized, so that the basis for the feature restoration processing of the optimized chip layout vector can be more reliable and sufficient. Therefore, the restored chip model optimization data can achieve the optimization of performance problems while ensuring the matching of the original chip layout characteristics, and no manual iteration and other operations are required during the optimization process, so that the problems of low efficiency and high cost of multi-chip integrated layout design in the prior art can be improved, that is, it has the advantages of reliability, efficiency and cost of solution optimization. BRIEF DESCRIPTION OF THE DRAWINGS
[0053] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, preferred embodiments are specifically cited below and described in detail with reference to the attached drawings.
[0054] Figure 1 A structural block diagram of an electronic device provided in an embodiment of the present application.
[0055] Figure 2 A schematic diagram of a flow chart of a multi-chip integrated layout design method provided in an embodiment of the present application.
[0056] Figure 3 A network architecture diagram for the first feature mining process provided in an embodiment of the present application.
[0057] Figure 4 Another network architecture diagram for the first feature mining process provided in an embodiment of the present application.
[0058] Figure 5 A network architecture diagram for the second feature mining process provided in an embodiment of the present application.
[0059] Figure 6A block diagram of a multi-chip integrated layout design device provided in an embodiment of the present application. DETAILED DESCRIPTION
[0060] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.
[0061] Therefore, the following detailed description of the embodiments of the present application provided in the accompanying drawings is not intended to limit the scope of the present application for which protection is sought, but merely represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.
[0062] like Figure 1 As shown, an embodiment of the present application provides an electronic device, wherein the electronic device may include a memory, a processor, and a multi-chip integrated layout design device.
[0063] In detail, the memory and the processor are electrically connected directly or indirectly to achieve data transmission or interaction. For example, the memory and the processor can be electrically connected through one or more communication buses or signal lines. The multi-chip integrated layout design device includes at least one software function module stored in the memory in the form of software or firmware. The processor is used to execute an executable computer program stored in the memory, for example, the software function module and computer program included in the multi-chip integrated layout design device, so as to implement the multi-chip integrated layout design method provided in the embodiment of the present application.
[0064] Optionally, the memory may be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), etc.
[0065] Furthermore, the processor may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), a system on chip (SoC), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.
[0066] Understandably, Figure 1 The structure shown is for illustration only, and the electronic device may also include Figure 1 More or fewer components as shown, or with Figure 1 The different configurations shown, for example, may also include a communication unit for exchanging information with other devices (such as other electronic devices, etc.).
[0067] Combination Figure 2 The embodiment of the present application also provides a multi-chip integrated layout design method applicable to the above electronic device. The method steps defined in the process related to the multi-chip integrated layout design method can be implemented by the electronic device.
[0068] The following will Figure 2 The specific process shown is explained in detail.
[0069] Step S110, determining chip model data and chip layout optimization data of the integrated chip to be optimized.
[0070] In an embodiment of the present application, the electronic device can determine the chip model data and chip layout optimization data of the integrated chip to be optimized. Among them, the integrated chip to be optimized is an integrated chip that stacks multiple sub-chips and / or multiple devices in multiple cavities on the front and back of the substrate (exemplarily, the layout can be a thermal layer, a cover plate, a frame, a chip, a substrate, a chip, a frame, and a cover plate; the thickness of the substrate can be 2.8mm to ensure the rigidity requirements of the substrate, and at the same time, long leads are avoided on the substrate. If it cannot be avoided, welding or bonding positions are left at intervals; the layout of components on the substrate is as symmetrical as possible, and the flip-chip FPGA chip is the heaviest device in the assembly. The FPGA chip is supported by chip bumps, and the number of bumps is 2700+. In order to ensure reliability, epoxy resin can be used for bottom filling reinforcement), and the chip model data is based on three-dimensional modeling of the integrated chip to be optimized (the specific modeling software is not limited, such as Cadence Virtuoso, an integrated circuit design platform, is widely used in the design, verification and implementation of analog, digital and mixed-signal circuits. It provides a complete set of tools to support a complete workflow from the initial concept of circuit design to the final layout implementation) to form, the chip layout optimization data is formed based on the simulation of the chip model data, the chip layout optimization data is used to reflect the performance problems of the integrated chip to be optimized, and the performance problems include at least one of problems related to airtightness (such as poor airtightness), problems related to chip volume (such as the volume can be reduced), problems related to structural stability (such as unstable connection), and problems related to heat dissipation (such as poor heat dissipation effect).
[0071] Step S120 , performing feature extraction processing on the chip model data and the chip layout optimization data respectively, and outputting a chip layout vector corresponding to the chip model data and a chip layout optimization vector corresponding to the chip layout optimization data.
[0072] In an embodiment of the present application, the electronic device can perform feature extraction processing on the chip model data and the chip layout optimization data, respectively, and output a chip layout vector corresponding to the chip model data and a chip layout optimization vector corresponding to the chip layout optimization data. In other words, feature extraction processing is performed on the chip model data to obtain corresponding features, which are represented in the form of vectors, that is, the corresponding chip layout vectors are obtained. Feature extraction processing is performed on the chip layout optimization data to obtain corresponding features, which are represented in the form of vectors, that is, the corresponding chip layout optimization vectors are obtained.
[0073] Step S130 , performing multi-level first feature mining processing on the chip layout vector, and outputting a deep chip layout vector of the integrated chip to be optimized at multiple levels reflected by the chip model data.
