CN112989756A - Coarse grain parallel iteration method and device for dynamically applying coupling between layers of integrated circuit - Google Patents
Coarse grain parallel iteration method and device for dynamically applying coupling between layers of integrated circuit Download PDFInfo
- Publication number
- CN112989756A CN112989756A CN202110425242.5A CN202110425242A CN112989756A CN 112989756 A CN112989756 A CN 112989756A CN 202110425242 A CN202110425242 A CN 202110425242A CN 112989756 A CN112989756 A CN 112989756A
- Authority
- CN
- China
- Prior art keywords
- parallel
- layer
- integrated circuit
- computing
- source
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/30—Circuit design
- G06F30/39—Circuit design at the physical level
- G06F30/398—Design verification or optimisation, e.g. using design rule check [DRC], layout versus schematics [LVS] or finite element methods [FEM]
Landscapes
- Engineering & Computer Science (AREA)
- Computer Hardware Design (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Design And Manufacture Of Integrated Circuits (AREA)
Abstract
The invention provides a coarse grain parallel iteration method and a coarse grain parallel iteration device for interlayer coupling dynamic application of an integrated circuit, wherein the coarse grain parallel iteration method comprises the following steps: firstly, dividing iterative computation of interlayer coupling of an integrated circuit into parallel computation particles, and combining the parallel computation particles into parallel computation coarse particles according to CPU weighted time; secondly, distributing a calculation task by taking parallel calculation particles as a unit, performing iteration on current distribution of the source layer in parallel, calculating the influence of other layers on the source layer through a parallel vector Green function during each iteration, accumulating the influence as a source item of the source layer, applying a two-dimensional finite element to the source layer to calculate field distribution of the source layer so as to update the field and current distribution of the source layer, obtaining the change amount of the field of the source layer, and dynamically modifying the range of an action layer of the source layer influenced by other layers; and repeating the iteration until the change amount of the fields of all the layers is smaller than a specified threshold value, and finishing the iteration. According to the method and the device, the electromagnetic response calculation time of the three-dimensional multilayer integrated circuit is linearly reduced along with the parallel process number.
Description
Technical Field
The invention relates to the technical field of integrated circuits, in particular to a coarse grain parallel iteration method and device for interlayer coupling dynamic application of an integrated circuit.
Background
When the integrated circuit works, a high-frequency alternating electromagnetic field can be formed on a multilayer layout of the integrated circuit due to the transmission of high-speed signals, and meanwhile, in order to improve the performance of electronic equipment, reduce the volume and reduce the cost, transistors, other components and circuits are integrated on a small semiconductor substrate. In order to realize more functions, the ultra-large scale integrated circuit has a structure from tens of layers to hundreds of layers, each layer of structure is extremely complex, millions or even tens of millions of transistors are integrated, and the ultra-large scale integrated circuit has a multi-scale structure from a centimeter level to the latest nanometer level at present. In order to ensure that the integrated circuit can normally work and realize the function designed in advance, the power integrity and the signal integrity of the integrated circuit need to be ensured firstly, so that the power integrity and the signal integrity of the integrated circuit with a multi-scale structure of tens of layers and hundreds of layers need to be accurately analyzed by adopting an electromagnetic field analysis method, which is a great problem of the electromagnetic field analysis of the ultra-large scale integrated circuit.
The method comprises the steps of performing electromagnetic field analysis on a three-dimensional large-scale integrated circuit by adopting a traditional method, further calculating the electromagnetic response of the three-dimensional large-scale integrated circuit, generally determining the whole three-dimensional integrated circuit and a limited region outside the integrated circuit as a calculation region after setting a truncation error of a certain region, then performing mesh division on the whole calculation region, calculating the electromagnetic field distribution of the whole calculation region, and further calculating the electromagnetic response of each layer of the integrated circuit, such as the electromagnetic field distribution, the current voltage of a designated port and the like. However, the characteristic dimensions of the via holes, the wires and the like of the integrated circuit are nano-scale, the dimension of the whole integrated circuit is centimeter-scale, the calculation area determined according to the truncation error is decimeter-scale and meter-scale, and hundreds of millions of grids and unknown quantities can be generated by carrying out uniform grid subdivision on the multi-scale space and then analyzing the space electromagnetic radiation of the multi-scale space, so that the hardware (memory) cost and the CPU time cost are overlarge. Therefore, the electromagnetic response of the three-dimensional large-scale integrated circuit can be calculated by adopting a method combining a finite element method and a moment method. In the three-dimensional large-scale integrated circuit area, a finite element method is adopted; in a large-scale area outside the integrated circuit, a moment method is adopted; the finite element method and the moment method are coupled at the interface of the integrated circuit and the external space. Because the moment method only integrates aiming at the interface, a large number of grid units and unknowns can be reduced, but because the scale range of the integrated circuit is from nano-scale to centimeter-scale, the finite element method directly used for solving the integrated circuit can generate a huge sparse matrix, and because the finite element method and the moment method are coupled, the formed coupling matrix is a dense matrix at the interface, the non-zero element number of the whole sparse matrix and the solving complexity of the sparse matrix are greatly increased, and the calculation time is still long.
Disclosure of Invention
Objects of the invention
Based on the problems, the invention provides a coarse grain parallel iteration method and a coarse grain parallel iteration device dynamically applied by interlayer coupling of an integrated circuit, on one hand, the communication among processes is reduced to the maximum extent in the process of iterative computation of interlayer coupling of a multilayer ultra-large scale integrated circuit, the hard disk read-write bottleneck caused by the fact that the peak value of a memory is larger than the available physical memory during the multi-process parallel computation is avoided, the problem of process waiting caused by the fact that the complexity of different computation examples is not equal is perfectly solved, and the parallel computation efficiency is greatly improved; on the other hand, according to the characteristics of interlayer coupling of the integrated circuit, all the calculation particles are combined into two-stage parallel calculation coarse particles with calculation task sets in different ranges, and calculation processes are distributed according to the two-stage parallel calculation coarse particles; finally, based on the attenuation rules of electromagnetic fields and electromagnetic waves in space, when a parallel iteration solving method is designed, the influence of other layers on the source layer is calculated through a parallel vector Green function in each iteration, the influence is accumulated to be used as a source item of the source layer, a two-dimensional finite element is applied to the source layer to calculate the field distribution of the source layer so as to update the field and current distribution of the layer, the change quantity of the source layer field is obtained, the range of an action layer of the source layer influenced by other layers is dynamically modified, and the iteration solving time is greatly accelerated. The present application discloses the following technical solutions.