[0074] In the embodiment of the present application, after outputting the chip layout vector, the electronic device may perform multi-level first feature mining processing on the chip layout vector, and output the deep chip layout vector of the integrated chip to be optimized at multiple levels reflected by the chip model data. In other words, multiple vectors of different depths may be obtained, with different emphases when performing information representation.
[0075] Step S140 , combining the deep chip layout vectors at the multiple levels, performing multi-level second feature mining processing on the chip layout optimization vector to obtain the optimized chip layout vectors at the multiple levels.
[0076] In an embodiment of the present application, after outputting the deep chip layout vector and the chip layout optimization vector, the electronic device can combine the deep chip layout vectors at the multiple levels, perform multi-level second feature mining processing on the chip layout optimization vector, and obtain optimized chip layout vectors at multiple levels. In other words, multiple vectors of different depths can be obtained to have different emphases when performing information characterization. Among them, the optimized chip layout vector is used to reflect the feature information of the integrated chip to be optimized in two dimensions of the chip model data and the chip layout optimization data, that is, to reflect the original layout characteristics of the chip and the performance characteristics that need to be optimized.
[0077] Step S150 , performing feature restoration processing on the optimized chip layout vectors at the multiple levels to form chip model optimization data that incorporates the chip layout feature information of the integrated chip to be optimized and the layout optimization feature information of the chip layout optimization data.
[0078] In an embodiment of the present application, after outputting the optimized chip layout vector, the electronic device may perform feature restoration processing on the optimized chip layout vectors at the multiple levels to form chip model optimization data that integrates the chip layout feature information of the integrated chip to be optimized and the layout optimization feature information of the chip layout optimization data. The chip model optimization data is used to reflect the layout of the optimized integrated chip, such as the optimized integrated chip has better air tightness, smaller volume, better structural stability, better heat dissipation effect, etc. Exemplarily, the optimized chip layout vector at the last level of the optimized chip layout vectors at the multiple levels may be subjected to feature restoration processing to output the corresponding chip model optimization data. Specifically, the optimized chip layout vector at the last level may be mapped to the original data space (for example, it may be implemented by a corresponding decoding neural network, and correspondingly, in step S120, the encoding neural network may be used to perform feature extraction processing on the data, such as convolution operation and activation processing, etc.), to obtain model data similar to the chip model data, that is, to obtain the chip model optimization data, such as the solder column height, solder column diameter, solder pad height, thermal layer material, etc. in the chip model data are adjusted accordingly, thereby obtaining the chip model optimization data.
[0079] Based on the above content, since the chip layout optimization vector will be subjected to multi-level second feature mining processing in combination with the deep chip layout vectors at multiple levels, the optimized chip layout vectors at multiple levels obtained can reflect the feature information of the integrated chip to be optimized in two dimensions of chip model data and chip layout optimization data. That is to say, it can not only characterize the original chip layout characteristics, but also reflect the characteristics of the performance problems that need to be optimized, so that the basis for the feature restoration processing of the optimized chip layout vector can be more reliable and sufficient. Therefore, the restored chip model optimization data can achieve the optimization of performance problems while ensuring the matching of the original chip layout characteristics, and no manual iteration and other operations are required during the optimization process, so that the problems of low efficiency and high cost of multi-chip integrated layout design existing in the prior art can be improved.
[0080] It should be noted that for the above step S130, the specific manner of performing multi-level first feature mining processing on the chip layout vector is not limited and can be selected according to actual needs.
[0081] For example, in an alternative implementation, in order to ensure that the obtained deep chip layout vector has better feature representation capability, the above step S130 may further include the following contents:
[0082] The chip layout vector may be subjected to multi-level first feature mining processing by using multiple intermediate feature output units in a deep feature mining model formed by optimization operations (i.e., iterative training of the network is performed by using corresponding training data, and the training data may be referred to as samples), and a deep chip layout vector of the integrated chip to be optimized at multiple levels reflected by the chip model data may be output;
[0083] wherein, for the first feature mining process of the first level performed on the chip layout vector, the first intermediate feature output unit among the plurality of intermediate feature output units is used to perform the first feature mining process of the chip layout vector at the first level, and the deep chip layout vector of the integrated chip to be optimized at the first level reflected by the chip model data is output;
[0084] Furthermore, for the first feature mining process of the ath level performed on the chip layout vector, the ath intermediate feature output unit among the multiple intermediate feature output units is used to perform the first feature mining process of the ath level on the deep chip layout vector of the bth level, and the deep chip layout vector of the integrated chip to be optimized at the ath level reflected by the chip model data is output, ab=1, and a is greater than 1, b is greater than or equal to 1, and a and b are both integers; that is, starting from the second intermediate feature output unit, the input of the intermediate feature output unit is the output of the previous intermediate feature output unit, and the output is used as the input of the next intermediate feature output unit. Therefore, in fact, multiple intermediate feature output units belong to a cascade connection, such as Figure 3 shown.
[0085] It is understandable that, in an alternative implementation, in order to further improve the feature characterization capability of the obtained deep chip layout vector, the intermediate feature output unit may be a first gradient optimization unit, and the first gradient optimization unit may include a gradient optimization relationship (which may be a function for gradient optimization, i.e., a corresponding vector or feature mapping may be implemented based on the function). Based on this, the step of using multiple intermediate feature output units in the deep feature mining model formed by the optimization operation to perform multi-level first feature mining processing on the chip layout vector and outputting the deep chip layout vector of the integrated chip to be optimized at multiple levels reflected by the chip model data may further include the following specific implementation process:
[0086] The chip layout vector can be subjected to relationship mapping processing by utilizing the gradient optimization relationship in each first gradient optimization unit (i.e., the intermediate feature output unit) in the multiple intermediate feature output units to implement multi-level first feature mining processing, thereby outputting the deep chip layout vector of the integrated chip to be optimized at multiple levels reflected by the chip model data;
[0087] Among them, each of the first gradient optimization units includes multiple convolution sub-units. For any of the first gradient optimization units, after the input vector of the first gradient optimization unit is convolved by the multiple convolution sub-units (exemplarily, the number of convolution sub-units in every two first gradient optimization units can be the same, but the corresponding convolution kernel sizes are different), the obtained convolution vector and the input vector will be vector aggregated (such as performing a cascade combination of vectors, that is, concatenating the convolution vector and the input vector, such as {convolution vector, input vector}), and the result of the vector aggregation processing will be used as the output vector of the first gradient optimization unit; exemplarily, after the first first gradient optimization unit performs a convolution operation on the chip layout vector, the obtained convolution vector and the chip layout vector can be vector aggregated to obtain a deep chip layout vector at the first level, which is also the output vector of the first first gradient optimization unit, such as Figure 4 As shown, the deep chip layout vector at the second level (ie, the output vector of the second first gradient optimization unit) and the deep chip layout vector at the third level can be obtained by the same token.
[0088] It should be noted that for the above step S140, the specific manner of performing multi-level second feature mining processing on the chip layout optimization vector in combination with the deep chip layout vectors at the multiple levels is not limited and can be selected according to actual needs.
[0089] For example, in an alternative implementation, in order to make the obtained optimized chip layout vector have better feature characterization capability, the above step S140 may further include the following contents:
[0090] The optimized feature mining model formed by the optimization operation (i.e., iterative training of the network using the corresponding training data) can be used to combine the deep chip layout vectors at the multiple levels to perform multi-level second feature mining processing on the chip layout optimization vectors (that is, in the process of performing feature mining on the chip layout optimization vectors, it is necessary to integrate the deep chip layout vectors so that the information represented by the mining results can be richer and more comprehensive), and obtain the optimized chip layout vectors at the multiple levels; wherein the optimized feature mining model includes a plurality of second gradient optimization units connected in cascade (e.g., the output of the first second gradient optimization unit is connected to the input of the second second gradient optimization unit, and the output of the second second gradient optimization unit is connected to the input of the third second gradient optimization unit), and for every two adjacent second gradient optimization units, a focused feature mining unit is configured between the two second gradient optimization units (e.g., a focused feature mining unit is configured between the first second gradient optimization unit and the second second gradient optimization unit A feature mining unit, a focused feature mining unit is configured between the second second gradient optimization unit and the third second gradient optimization unit), for every two adjacent second gradient optimization units, a focused feature mining unit is configured between the two second gradient optimization units (such as a focused feature mining unit is configured between the first second gradient optimization unit and the second second gradient optimization unit, and a focused feature mining unit is configured between the second second gradient optimization unit and the third second gradient optimization unit), so that the output vector of the previous second gradient optimization unit is processed by the focused feature mining unit (that is, the focused features and significant features are mined out, so that the feature characterization ability of the result is better), and then output to the next second gradient optimization unit for corresponding processing, the number of the second gradient optimization units is the same as the number of the first gradient optimization units in the deep feature mining model, so that the deep chip layout vectors at multiple levels output by the deep feature mining model can be fused respectively.
[0091] It can be understood that, in an alternative implementation, the step of using the optimized feature mining model formed by the optimization operation, combining the deep chip layout vectors at multiple levels, performing multi-level second feature mining processing on the chip layout optimization vector, and obtaining the optimized chip layout vectors at multiple levels may further include the following specific implementation process:
[0092] First, for the second feature mining process of the first level performed on the chip layout optimization vector, a first second gradient optimization unit among the plurality of second gradient optimization units may be used to perform the second feature mining process of the first level on the chip layout optimization vector, and output a first level intermediate chip layout optimization vector of the chip layout optimization data after the second feature mining process of the first level;
[0093] Secondly, after the chip layout optimization vector is subjected to the second feature mining processing of the cth level, the cth focused feature mining unit in the optimization feature mining model is used to perform vector aggregation processing on the cth level intermediate chip layout optimization vector after the second feature mining processing of the cth level and the deep chip layout vector output by the cth first gradient optimization unit in the deep feature mining model, and the corresponding cth chip layout aggregation vector is output, wherein c is greater than or equal to 1 and is an integer, that is, starting from the first second gradient optimization unit, after the intermediate chip layout optimization vector of the corresponding level is obtained by using the second gradient optimization unit, the corresponding focused feature mining unit can be used to perform vector aggregation processing on the intermediate chip layout optimization vector and the deep chip layout vector of the corresponding level, and the corresponding chip layout aggregation vector is output; for example, after the first second gradient optimization unit performs the second feature mining processing of the first level, the first focused feature mining unit can be used to perform vector aggregation processing on the first level intermediate chip layout optimization vector and the deep chip layout vector output by the first first gradient optimization unit, and the corresponding first chip layout aggregation vector is output;
[0094] Then, in the process of performing the second feature mining processing of the dth level on the chip layout optimization vector, the dth second gradient optimization unit among the multiple second gradient optimization units is used to perform the second feature mining processing of the dth level on the chip layout optimization vector and the eth chip layout aggregation vector output by the eth focused feature mining unit, and output the dth level intermediate chip layout optimization vector after the second feature mining processing of the dth level of the chip layout optimization data, wherein d is greater than 1, de=1, and d and e are both integers; that is, starting from the second second gradient optimization unit, the second gradient optimization unit will perform feature mining on the chip layout aggregation vector output by the previous focused feature mining unit and the chip layout optimization vector, thereby outputting the intermediate chip layout optimization vector of the corresponding level, such as Figure 5 As shown;
[0095] Finally, after the second feature mining process of the last level is performed, the output intermediate chip layout optimization vectors of multiple levels are used as the optimized chip layout vectors at the multiple levels. That is to say, the intermediate chip layout optimization vectors of one level can be used as the optimized chip layout vectors at the corresponding level, or the intermediate chip layout optimization vectors of multiple levels can be screened and processed to obtain the optimized chip layout vectors at multiple levels.