(II) technical scheme
As a first aspect of the invention, the invention discloses a coarse grain parallel iteration method for the interlayer coupling dynamic application of an integrated circuit, which comprises the following steps:
step S100, dividing iterative computation of coupling between integrated circuit layers into parallel computation particles, combining the parallel computation particles into two-stage parallel computation coarse particles according to weighted CPU time of the parallel computation particles, wherein the first-stage parallel computation coarse particles are used for computing electromagnetic field and current distribution of each source layer, and the first-stage parallel computation coarse particles comprise the steps of updating source items acting on the source layers in a coupling mode by computing influences of other layers on the source layers and computing the electromagnetic field and current distribution of the source layers by utilizing two-dimensional finite elements; the second level of parallel computing coarse grains is used for computing the layer-layer mutual influence of the integrated circuit;
step S200, the large scale integrated circuit is summarizedN+1 layer, each layer numberedWhen considering the second aspect of LSImWhen the current source of the layer is in the current source of the layer, the layer is called the first layermA source layer provided with a secondmActive layer of source layerIs divided bymOthers of the source layerNLayer integrated circuits, i.e.WhereinThe 0 th layer is a bottom layer; setting the number of iterations;
Step S300, ifComputing the first based on the coarse grain of the first stage parallel computingmThe electromagnetic field and current distribution of the source layer are calculated, and the change amount of the electromagnetic field of the layer is calculateddE m And a firstlLayer to layer ofInfluence value of layer-by-layer vector green's functionG lm Maximum value ofG m,maxAnd minimum valueG m,minWherein 0 is less than or equal tom≤N;
Step S400, ifAnd isAnd (5) finishing the iteration, and outputting electromagnetic fields and current distributions of each layer, whereinThe iteration precision is preset;
step S500, ifMerging all the influence values of the dyadic Green function obtained by the parallel calculation of the coarse grains at the first stage, and obtaining the maximum value of the influence values of all the dyadic Green functions of the current iterationG maxAnd minimum valueG minCalculating the effective influence value of the dyadic Green functionHere, thethredsholdA discarding threshold value for the influence of a preset dyadic Green function;
step S600, ifCalculating satisfyOf the conditionG lm Middle distance layermNearest layerl near Will beIs updated to;
Further, in step S100, the formula for calculating the weighted CPU time of the parallel calculation grain is:in the formula:is as followsiThe weighted CPU times of the particles are calculated in parallel,is as followsiThe CPU time of a single calculation of a grain is calculated in parallel,is as followsiThe number of particle executions is calculated in parallel.
Furthermore, the first-stage parallel computing coarse grains and the second-stage parallel computing coarse grains are computing task sets in different ranges, the first-stage parallel computing coarse grains comprise a plurality of second-stage parallel computing coarse grains and a computing task for computing electromagnetic fields and current distribution of an integrated circuit layer by using two-dimensional finite elements, the parallel computing grains belonging to the same stage can be computed in parallel, the first-stage parallel computing grains can be distributed with a plurality of computing processes, and the second-stage parallel computing grains can be distributed with only one computing process.
Further, the step S300 includes:
step S310, calculating coarse grains based on the second-stage parallel calculation, and calculating the second stage by utilizing a dyadic Green functionlLayer to layermInfluence of the Source layer, isG lm Wherein, in the step (A),(ii) a All will beG lm Overlapping to obtain the other layer pairs within the range of the action layermSum of influence of source layers:(ii) a Finishing the second-stage parallel calculation of coarse grains;
step S320, ifiter>0, willG m As a firstmAdditional sources of source layers added to the secondmIn the source item of the source layer;
step S330, for the secondmThe source layer applies two-dimensional finite element to calculate the electromagnetic field distribution so as to update the electromagnetic field and current distribution of the layer, and the change quantity of the electromagnetic field of the layer is calculateddE m ;
Step S340, according tolInfluence value of layer on dyadic Green function of layerG lm CalculatingG lm Maximum value ofG m,maxAnd minimum valueG m,min。
Further, the specific steps of the second-stage parallel computation of coarse grains include:
step S311, calculating the electric field generated by the point current source at the field point, wherein the electric field expression generated by the point current source at the field point is a special analytic expression formed according to the layered special structure of the integrated circuit, and the current sources of the multilayer integrated circuit are layered, that is, the current density distributed on each metal layer of the integrated circuit layout with the complex shape is only equal to that of each metal layer of the integrated circuit layout with the complex shapexAndyis related tozIndependently, the current density distribution is onlyx, yA function of (a);
step S312, taking an electric field expression generated by the point current source at the field point as an integrand of two-dimensional gaussian integration, and calculating a field generated by the surface current source of the simple-shaped polygon at the same position based on a field linear superposition principle, including: the field generated by the current source in the two-dimensional plane S at any point in space can be calculated by the two-dimensional gaussian integral:
wherein the content of the first and second substances,at any point in space for the current source within the two-dimensional plane S (x,y,z) The field that is generated is,is an arbitrary position within the two-dimensional surface S: (u,v) At any point in space (a)x,y,z) The expression of the field that is generated,representing a gaussian integration point corresponding to a two-dimensional gaussian integration in the two-dimensional plane S,p,qrespectively representu,vIn the first directionpA first, aqThe number of the Gaussian integration points is equal to the total number of the points,is the weight factor corresponding to the gaussian integral point;
step S313, calculating fields generated by the current on the simple-shaped polygon at different positions of other layers of the integrated circuit, and determining the fields generated by the current on the simple-shaped polygon divided on the layout of other layers of the integrated circuit based on the linear superposition principle of the fields;
step S314, determining the second step based on the linear superposition principle of the fieldlIs layered onmThe electric field generated by the source layer.
Further, influence values according to the dyadic Green function in the iteration processGDetermines a layer that can be ignored, adaptively adjustsmExtent of influence of source layer by other layers adjacent thereto。
The other side fully discloses a coarse grain parallel iteration device for the interlayer coupling dynamic application of the integrated circuit, which comprises a parallel calculation grain division module, a parallel calculation coarse grain division module and a parallel calculation coarse grain operation module;
the parallel computing particle dividing module is used for dividing iterative computing of interlayer coupling of the integrated circuit into parallel computing particles;
the parallel computing coarse grain dividing module is used for acquiring weighted CPU time of each parallel computing grain and total CPU time of an iteration method for coupling real-time updating between layers of an integrated circuit based on one-time complete serial iterative computation, and combining the parallel computing grains into two-stage parallel computing coarse grains according to the ratio of the weighted CPU time to the total CPU time; the first stage of parallel coarse grain calculation is to calculate the electromagnetic field and current distribution of each source layer, and the first stage of parallel coarse grain calculation comprises the steps of calculating the influence of other layers on the source layer to update source items acting on the source layer in a coupling mode and calculating the electromagnetic field and current distribution of the source layer by utilizing two-dimensional finite elements; the second level of parallel computing coarse grains is used for computing the layer-layer mutual influence of the integrated circuit;
the parallel computing coarse grain operation module is used for randomly disordering the sequences of all computing tasks executed by the parallel computing coarse grains at the same level in the process of executing the parallel computing coarse grains to form a new computing task sequence, and dynamically distributing the new computing task sequence to different computing processes to complete the parallel computing of the computing tasks.