[0096] It can be understood that, in an alternative implementation, in order to further improve the feature characterization capability of the determined optimized chip layout vector, before the step of performing the first-level second feature mining processing on the chip layout optimization vector, performing the first-level second feature mining processing on the chip layout optimization vector using the first second gradient optimization unit among the plurality of second gradient optimization units, and outputting the first-level intermediate chip layout optimization vector of the chip layout optimization data after the first-level second feature mining processing, the multi-chip integrated layout design method further includes the following steps:
[0097] First, a predetermined random vector determination rule can be used to output a random semantic vector (such as blind noise or white noise) corresponding to the optimized feature mining model, that is, a random vector can be generated to improve the generalization ability; the random semantic vector and the chip layout optimization vector are marked as vectors to be processed and loaded into the optimized feature mining model.
[0098] Based on this, the step of performing a first-level second feature mining process on the chip layout optimization vector, using a first second gradient optimization unit among the plurality of second gradient optimization units, performing a first-level second feature mining process on the chip layout optimization vector, and outputting a first-level intermediate chip layout optimization vector of the chip layout optimization data after the first-level second feature mining process, may further include the following contents:
[0099] For the first-level second feature mining processing of the chip layout optimization vector, the first second gradient optimization unit among the multiple second gradient optimization units is used to perform the first-level second feature mining processing on the chip layout optimization vector and the random semantic vector, and the first-level intermediate chip layout optimization vector of the chip layout optimization data after the first-level second feature mining processing is output; illustratively, the chip layout optimization vector and the random semantic vector can be spliced, and then the first second gradient optimization unit can be used to perform convolution processing on the spliced vector, so that the result of the convolution processing can be used as the first-level intermediate chip layout optimization vector.
[0100] It can be understood that, in an alternative implementation, in order to improve the reliability of the vector aggregation processing performed by the focused feature mining unit, so that the feature characterization capability of the obtained chip layout aggregation vector can be better, the intermediate chip layout optimization vector of the cth level and the deep chip layout vector output by the cth first gradient optimization unit have the same number of dimensions; based on this, after the chip layout optimization vector is subjected to the second feature mining processing of the cth level, the cth focused feature mining unit in the optimized feature mining model is used to perform vector aggregation processing on the intermediate chip layout optimization vector of the cth level after the second feature mining processing of the cth level and the deep chip layout vector output by the cth first gradient optimization unit in the deep feature mining model, and the step of outputting the corresponding cth chip layout aggregation vector can further include the following specific implementation process:
[0101] First, the c-th focused feature mining unit in the optimized feature mining model is used to perform vector cascade combination processing on the c-th level intermediate chip layout optimization vector after the c-th level second feature mining processing and the deep chip layout vector output by the c-th first gradient optimization unit in the deep feature mining model to form a corresponding chip layout cascade vector; for example, the first focused feature mining unit can be used to perform vector cascade combination processing on the first level intermediate chip layout optimization vector and the first level deep chip layout vector (since the number of dimensions is the same, they can be spliced to achieve cascade combination);
[0102] Secondly, the chip layout cascade vector may be compressed to output a corresponding chip layout compression vector. For example, a Pooling operation may be performed to achieve vector compression, thereby reducing the amount of vector data while retaining important information.
[0103] Then, the chip layout compression vector may be subjected to a convolution operation (the size and number of the specific convolution kernels may be configured according to actual needs, or adjusted during the corresponding network optimization operation), and a focusing weight parameter distribution between the intermediate chip layout optimization vector of the c-th level and the deep chip layout vector corresponding to the c-th first gradient optimization unit may be output;
[0104] Finally, the chip layout cascade vectors may be weighted according to the focus weight parameter distribution, and the corresponding c-th chip layout aggregation vector may be output; illustratively, the focus weight parameter distribution and the chip layout cascade vector (such as the result of vector cascade combination processing of the first-level intermediate chip layout optimization vector and the first-level deep chip layout vector) may be multiplied to obtain the first chip layout aggregation vector.
[0105] It can be understood that, in an alternative embodiment, in order to effectively reconstruct or restore the input vector while retaining important information to improve the generalization ability of the model, the multiple second gradient optimization units include at least one second gradient optimization unit for convolution operation and at least one second gradient optimization unit for transposed convolution operation, wherein the number of the at least one second gradient optimization unit for convolution operation is equal to the number of the at least one second gradient optimization unit for transposed convolution operation, and the at least one second gradient optimization unit for transposed convolution operation is cascaded behind the position of the at least one second gradient optimization unit for convolution operation. That is to say, in the process of processing by the multiple second gradient optimization units, a convolution operation is first performed to reduce the amount of data while retaining important information, and then a transposed convolution operation is performed so that the size of the vector can be increased to achieve reliable restoration of the vector.