Furthermore, the first-stage parallel computing coarse grains and the second-stage parallel computing coarse grains are computing task sets in different ranges, the first-stage parallel computing coarse grains comprise a plurality of second-stage parallel computing coarse grains and a computing task for computing electromagnetic fields and current distribution of an integrated circuit layer by using two-dimensional finite elements, the parallel computing grains belonging to the same stage can be computed in parallel, the first-stage parallel computing grains can be distributed with a plurality of computing processes, and the second-stage parallel computing grains can be distributed with only one computing process.
Further, the specific steps of the second-stage parallel computation of coarse grains include:
step 1, calculating an electric field generated by a point current source at a field point, wherein an electric field expression generated by the point current source at the field point is formed according to a layered special structure of an integrated circuitAnalyzing the expression, the current sources of the multilayer integrated circuit are distributed in a layered way, namely, the current density distributed on each metal layer of the integrated circuit layout with the complex shape is only equal to that of the current density distributed on each metal layer of the integrated circuit layout with the complex shapexAndyis related tozIndependently, the current density distribution is onlyx, yA function of (a);
step 2, taking an electric field expression generated by the point current source at the field point as an integrand function of two-dimensional Gaussian integration, and calculating fields generated by the surface current source of the simple-shaped polygon at the same position based on a linear superposition principle of the fields, wherein the field expression comprises the following steps: the field generated by the current source in the two-dimensional plane S at any point in space can be calculated by the two-dimensional gaussian integral:
wherein the content of the first and second substances,at any point in space for the current source within the two-dimensional plane S (x,y,z) The field that is generated is,is an arbitrary position within the two-dimensional surface S: (u,v) At any point in space (a)x,y,z) The expression of the field that is generated,representing a gaussian integration point corresponding to a two-dimensional gaussian integration in the two-dimensional plane S,p,qrespectively representu,vIn the first directionpA first, aqThe number of the Gaussian integration points is equal to the total number of the points,is the weight factor corresponding to the gaussian integral point;
step 3, calculating fields generated by the current on the simple-shaped polygon at different positions of other layers of the integrated circuit, and determining the fields generated by the current on the simple-shaped polygon divided on the layout of other layers of the integrated circuit based on the linear superposition principle of the fields;
step 4, determining the first step based on the linear superposition principle of the fieldlIs layered onmThe electric field generated by the source layer.
Further, the calculation formula of the weighted CPU time of the parallel calculation grain is:in the formula:is as followsiThe weighted CPU times of the particles are calculated in parallel,is as followsiThe CPU time of a single calculation of a grain is calculated in parallel,is as followsiThe number of particle executions is calculated in parallel.
(III) advantageous effects
In the process of dynamically applying the calculation by coupling among the integrated circuit layers, the invention greatly reduces the communication among the processes and the waiting time generated by synchronization, and simultaneously, because of adopting a random dynamic allocation method of the calculation tasks, the invention ensures that the calculation models with unequal complexity are randomly and uniformly distributed on each calculation node, and avoids the bottleneck of hard disk reading and writing caused by virtual memory access due to overhigh peak memory.
Drawings
The embodiments described below with reference to the drawings are exemplary and intended to be used for explaining and illustrating the present invention and should not be construed as limiting the scope of the present invention.
FIG. 1 is a block diagram of the main steps of a first embodiment of the present invention;
FIG. 2 is a flow chart of the second stage parallel computation coarse grain in the first embodiment of the present invention;
FIG. 3 is a block diagram of the modules of a second embodiment of the present invention;
fig. 4 is an exploded view of the electric field generated at the field point of the point source of the present invention.
Detailed Description
In order to make the implementation objects, technical solutions and advantages of the present invention clearer, the technical solutions in the embodiments of the present invention will be described in more detail below with reference to the accompanying drawings in the embodiments of the present invention.
It should be noted that: in the drawings, the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described are some embodiments of the present invention, not all embodiments, and features in embodiments and embodiments in the present application may be combined with each other without conflict. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In the description of the present invention, it is to be understood that the terms "central," "longitudinal," "lateral," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," and the like are used in the orientation or positional relationship indicated in the drawings, which are used for convenience in describing the invention and for simplicity in description, and are not intended to indicate or imply that the referenced device or element must have a particular orientation, be constructed and operated in a particular orientation, and are not to be considered limiting of the scope of the invention.
The following describes in detail a first embodiment of the coarse grain parallel iterative method and apparatus for integrated circuit interlayer coupling dynamic application provided by the present invention with reference to fig. 1, 2 and 4. The coarse grain parallel iteration method for the interlayer coupling dynamic application of the integrated circuit provided by the embodiment, as shown in fig. 1 and 2, includes the following steps:
step S100, dividing iterative computation of coupling between integrated circuit layers into parallel computation particles, combining the parallel computation particles into two-stage parallel computation coarse particles according to weighted CPU time of the parallel computation particles, wherein the first-stage parallel computation coarse particles are used for computing electromagnetic field and current distribution of each source layer, and the first-stage parallel computation coarse particles comprise the steps of updating source items acting on the source layers in a coupling mode by computing influences of other layers on the source layers and computing the electromagnetic field and current distribution of the source layers by utilizing two-dimensional finite elements; the second level of parallel computing coarse grains is used for computing the layer-layer mutual influence of the integrated circuit;
furthermore, the first-stage parallel computing coarse grains and the second-stage parallel computing coarse grains are computing task sets in different ranges, the first-stage parallel computing coarse grains comprise a plurality of second-stage parallel computing coarse grains and a computing task for computing electromagnetic fields and current distribution of an integrated circuit layer by using a two-dimensional finite element, the parallel computing grains belonging to the same stage can be computed in parallel, the first-stage parallel computing coarse grains can be distributed with a plurality of computing processes, and the second-stage parallel computing coarse grains can be distributed with only one computing process.
Specifically, the weighted CPU time of each parallel computation grain is sorted in descending order and sequentially accumulated until the accumulated sum exceeds 90% of the total CPU time, and each parallel computation grain in the accumulated sum is taken as a parallel computation coarse grain. Wherein, the calculation formula of the weighted CPU time of the parallel calculation particles is as follows:in the formula:is as followsiThe weighted CPU times of the particles are calculated in parallel,T i the CPU time of a single calculation for the ith parallel-calculated grain,is as followsiThe number of times the parallel computation particles are executed; the calculation formula of the total CPU time in the whole calculation process is as follows:wherein, in the step (A),Tfor the total CPU time of the entire calculation process,mparallel computing grain partitioned for an entire computing programThe number of the first and second groups is,is as followsiThe weighted CPU times of the particles are calculated in parallel.