[0106] It should be further explained that the specific execution of steps S120-S150 can be implemented using a chip layout optimization network (a neural network). The chip layout optimization network includes a first network part and a second network part, the first network part includes a first feature extraction model and an optimization feature mining model, and the second network part includes a second feature extraction model and a deep feature mining model. In addition, the chip layout optimization network is formed through a network optimization operation, wherein the step of forming the chip layout optimization network through a network optimization operation may include the following:
[0107] First, the training chip model data (corresponding to the aforementioned chip model data) in the training chip layout data combination may be loaded into the second network part, and the training chip model data may be subjected to feature extraction processing using the second feature extraction model in the second network part to output a training chip layout vector;
[0108] Secondly, the deep feature mining model in the second network part can be used to perform multi-level first feature mining processing on the training chip layout vector, and output the training deep chip layout vector of the training integrated chip to be optimized at multiple levels reflected by the training chip model data;
[0109] Then, the training chip layout optimization data (corresponding to the aforementioned chip layout optimization data) in the training chip layout data combination may be loaded into the first network part, and the training chip layout optimization data may be subjected to feature extraction processing using the first feature extraction model in the first network part to output a training chip layout optimization vector;
[0110] Then, using the optimized feature mining model in the first network part, combined with the training deep chip layout vectors at the multiple levels, the training chip layout optimization vectors are subjected to multi-level second feature mining processing, and the corresponding training optimized chip layout vectors at the multiple levels are output;
[0111] Further, a network optimization error of the chip layout optimization network may be determined based on the training deep chip layout vectors at the multiple levels and the training optimized chip layout vectors at the multiple levels (which may be used to reflect the difference between the training deep chip layout vectors at the multiple levels and the training optimized chip layout vectors at the multiple levels);
[0112] Finally, the network parameters in the chip layout optimization network can be optimized and adjusted according to the network optimization error, and the chip layout optimization network after the network optimization operation can be output; illustratively, the network parameters in the chip layout optimization network can be adjusted in the direction of reducing the network optimization error until the number of adjustments reaches a preset number or the current network optimization error is reduced to a preset error or the reduction range of the network optimization error converges, etc.
[0113] It can be understood that, in an alternative implementation, in order to improve the ability of the network optimization error to characterize the difference between the training deep chip layout vector and the training optimized chip layout vector, the step of determining the network optimization error of the chip layout optimization network based on the training deep chip layout vectors at the multiple levels and the training optimized chip layout vectors at the multiple levels may further include the following contents:
[0114] First, according to the training optimized chip layout vectors at the multiple levels and the training chip model data, a first network optimization error of the chip layout optimization network can be determined, that is, the difference between the training optimized chip layout vectors and the training chip model data can be determined;
[0115] Secondly, a second network optimization error of the chip layout optimization network can be determined according to the training depth chip layout vectors at the multiple levels and the training optimization chip layout vectors at the multiple levels, that is, the difference between the training depth chip layout vector and the training optimization chip layout vector at the multiple levels is determined;
[0116] Finally, the first network optimization error and the second network optimization error can be fused to output the network optimization error of the chip layout optimization network; illustratively, the first network optimization error and the second network optimization error can be added or weighted summed, and the result of the calculation is used as the network optimization error of the chip layout optimization network.
[0117] It can be understood that, in an alternative implementation, the step of determining the first network optimization error of the chip layout optimization network based on the training optimized chip layout vectors at the multiple levels and the training chip model data may include the following implementation process:
[0118] First, the training optimized chip layout vectors at the multiple levels may be subjected to feature restoration processing (e.g., feature restoration processing may be performed on the training optimized chip layout vectors at the last level through a corresponding decoding neural network), to form training chip model optimization data that incorporates the chip layout feature information of the training integrated chip to be optimized and the layout optimization feature information of the training chip layout optimization data;
[0119] Secondly, the training chip model optimization data and the reference chip model optimization data corresponding to the training chip model data can be processed for difference determination (for example, the edit distance between the training chip model optimization data and the reference chip model optimization data can be directly calculated, or feature extraction processing can be performed, and then the cosine distance between the vectors obtained by the feature extraction processing is calculated, etc.), and the first network optimization error of the chip layout optimization network is output, wherein the reference chip model optimization data is used to describe the integrated chip after optimization for the performance problem existing in the training integrated chip to be optimized reflected by the training chip model data (which can be understood as a label, such as formed by adjusting the training chip model data based on the training chip layout optimization data by corresponding experts or other reliable neural networks).
[0120] It can be understood that, in an alternative implementation, the step of determining the second network optimization error of the chip layout optimization network based on the training depth chip layout vectors at the multiple levels and the training optimization chip layout vectors at the multiple levels may include the following implementation process:
[0121] First, the training deep chip layout vectors at the multiple levels may be subjected to feature restoration processing (e.g., by decoding a neural network to perform feature restoration processing on the training deep chip layout vectors at the last level) to form corresponding training chip model restoration data;
[0122] Secondly, the training optimized chip layout vectors at the multiple levels may be subjected to feature restoration processing (e.g., by decoding a neural network to perform feature restoration processing on the training optimized chip layout vectors at the last level), to form training chip model optimization data that incorporates the chip layout feature information of the training integrated chip to be optimized and the layout optimization feature information of the training chip layout optimization data;
[0123] Finally, the training chip model restoration data and the training chip model optimization data may be subjected to feature extraction processing respectively, and then the output vectors may be subjected to difference determination processing (such as cosine distance between vectors) to output the second network optimization error of the chip layout optimization network.