For example: if the iterative computation of the integrated circuit interlayer coupling immediate update is divided into 3 parallel computation particles of c1, c2 and c3 according to the definition of the parallel computation particles, the 3 parallel computation particles can execute the computation task of the whole operation process; if c1 executes 500 computation tasks, c2 executes 200 computation tasks, and c3 executes 5 computation tasks; then 705 total computing tasks constitute the whole computation process, which only needs 3 parallel computing granules of c1, c2 and c 3. The whole operation process is executed by 3 parallel computing particles of c1, c2 and c3, and each of c1, c2 and c3 comprises at least 1 independent operation (computing task).
Sorting according to the weighted CPU time obtained by each parallel computing grain operation, wherein if the c1 weighted CPU time is 0.1s, the c2 weighted CPU time is 100s and the c3 weighted CPU time is 0.2s, the final sorting result is c2> c3> c 1; the weighted CPU times for the 3 parallel-calculated particles add sequentially from large to small, i.e., T (c2) + T (c3) + …, until the sum of the times is greater than 90% of the total CPU time; if T (c2) + T (c3) > 90%, then c2, c3 are each taken as a parallel calculation coarse grain; if T (c2) > 90% of the total CPU time, then c2 is parallel computing coarse grain.
The parallel computing coarse grains are classified, the similar parallel computing coarse grains are mutually independent, the corresponding computing task sequences can be randomly disordered, in the process of executing the parallel computing coarse grains, the sequences of all computing tasks executed by the similar parallel computing coarse grains are randomly disordered to form new computing task sequences, and the new computing task sequences are dynamically distributed to different computing processes to complete the parallel computing of the computing tasks.
Specifically, the way of randomly scrambling the operation task sequence is as follows:
firstly, the sequence of operation tasksCorrespondingly generating random number sequences,m=1,2,3,…,M. Then to the sequenceThe sequences are sorted from small to large, and the sorted sequences are. Finally, generating new non-repeated operation task sequence,Is composed ofIn thatOf (c) is used.
The key point is to make all the operation tasks in the parallel particles in sequenceRandomly disorganized to generate new non-repetitive operation task sequenceAnd then distributing the operation tasks according to the sequence, namely equivalently distributing the original operation tasks randomly, wherein the random distribution strategy is characterized in that a random distribution scheme can completely disturb the distribution sequence of all the operation tasks, so that the sum of the peak value memory occupied by the tasks operated by all the operation nodes at the same time is determined by the average value of the process number and the peak value memory occupied by all the models (calculation particles) rather than the maximum value.
And the main process distributes all the operation tasks required to be executed by the parallel coarse grain calculation to all the processes including the main process according to the formed new calculation task sequence, and completes the parallel operation of all the operation tasks executed by the parallel coarse grain calculation.
In addition, if a certain operation task in the parallel computing coarse grain is distributed to a process, a mark file which is used for indicating that the operation task is already distributed to the operation task is generated; when applying for distributing a certain calculation task, the other process tries to generate a mark file of the calculation task, and automatically applies for distributing the next calculation task by the other process under the condition that the mark file exists.
In the multi-process parallel operation process, the chances of allocating a certain operation task to each process are equal, if no measure is taken, multiple processes may be allocated to the same operation task, and the waste of operation resources is caused, so that some measure must be taken, and all operation tasks are uniquely allocated to a certain process. The simplest and most intuitive measure for achieving this is to assign a task a time stamp, i.e. a task is assigned to a process at the same time as it is marked so that other processes are no longer assigned the task. However, because the variables of each process are generally independent of each other during parallel operation, the operation tasks are asymmetric, the operation states of each process are different, and information distributed by any process through the variable marking task cannot be immediately transmitted to other processes, an external explicit marking method is needed to be adopted so that all processes can obtain the information once the operation tasks are marked. Therefore, if the operation tasks in the parallel computation coarse grains are distributed to the processes, the mark files of the operation tasks are immediately generated; when a process applies for distributing a certain operation task, the process will try to generate a mark file of the operation task, if the mark file exists, the operation task is indicated to be distributed, and the process will automatically apply for distributing the next operation task.
The specific implementation steps for realizing the correct allocation of the operation tasks by utilizing the marker files are as follows:
step A1, a process applies for distributioniAn arithmetic task;
step A2,Judgment ofiSign file of individual operation taskFiIf the current state does not exist, jumping to the step A8, and if the current state does not exist, jumping to the step A3;
step A3, judging the flag file FiWhether the lock is locked or not, jumping to the step A8 if the lock is locked, and jumping to the step A4 if the lock is not locked;
step A4, locking the logo fileFi;
Step A5, generating a logo fileFi;
Step A6, marking fileFiUnlocking;
step A7, completing the first stepiCalculating the operation tasks;
step A8, judging whether all the operation tasks in the parallel computation coarse grain are completed or not, if not, theni=i+1 and returning to step a1, if finished, jumping to step a 9;
all operation tasks required to be executed by the parallel calculation coarse grain are all distributed to all processes, and the distribution of the parallel calculation coarse grain is finished; it returns to performing all the computational tasks that the other parallel computing coarse grains need to perform individually.
Step S200, the large scale integrated circuit is summarizedN+1 layer, each layer numberedWhen considering the second aspect of LSImWhen the current source of the layer is in the current source of the layer, the layer is called the first layermA source layer provided with a secondmActive layer of source layerIs divided bymOthers of the source layerNLayer integrated circuits, i.e.WhereinThe 0 th layer is a bottom layer; setting the number of iterations;
Step S300, ifComputing the first based on the coarse grain of the first stage parallel computingmThe electromagnetic field and current distribution of the source layer are calculated, and the change amount of the electromagnetic field of the layer is calculateddE m And a firstlInfluence value of layer on dyadic Green function of layerG lm Maximum value ofG m,maxAnd minimum valueG m,minWherein 0 is less than or equal tom≤N;
Further, the step S300 includes:
step S310, calculating coarse grains based on the second-stage parallel calculation, and calculating the second stage by utilizing a dyadic Green functionlLayer to layermInfluence of the Source layer, isG lm Wherein, in the step (A),(ii) a All will beG lm Overlapping to obtain the other layer pairs within the range of the action layermSum of influence of source layers:(ii) a Finishing the second-stage parallel calculation of coarse grains;
step S320, ifiter>0, willG m As a firstmAdditional sources of source layers added to the secondmIn the source item of the source layer;
step S330, for the secondmThe source layer applies two-dimensional finite element to calculate the electromagnetic field distribution so as to update the electromagnetic field and current distribution of the layer, and the change quantity of the electromagnetic field of the layer is calculateddE m ;
Step S340, according tolInfluence value of layer on dyadic Green function of layerG lm CalculatingG lm Maximum value ofG m,maxAnd minimum valueG m,min。
Step S400, ifAnd isAnd (5) finishing the iteration, and outputting electromagnetic fields and current distributions of each layer, whereinThe iteration precision is preset;
step S500, ifMerging all the influence values of the dyadic Green function obtained by the parallel calculation of the coarse grains at the first stage, and obtaining the maximum value of the influence values of all the dyadic Green functions of the current iterationG maxAnd minimum valueG minCalculating the effective influence value of the dyadic Green functionHere, thethredsholdA discarding threshold value for the influence of a preset dyadic Green function;
step S600, ifCalculating satisfyOf the conditionG lm Middle distance layermNearest layerl near Will beIs updated to(ii) a Because the influence of the point source on any point in the space is weakened along with the distance between the point source and the point as known by the attenuation rule of the electromagnetic field and the electromagnetic wave in the space, the influence value is in inverse proportion to the distance, and the influence value of the layer interface isReflection makes the electromagnetic wave from the source point to the field point in the space weaken faster, therefore, when using the dyadic Green function to calculate the effect of the point source to the space point, it can be considered that the effect of the point source to the space point can be ignored after the distance between the space point and the point source is greater than a certain degree, or the number of the medium layers separated from each other reaches a certain degree. Therefore, in the iteration process, a layer capable of being ignored is determined according to the magnitude of the influence value G of the dyadic Green function, and the second step of self-adaptive adjustment is carried outmExtent of influence of source layer by other layers adjacent thereto。
Step S700, settingiter=iter+1, the process proceeds to step S300.
Further, the specific steps of parallel coarse grain calculation in the second stage as shown in fig. 2 include:
step S311, calculating an electric field generated by the point current source at the field point, where an electric field expression generated by the point current source at the field point is a special analytic expression formed according to a special structure of the integrated circuit layer, as shown in fig. 4, the concrete analytic expression is as follows: aiming at the frequency domain electromagnetic field of the multilayer integrated circuit layout, the electric field intensity generated by the point source at any layer of field point is calculated by adopting a dyadic Green function, and the electric field intensity in nine directions of any point of any layer of the multilayer integrated circuit layout can be solved through the following formula to express that the electric field expression of the point source to the field point is solved:
the electric field expression generated by the point current source at the field point is as follows:
wherein the content of the first and second substances,
iis the unit of an imaginary number,i 2=-1;representing a Bessel function of order 0;representing a Bessel function of order 1;expressed as a function of the Bessel integral coefficient,;x, y, zthe coordinates of the field points are represented,, , representing source point coordinates; angular frequency,Represents a frequency;indicating that the site is at the secondA layer of a material selected from the group consisting of,is as followsAt layer boundarieszCoordinates;, respectively representThe number of complex waves in the horizontal and vertical directions of the layer;respectively representA layer horizontal dielectric constant, a vertical dielectric constant;, respectively representlHorizontal magnetic conductivity and vertical magnetic conductivity of the layer;is shown aslThe anisotropy coefficient of the layer;, respectively representlIntegral coefficients of complex wave numbers of the horizontal and vertical layers;respectively representlThe undetermined coefficient of a layer,A l , B l the following linear equation is solved:
to representxOriented electric dipole in the second placelOf said electric field generated by said field points of the layerxA component;
to representxOriented electric dipole in the second placelOf said electric field generated by said field points of the layeryA component;
to representxOriented electric dipole in the second placelOf said electric field generated by said field points of the layerzA component;
to representyOriented electric dipole in the second placelOf said electric field generated by said field points of the layerxA component;
to representyOriented electric dipole in the second placelOf said electric field generated by said field points of the layeryA component;
to representyOriented electric dipole in the second placelOf said electric field generated by said field points of the layerzA component;
to representzOriented electric dipole in the second placelOf said electric field generated by said field points of the layerxA component;
to representzOriented electric dipole in the second placelOf said electric field generated by said field points of the layeryA component;
to representzOriented electric dipole in the second placelOf said electric field generated by said field points of the layerzAnd (4) components.
The current sources of the multilayer integrated circuit are distributed in a layered manner, namely the current density distributed on each metal layer of the integrated circuit layout with the complex shape is only equal to that of the current density distributed on each metal layer of the integrated circuit layout with the complex shapexAndyis related tozIndependently, the current density distribution is onlyx, yAs a function of (c).
Step S312, taking an electric field expression generated by the point current source at the field point as an integrand of two-dimensional gaussian integration, and calculating a field generated by the surface current source of the simple-shaped polygon at the same position based on a field linear superposition principle, including: the field generated by the current source in the two-dimensional plane S at any point in space can be calculated by the two-dimensional gaussian integral:
wherein the content of the first and second substances,E(x,y,z) At any point in space for the current source in the two-dimensional plane Sx,y,z) The field that is generated is,is an arbitrary position within the two-dimensional surface S: (u,v) At any point in space (x,y,z) The expression of the dyadic green function of the generated field,representing a gaussian integration point corresponding to a two-dimensional gaussian integration in the two-dimensional plane S,p,qrespectively representu,vIn the first directionpA first, aqThe number of the Gaussian integration points is equal to the total number of the points,is the weight factor corresponding to the gaussian integral point;
step S313, calculating fields generated by the current on the simple-shaped polygon at different positions of other layers of the integrated circuit, and determining the fields generated by the current on the simple-shaped polygon divided on the layout of other layers of the integrated circuit based on the linear superposition principle of the fields;
step S314, determining the second step based on the linear superposition principle of the fieldlIs layered onmThe electric field generated by the source layer.