[0124] Combination Figure 6 The embodiment of the present application also provides a multi-chip integrated layout design device applicable to the above electronic device. The multi-chip integrated layout design device includes a data determination module, a feature extraction module, a first feature mining module, a second feature mining module and a feature restoration module.
[0125] The data determination module can be used to determine the chip model data and chip layout optimization data of the integrated chip to be optimized, wherein the integrated chip to be optimized is an integrated chip with multiple cavities stacked on the front and back sides of a substrate with multiple sub-chips and / or multiple devices, the chip model data is formed based on three-dimensional modeling of the integrated chip to be optimized, the chip layout optimization data is formed based on simulation of the chip model data, and the chip layout optimization data is used to reflect the performance problems of the integrated chip to be optimized, and the performance problems include at least one of problems related to air tightness, problems related to chip volume, problems related to structural stability, and problems related to heat dissipation. In an embodiment of the present application, the data determination module can be used to perform Figure 2 As shown in step S110, the relevant contents of the data determination module can refer to the above description of step S110.
[0126] The feature extraction module can be used to perform feature extraction processing on the chip model data and the chip layout optimization data, and output a chip layout vector corresponding to the chip model data and a chip layout optimization vector corresponding to the chip layout optimization data. Figure 2As shown in step S120, the relevant contents of the feature extraction module can refer to the above description of step S120.
[0127] The first feature mining module can be used to perform multi-level first feature mining processing on the chip layout vector, and output the deep chip layout vector of the integrated chip to be optimized at multiple levels reflected by the chip model data. Figure 2 As shown in step S130, the relevant contents of the first feature mining module can refer to the above description of step S130.
[0128] The second feature mining module can be used to combine the deep chip layout vectors at multiple levels, perform multi-level second feature mining on the chip layout optimization vector, and obtain optimized chip layout vectors at multiple levels, wherein the optimized chip layout vector is used to reflect the feature information of the integrated chip to be optimized in two dimensions of the chip model data and the chip layout optimization data. In the embodiment of the present application, the second feature mining module can be used to perform Figure 2 As shown in step S140, the relevant contents of the second feature mining module can refer to the above description of step S140.
[0129] The feature restoration module can be used to perform feature restoration processing on the optimized chip layout vectors at the multiple levels to form chip model optimization data that incorporates the chip layout feature information of the integrated chip to be optimized and the layout optimization feature information of the chip layout optimization data, wherein the chip model optimization data is used to reflect the layout of the optimized integrated chip. Figure 2 As shown in step S150, the relevant contents of the feature restoration module can refer to the above description of step S150.
[0130] In summary, the multi-chip integrated layout design method provided by the present application first determines chip model data and chip layout optimization data; secondly, performs feature extraction processing on the chip model data and chip layout optimization data respectively, and outputs a chip layout vector and a chip layout optimization vector; then, performs multi-level first feature mining processing on the chip layout vector, and outputs deep chip layout vectors at multiple levels; further, in combination with the deep chip layout vectors at multiple levels, performs multi-level second feature mining processing on the chip layout optimization vector, and obtains optimized chip layout vectors at multiple levels; finally, performs feature restoration processing on the optimized chip layout vectors at multiple levels to form chip model optimization data. Based on the above content, since the chip layout optimization vector will be subjected to multi-level second feature mining processing in combination with the deep chip layout vectors at multiple levels, the optimized chip layout vectors at multiple levels obtained can reflect the feature information of the integrated chip to be optimized in two dimensions of chip model data and chip layout optimization data. That is to say, it can not only characterize the original chip layout characteristics, but also reflect the characteristics of the performance problems that need to be optimized, so that the basis for the feature restoration processing of the optimized chip layout vector can be more reliable and sufficient. Therefore, the restored chip model optimization data can achieve the optimization of performance problems while ensuring the matching of the original chip layout characteristics, and no manual iteration and other operations are required during the optimization process, so that the problems of low efficiency and high cost of multi-chip integrated layout design in the prior art can be improved, that is, it has the advantages of reliability, efficiency and cost of solution optimization.
[0131] In several embodiments provided in the embodiments of the present application, it should be understood that the disclosed device and method can also be implemented in other ways. The device and method embodiments described above are merely schematic, for example, the flowcharts and block diagrams in the accompanying drawings show the possible architecture, functions and operations of the device, method and computer program product according to multiple embodiments of the present application. 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 a specified logical function. It should also be noted that in some alternative implementations, 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 block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart 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.
[0132] In addition, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.
[0133] If the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially 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, which is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, electronic device, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned 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 code. It should be noted that in this article, the term "include", "include" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements includes not only those elements, but also includes other elements that are not explicitly listed, or also includes elements inherent to such process, method, article or device. Without more constraints, an element defined by the phrase "comprising a..." does not exclude the existence of other identical elements in the process, method, article or apparatus comprising the element.
[0134] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.