Further, influence values according to the dyadic Green function in the iteration processGDetermines a layer that can be ignored, adaptively adjustsmExtent of influence of source layer by other layers adjacent thereto。
Further, the coarse grains of the first-stage parallel computation are computed by two-dimensional finite element methodmThe electromagnetic field and current distribution of the source layer are as follows:
for the direct current electric field model, the three-dimensional model of the multilayer integrated circuit refers to the conductivity in the direct current electric field modelPotential of the electrodeuAll the distributions of (A) and (B) are three-dimensional space coordinatesx,y,z) I.e.:,the function of the three-dimensional model satisfies the following equation (1):
and boundary condition (2):
in the formulaIs a boundary of the first type and is,nis normal to the boundary of the second type,represents a potentialuAt the first kind boundaryValue of above, usingIt is shown that,bulk current density for external circuits;
the dimension of an actual PCB or a chip packaged board in the multilayer super large scale integrated circuit is far larger than the thickness of the metal layer, so that the three-dimensional direct current field problem of the multilayer integrated circuit is simplified into a two-dimensional direct current field problem;
the field solving equation set established by the finite element method for the two-dimensional model is an equation set (3):
in the formula (I), theI(u) In order to be a functional function,tis the thickness of the metal layer or layers,as a grid celleThe electrical conductivity of (a) a (b),as a grid celleThe potential of (a) is set to be,as a grid celleThe area of (a) is,as the density of the surface current, the current density,representing grid cellseThe edge of (1);
for the alternating electromagnetic field model, the three-dimensional model of the multilayer integrated circuit refers to the dielectric constant in the three-dimensional model of the electromagnetic response characteristic in the frequency domain simulation of the multilayer VLSIMagnetic permeability ofElectric field intensityEMagnetic field intensityHAll the distributions of (A) and (B) are three-dimensional space coordinatesx,y,z) I.e.:, , ,the function of the three-dimensional model satisfies the following equation:
in the formulaJFor the purpose of the applied current density distribution,for the angular frequency simulated for the integrated circuit,indicating the strength of the magnetic fieldHRotation of,Indicates the electric field intensityEThe degree of rotation of the screw is reduced,jis the unit of an imaginary number,j 2=-1;
the board size of the actual PCB or chip package in the multilayer VLSI is far larger than the metal layer spacing, the three-dimensional model of the electromagnetic response characteristics in the frequency domain simulation of the multilayer VLSI is simplified into a two-dimensional model, and the dielectric constant in the model is at the momentMagnetic permeability ofElectric field intensityEMagnetic field intensityHAll the distributions are two-dimensional plane coordinates (x,y) I.e.:,,,distribution thereof andzindependent of and potential in the fielduAnd surface current densityJ sSatisfies the following conditions:
in the formula (I), the compound is shown in the specification,respectively representx, y, zThe unit vector of the direction is,E zof electric field strengthzThe direction component of the light beam is,H xandH yrespectively of magnetic field strengthxAndythe direction component of the light beam is,his the metal layer spacing;
through the simplification from the three-dimensional model to the two-dimensional model, the two-dimensional finite element functional extreme value formula corresponding to the two-dimensional model is obtained as follows:
in the formula (I), the compound is shown in the specification,
in order to be a functional function,it is shown that the extreme value is taken for the functional,as a grid celliThe surface admittance of the first and second electrodes,is a boundaryThe boundary condition of the opening of (a),u kis a boundaryThe distribution of the electric potential on the upper side,indicating a position to the right of the boundary and infinitely close to the boundary,indicating a position to the left of the boundary and infinitely close to the boundary,representing grid cellsiThe area of (a) is,as a grid celliThe current density of (a) is,as a grid celliThe surface resistance of the glass substrate is higher than the surface resistance of the glass substrate,as a grid celliThe potential of (a) is set to be,kis referred to askAnd (4) a boundary.
From the above iteration steps, it can be seen that each iteration updates the electromagnetic field distribution and current distribution of each layer, rather than actively calculating the influence of the source layer on other layers. When the electromagnetic field distribution and the current distribution of each layer are calculated, the source item of the layer is determined to be the excitation source corresponding to the external circuit and the source item corresponding to the influence of other layers on the excitation source. The influence of other layers on the electromagnetic field and the electromagnetic wave can be determined according to the attenuation rule of the electromagnetic field and the electromagnetic wave in the space. And meanwhile, in the iteration process, the range of the influence exerted on other layers by each source layer is adaptively adjusted according to the magnitude of the influence value of the dyadic Green function of each layer.
Particularly, when the voltage drop and the current distribution of a power supply layer of the integrated circuit are analyzed, the working frequency is low frequency, the direct current field model is adopted for analysis, no space coupling exists between the integrated circuit layers at the moment, only physical coupling exists, namely, the layers which are connected with each other through the through hole and the external circuit are mutually coupled, at the moment, the mutual influence layers between the integrated circuit layers are determined, and iteration is not needed to influence the influence rangeN eff,m And (6) correcting.
A second embodiment of the coarse grain parallel iterative method and apparatus for interlayer coupling dynamic application of an integrated circuit provided by the present invention is described in detail below with reference to fig. 3 and 4. As shown in fig. 3 and 4, the coarse grain parallel iteration apparatus for interlayer coupling dynamic application of an integrated circuit provided in this embodiment includes, as shown in fig. 3, a parallel calculation grain division module, a parallel calculation coarse grain division module, and a parallel calculation coarse grain operation module;
the parallel computing particle dividing module is used for dividing iterative computing of interlayer coupling of the integrated circuit into parallel computing particles;
the parallel computing coarse grain dividing module is used for acquiring weighted CPU time of each parallel computing grain and total CPU time of an iteration method for coupling real-time updating between layers of an integrated circuit based on one-time complete serial iterative computation, and combining the parallel computing grains into two-stage parallel computing coarse grains according to the ratio of the weighted CPU time to the total CPU time; the first stage of parallel coarse grain calculation is to calculate the electromagnetic field and current distribution of each source layer, and the first stage of parallel coarse grain calculation comprises the steps of calculating the influence of other layers on the source layer to update source items acting on the source layer in a coupling mode and calculating the electromagnetic field and current distribution of the source layer by utilizing two-dimensional finite elements; the second level of parallel computing coarse grains is used for computing the layer-layer mutual influence of the integrated circuit;
specifically, the weighted CPU time of each parallel computation grain is sorted in descending order and sequentially accumulated until the accumulated sum exceeds 90% of the total CPU time, and each parallel computation grain in the accumulated sum is taken as a parallel computation coarse grain. Wherein, the calculation formula of the weighted CPU time of the parallel calculation particles is as follows:in the formula:is as followsiThe weighted CPU times of the particles are calculated in parallel,is as followsiThe CPU time of a single calculation of a grain is calculated in parallel,is as followsiThe number of times the parallel computation particles are executed; the calculation formula of the total CPU time in the whole calculation process is as follows:wherein, in the step (A),Tfor the total CPU time of the entire calculation process,mthe number of parallel computing grains divided for the entire computing program,is as followsiThe weighted CPU times of the particles are calculated in parallel.