Claims
1. A multi-chip integrated layout design method, characterized in that: include: Determine chip model data and chip layout optimization data of the integrated chip to be optimized, wherein the integrated chip to be optimized is an integrated chip in which multiple sub-chips and / or multiple devices are stacked on multiple cavities on the front and back sides of a substrate, the chip model data is formed based on three-dimensional modeling of the integrated chip to be optimized, the chip layout optimization data is formed based on simulation of the chip model data, and the chip layout optimization data is used to reflect performance problems existing in the integrated chip to be optimized, and the performance problems include at least one of problems related to air tightness, problems related to chip volume, problems related to structural stability, and problems related to heat dissipation; Performing feature extraction processing on the chip model data and the chip layout optimization data respectively, and outputting a chip layout vector corresponding to the chip model data and a chip layout optimization vector corresponding to the chip layout optimization data; Perform multi-level first feature mining processing on the chip layout vector, and output a deep chip layout vector of the integrated chip to be optimized at multiple levels reflected by the chip model data; Combined with the deep chip layout vectors at the multiple levels, the chip layout optimization vector is subjected to multi-level second feature mining processing to obtain optimized chip layout vectors at the multiple levels, wherein the optimized chip layout vector is used to reflect the feature information of the integrated chip to be optimized in two dimensions of the chip model data and the chip layout optimization data; Performing feature restoration processing on the optimized chip layout vectors at the multiple levels to form chip model optimization data that incorporates the chip layout feature information of the integrated chip to be optimized and the layout optimization feature information of the chip layout optimization data, wherein the chip model optimization data is used to reflect the layout of the optimized integrated chip, and the chip model optimization data is obtained by mapping the optimized chip layout vector at the last level among the optimized chip layout vectors at the multiple levels to the original data space through a decoding neural network; The step of performing multi-level first feature mining processing on the chip layout vector and outputting the deep chip layout vector of the integrated chip to be optimized at multiple levels reflected by the chip model data comprises: performing multi-level first feature mining processing on the chip layout vector using multiple intermediate feature output units in the deep feature mining model formed by the optimization operation, and outputting the deep chip layout vector of the integrated chip to be optimized at multiple levels reflected by the chip model data; wherein, for the first feature mining processing of the first level performed on the chip layout vector, the first intermediate feature output unit among the multiple intermediate feature output units is used to perform the first feature mining processing on the chip layout vector at the first level, and the deep chip layout vector of the integrated chip to be optimized at the first level reflected by the chip model data is output; and, for the first feature mining processing of the ath level performed on the chip layout vector, the ath intermediate feature output unit among the multiple intermediate feature output units is used to perform the ath first feature mining processing on the deep chip layout vector of the bth level, and the deep chip layout vector of the integrated chip to be optimized at the ath level reflected by the chip model data is output, ab=1, and a is greater than 1, and b is greater than or equal to 1; Wherein, the intermediate feature output unit is a first gradient optimization unit, and the first gradient optimization unit includes a gradient optimization relationship; the step of using multiple intermediate feature output units in the deep feature mining model formed by the optimization operation to perform multi-level first feature mining processing on the chip layout vector, and outputting the deep chip layout vector of the integrated chip to be optimized at multiple levels reflected by the chip model data, includes: using the gradient optimization relationship in each first gradient optimization unit in the multiple intermediate feature output units, performing relationship mapping processing on the chip layout vector to achieve multi-level first feature mining processing, thereby outputting the deep chip layout vector of the integrated chip to be optimized at multiple levels reflected by the chip model data; wherein each of the first gradient optimization units includes multiple convolution sub-units, and for any one of the first gradient optimization units, after the input vector of the first gradient optimization unit is convolved by the multiple convolution sub-units, the obtained convolution vector will be subjected to vector aggregation processing with the input vector, and the result of the vector aggregation processing will be used as the output vector of the first gradient optimization unit; Among them, the step of combining the deep chip layout vectors at the multiple levels, performing multi-level second feature mining processing on the chip layout optimization vector, and obtaining the optimized chip layout vectors at the multiple levels, includes: using the optimized feature mining model formed by the optimization operation, combining the deep chip layout vectors at the multiple levels, performing multi-level second feature mining processing on the chip layout optimization vector, and obtaining the optimized chip layout vectors at the multiple levels; wherein the optimized feature mining model includes a plurality of second gradient optimization units connected in cascade, for each two adjacent second gradient optimization units, a focused feature mining unit is configured between the two second gradient optimization units, so that the output vector of the previous second gradient optimization unit is output to the next second gradient optimization unit for corresponding processing after being processed by the focused feature mining unit, and the number of the second gradient optimization units is the same as the number of the first gradient optimization units in the deep feature mining model.
2. The multi-chip integrated layout design method according to claim 1, characterized in that: The step of using the optimized feature mining model formed by the optimization operation, combining the deep chip layout vectors at multiple levels, performing multi-level second feature mining processing on the chip layout optimization vector, and obtaining the optimized chip layout vectors at multiple levels includes: For the chip layout optimization vector, a first second gradient optimization unit among the plurality of second gradient optimization units is used to perform a first-level second feature mining process on the chip layout optimization vector, and a first-level intermediate chip layout optimization vector of the chip layout optimization data after the first-level second feature mining process is output; After the chip layout optimization vector is subjected to the second feature mining processing of the cth level, the cth focused feature mining unit in the optimization feature mining model is used to perform vector aggregation processing on the cth level intermediate chip layout optimization vector after the second feature mining processing of the cth level and the deep chip layout vector output by the cth first gradient optimization unit in the deep feature mining model, and the corresponding cth chip layout aggregation vector is output, wherein c is greater than or equal to 1; In the process of performing the second feature mining processing of the dth level on the chip layout optimization vector, using the dth second gradient optimization unit among the multiple second gradient optimization units, the chip layout optimization vector and the eth chip layout aggregation vector output by the eth focused feature mining unit are subjected to the second feature mining processing of the dth level, and the intermediate chip layout optimization vector of the dth level after the second feature mining processing of the dth level of the chip layout optimization data is output, wherein d is greater than 1, and de=1; After the second feature mining process of the last level is performed, the output intermediate chip layout optimization vectors of multiple levels are used as the optimized chip layout vectors of the multiple levels.