For example: if the iterative computation of the integrated circuit interlayer coupling immediate update is divided into 3 parallel computation particles of c1, c2 and c3 according to the definition of the parallel computation particles, the 3 parallel computation particles can execute the computation task of the whole operation process; if c1 executes 500 computation tasks, c2 executes 200 computation tasks, and c3 executes 5 computation tasks; then 705 total computing tasks constitute the whole computation process, which only needs 3 parallel computing granules of c1, c2 and c 3. The whole operation process is executed by 3 parallel computing particles of c1, c2 and c3, and each of c1, c2 and c3 comprises at least 1 independent operation (computing task).
Sorting according to the weighted CPU time obtained by each parallel computing grain operation, wherein if the c1 weighted CPU time is 0.1s, the c2 weighted CPU time is 100s and the c3 weighted CPU time is 0.2s, the final sorting result is c2> c3> c 1; the weighted CPU times for the 3 parallel-calculated particles add sequentially from large to small, i.e., T (c2) + T (c3) + …, until the sum of the times is greater than 90% of the total CPU time; if T (c2) + T (c3) > 90%, then c2, c3 are each taken as a parallel calculation coarse grain; if T (c2) > 90% of the total CPU time, then c2 is parallel computing coarse grain.
The parallel computing coarse grains are classified, the similar parallel computing coarse grains are mutually independent, the corresponding computing task sequences can be randomly disordered, in the process of executing the parallel computing coarse grains, the sequences of all computing tasks executed by the similar parallel computing coarse grains are randomly disordered to form new computing task sequences, and the new computing task sequences are dynamically distributed to different computing processes to complete the parallel computing of the computing tasks.
Furthermore, the first-stage parallel computing coarse grains and the second-stage parallel computing coarse grains are computing task sets in different ranges, the first-stage parallel computing coarse grains comprise a plurality of second-stage parallel computing coarse grains and a computing task for computing electromagnetic fields and current distribution of an integrated circuit layer by using two-dimensional finite elements, the parallel computing grains belonging to the same stage can be computed in parallel, the first-stage parallel computing grains can be distributed with a plurality of computing processes, and the second-stage parallel computing grains can be distributed with only one computing process.
The parallel computing coarse grain operation module is used for randomly disordering the sequences of all computing tasks executed by the parallel computing coarse grains at the same level in the process of executing the parallel computing coarse grains to form a new computing task sequence, and dynamically distributing the new computing task sequence to different computing processes to complete the parallel computing of the computing tasks.
Specifically, the way of randomly scrambling the operation task sequence is as follows:
firstly, the sequence of operation tasksCorrespondingly generating random number sequences,m=1,2,3,…,M. Then to the sequenceThe sequences are sorted from small to large, and the sorted sequences are. Finally, generating new non-repeated operation task sequence,Is composed ofIn thatOf (c) is used.
The key point is to make all the operation tasks in the parallel particles in sequenceRandomly disorganized to generate new non-repetitive operation task sequenceAnd then distributing the operation tasks according to the sequence, namely equivalently distributing the original operation tasks randomly, wherein the random distribution strategy is characterized in that a random distribution scheme can completely disturb the distribution sequence of all the operation tasks, so that the sum of the peak value memory occupied by the tasks operated by all the operation nodes at the same time is determined by the average value of the process number and the peak value memory occupied by all the models (calculation particles) rather than the maximum value.
And the main process distributes all the operation tasks required to be executed by the parallel coarse grain calculation to all the processes including the main process according to the formed new calculation task sequence, and completes the parallel operation of all the operation tasks executed by the parallel coarse grain calculation.
In addition, if a certain operation task in the parallel computing coarse grain is distributed to a process, a mark file which is used for indicating that the operation task is already distributed to the operation task is generated; when applying for distributing a certain calculation task, the other process tries to generate a mark file of the calculation task, and automatically applies for distributing the next calculation task by the other process under the condition that the mark file exists.
In the multi-process parallel operation process, the chances of allocating a certain operation task to each process are equal, if no measure is taken, multiple processes may be allocated to the same operation task, and the waste of operation resources is caused, so that some measure must be taken, and all operation tasks are uniquely allocated to a certain process. The simplest and most intuitive measure for achieving this is to assign a task a time stamp, i.e. a task is assigned to a process at the same time as it is marked so that other processes are no longer assigned the task. However, because the variables of each process are generally independent of each other during parallel operation, the operation tasks are asymmetric, the operation states of each process are different, and information distributed by any process through the variable marking task cannot be immediately transmitted to other processes, an external explicit marking method is needed to be adopted so that all processes can obtain the information once the operation tasks are marked. Therefore, if the operation tasks in the parallel computation coarse grains are distributed to the processes, the mark files of the operation tasks are immediately generated; when a process applies for distributing a certain operation task, the process will try to generate a mark file of the operation task, if the mark file exists, the operation task is indicated to be distributed, and the process will automatically apply for distributing the next operation task.
The specific implementation steps for realizing the correct allocation of the operation tasks by utilizing the marker files are as follows:
step A1, a process applies for distributioniAn arithmetic task;
step A2, judgmentiSign file of individual operation taskFiIf the current state does not exist, jumping to the step A8, and if the current state does not exist, jumping to the step A3;
step A3, judging the mark fileFiWhether the lock is locked or not, jumping to the step A8 if the lock is locked, and jumping to the step A4 if the lock is not locked;
step A4, locking the logo fileFi;
Step A5, generating a logo fileFi;
Step A6, marking fileFiUnlocking;
step A7, completing the first stepiCalculating the operation tasks;
step A8, judging whether all the operation tasks in the parallel computation coarse grain are completed or not, if not, theni=i+1 and returning to step a1, if finished, jumping to step a 9;
all operation tasks required to be executed by the parallel calculation coarse grain are all distributed to all processes, and the distribution of the parallel calculation coarse grain is finished; it returns to performing all the computational tasks that the other parallel computing coarse grains need to perform individually.