3. The multi-chip integrated layout design method according to claim 2, characterized in that: Before the step of performing a first-level second feature mining process on the chip layout optimization vector, performing a first-level second feature mining process on the chip layout optimization vector using a first second gradient optimization unit among the plurality of second gradient optimization units, and outputting a first-level intermediate chip layout optimization vector of the chip layout optimization data after the first-level second feature mining process, the multi-chip integrated layout design method further includes: Outputting a random semantic vector corresponding to the optimized feature mining model using a predetermined random vector determination rule; Marking the random semantic vector and the chip layout optimization vector as vectors to be processed, and loading them into the optimization feature mining model; The step of performing a first-level second feature mining process on the chip layout optimization vector, using a first second gradient optimization unit among the plurality of second gradient optimization units, performing a first-level second feature mining process on the chip layout optimization vector, and outputting a first-level intermediate chip layout optimization vector of the chip layout optimization data after the first-level second feature mining process, comprises: With respect to the first-level second feature mining processing of the chip layout optimization vector, the first second gradient optimization unit among the plurality of second gradient optimization units is used to perform the first-level second feature mining processing on the chip layout optimization vector and the random semantic vector, and the first-level intermediate chip layout optimization vector of the chip layout optimization data after the first-level second feature mining processing is output.
4. The multi-chip integrated layout design method according to any one of claims 1 to 3, characterized in that: The multi-chip integrated layout design method is implemented based on a chip layout optimization network formed through a network optimization operation, wherein the chip layout optimization network includes a first network part and a second network part, wherein the first network part includes a first feature extraction model and an optimization feature mining model, and the second network part includes a second feature extraction model and a deep feature mining model; The step of forming the chip layout optimization network through network optimization operation includes: Loading the training chip model data in the training chip layout data combination to load into the second network part, performing feature extraction processing on the training chip model data using a second feature extraction model in the second network part, and outputting a training chip layout vector; Using the deep feature mining model in the second network part, the training chip layout vector is subjected to multi-level first feature mining processing, and the training deep chip layout vector of the training integrated chip to be optimized at multiple levels reflected by the training chip model data is output; Loading the training chip layout optimization data in the training chip layout data combination to load into the first network part, performing feature extraction processing on the training chip layout optimization data using a first feature extraction model in the first network part, and outputting a training chip layout optimization vector; Utilizing the optimized feature mining model in the first network part, combined with the training deep chip layout vectors at the multiple levels, the training chip layout optimization vectors are subjected to multi-level second feature mining processing, and the corresponding training optimized chip layout vectors at the multiple levels are output; Determining a network optimization error of the chip layout optimization network according to the training depth chip layout vectors at the multiple levels and the training optimization chip layout vectors at the multiple levels; According to the network optimization error, the network parameters in the chip layout optimization network are optimized and adjusted, and the chip layout optimization network after the network optimization operation is output.
5. The multi-chip integrated layout design method according to claim 4, characterized in that: The step of determining the network optimization error of the chip layout optimization network according to the training depth chip layout vectors at the multiple levels and the training optimization chip layout vectors at the multiple levels comprises: Determining a first network optimization error of the chip layout optimization network according to the training optimized chip layout vectors at the multiple levels and the training chip model data; Determining a second network optimization error of the chip layout optimization network according to the training depth chip layout vectors at the multiple levels and the training optimization chip layout vectors at the multiple levels; The first network optimization error and the second network optimization error are integrated to output a network optimization error of the chip layout optimization network.
6. The multi-chip integrated layout design method according to claim 5, characterized in that: The step of determining a first network optimization error of the chip layout optimization network based on the training optimized chip layout vectors at the multiple levels and the training chip model data comprises: Performing feature restoration processing on the training optimized chip layout vectors at the multiple levels to form training chip model optimization data that incorporates the chip layout feature information of the training integrated chip to be optimized and the layout optimization feature information of the training chip layout optimization data; The training chip model optimization data and the reference chip model optimization data corresponding to the training chip model data are subjected to difference determination processing, and a first network optimization error of the chip layout optimization network is output, wherein the reference chip model optimization data is used to describe an integrated chip after optimization for performance problems existing in the training integrated chip to be optimized as reflected by the training chip model data.
7. The multi-chip integrated layout design method according to claim 5, characterized in that: The step of determining a second network optimization error of the chip layout optimization network according to the training depth chip layout vectors at the multiple levels and the training optimization chip layout vectors at the multiple levels comprises: Performing feature restoration processing on the training deep chip layout vectors at the multiple levels to form corresponding training chip model restoration data; Performing feature restoration processing on the training optimized chip layout vectors at the multiple levels to form training chip model optimization data that incorporates the chip layout feature information of the training integrated chip to be optimized and the layout optimization feature information of the training chip layout optimization data; After performing feature extraction processing on the training chip model restoration data and the training chip model optimization data respectively, performing difference determination processing on the output vector, and outputting the second network optimization error of the chip layout optimization network.