Further, the specific steps of the second-stage parallel computation of coarse grains include:
step 1, calculating an electric field generated at a field point by a point current source, wherein an electric field expression generated at the field point by the point current source is a special analytical expression formed according to a special layered structure of an integrated circuit, as shown in fig. 4, the analytical expression is as follows: aiming at the frequency domain electromagnetic field of the multilayer integrated circuit layout, the electric field intensity generated by the point source at any layer of field point is calculated by adopting a dyadic Green function, and the electric field intensity in nine directions of any point of any layer of the multilayer integrated circuit layout can be solved through the following formula to express that the electric field expression of the point source to the field point is solved:
the electric field expression generated by the point current source at the field point is as follows:
wherein the content of the first and second substances,
iis the unit of an imaginary number,i 2=-1;representing a Bessel function of order 0;representing a Bessel function of order 1;expressed as a function of the Bessel integral coefficient,;x, y, zthe coordinates of the field points are represented,, , representing source point coordinates; angular frequency,Represents a frequency;indicating that the site is at the secondA layer of a material selected from the group consisting of,is as followsAt layer boundarieszCoordinates;, respectively representThe number of complex waves in the horizontal and vertical directions of the layer;respectively representA layer horizontal dielectric constant, a vertical dielectric constant;, respectively representlHorizontal magnetic conductivity and vertical magnetic conductivity of the layer;is shown aslThe anisotropy coefficient of the layer;, respectively representlIntegral coefficients of complex wave numbers of the horizontal and vertical layers;respectively representlThe undetermined coefficient of a layer,A l , B l the following linear equation is solved:
to representxOriented electric dipole in the second placelOf said electric field generated by said field points of the layerxA component;
to representxOriented electric dipole in the second placelOf said electric field generated by said field points of the layeryA component;
to representxOriented electric dipole in the second placelOf said electric field generated by said field points of the layerzA component;
to representyOriented electric dipole in the second placelSaid field of the layerOf said electric field generated by the dotsxA component;
to representyOriented electric dipole in the second placelOf said electric field generated by said field points of the layeryA component;
to representyOriented electric dipole in the second placelOf said electric field generated by said field points of the layerzA component;
to representzOriented electric dipole in the second placelOf said electric field generated by said field points of the layerxA component;
to representzOriented electric dipole in the second placelOf said electric field generated by said field points of the layeryA component;
to representzOriented electric dipole in the second placelOf said electric field generated by said field points of the layerzAnd (4) components.
The current sources of the multilayer integrated circuit are distributed in a layered manner, namely the current density distributed on each metal layer of the integrated circuit layout with the complex shape is only equal to that of the current density distributed on each metal layer of the integrated circuit layout with the complex shapexAndyis related tozIndependently, the current density distribution is onlyx, yAs a function of (c).
Step 2, taking an electric field expression generated by the point current source at the field point as an integrand function of two-dimensional Gaussian integration, and calculating fields generated by the surface current source of the simple-shaped polygon at the same position based on a linear superposition principle of the fields, wherein the field expression comprises the following steps: the field generated by the current source in the two-dimensional plane S at any point in space can be calculated by the two-dimensional gaussian integral:
wherein the content of the first and second substances,E(x,y,z) At any point in space for the current source in the two-dimensional plane Sx,y,z) The field that is generated is,is an arbitrary position within the two-dimensional surface S: (u,v) At any point in space (x,y,z) The expression of the dyadic green function of the generated field,representing a gaussian integration point corresponding to a two-dimensional gaussian integration in the two-dimensional plane S,p,qrespectively representu,vIn the first directionpA first, aqThe number of the Gaussian integration points is equal to the total number of the points,is the weight factor corresponding to the gaussian integral point;
step 3, calculating fields generated by the current on the simple-shaped polygon at different positions of other layers of the integrated circuit, and determining the fields generated by the current on the simple-shaped polygon divided on the layout of other layers of the integrated circuit based on the linear superposition principle of the fields;
step 4, determining the first step based on the linear superposition principle of the fieldlIs layered onmThe electric field generated by the source layer.
Further, influence values according to the dyadic Green function in the iteration processGDetermines a layer that can be ignored, adaptively adjustsmExtent of influence of source layer by other layers adjacent thereto。
Further, the above-mentionedThe coarse grains are calculated in parallel in the first stage through two-dimensional finite element calculationmThe electromagnetic field and current distribution of the source layer are as follows:
for the direct current electric field model, the three-dimensional model of the multilayer integrated circuit refers to the conductivity in the direct current electric field modelPotential of the electrodeuAll the distributions of (A) and (B) are three-dimensional space coordinatesx,y,z) I.e.:,the function of the three-dimensional model satisfies the following equation (1):
and boundary condition (2):
in the formulaIs a boundary of the first type and is,nis normal to the boundary of the second type,represents a potentialuAt the first kind boundaryValue of above, usingIt is shown that,bulk current density for external circuits;
the dimension of an actual PCB or a chip packaged board in the multilayer super large scale integrated circuit is far larger than the thickness of the metal layer, so that the three-dimensional direct current field problem of the multilayer integrated circuit is simplified into a two-dimensional direct current field problem;
the field solving equation set established by the finite element method for the two-dimensional model is an equation set (3):
in the formula (I), theI(u) In order to be a functional function,tis the thickness of the metal layer or layers,as a grid celleThe electrical conductivity of (a) a (b),as a grid celleThe potential of (a) is set to be,as a grid celleThe area of (a) is,as the density of the surface current, the current density,representing grid cellseThe edge of (1);
for the alternating electromagnetic field model, the three-dimensional model of the multilayer integrated circuit refers to the dielectric constant in the three-dimensional model of the electromagnetic response characteristic in the frequency domain simulation of the multilayer VLSIMagnetic permeability ofElectric field intensityEMagnetic field intensityHAre all distributedAs three-dimensional space coordinates (x,y,z) I.e.:, , ,the function of the three-dimensional model satisfies the following equation:
in the formulaJFor the purpose of the applied current density distribution,for the angular frequency simulated for the integrated circuit,indicating the strength of the magnetic fieldHThe degree of rotation of the screw is reduced,indicates the electric field intensityEThe degree of rotation of the screw is reduced,jis the unit of an imaginary number,j 2=-1;
the board size of the actual PCB or chip package in the multilayer VLSI is far larger than the metal layer spacing, the three-dimensional model of the electromagnetic response characteristics in the frequency domain simulation of the multilayer VLSI is simplified into a two-dimensional model, and the dielectric constant in the model is at the momentMagnetic permeability ofElectric field intensityEMagnetic field intensityHAll the distributions are two-dimensional plane coordinates (x,y) I.e.:,,,distribution thereof andzindependent of and potential in the fielduAnd surface current densityJ sSatisfies the following conditions:
in the formula (I), the compound is shown in the specification,respectively representx, y, zThe unit vector of the direction is,E zof electric field strengthzThe direction component of the light beam is,H xandH yrespectively of magnetic field strengthxAndythe direction component of the light beam is,his the metal layer spacing;
through the simplification from the three-dimensional model to the two-dimensional model, the two-dimensional finite element functional extreme value formula corresponding to the two-dimensional model is obtained as follows:
in the formula (I), the compound is shown in the specification,
in order to be a functional function,it is shown that the extreme value is taken for the functional,as a grid celliThe surface admittance of the first and second electrodes,is a boundaryThe boundary condition of the opening of (a),u kis a boundaryThe distribution of the electric potential on the upper side,indicating a position to the right of the boundary and infinitely close to the boundary,indicating a position to the left of the boundary and infinitely close to the boundary,representing grid cellsiThe area of (a) is,as a grid celliThe current density of (a) is,as a grid celliThe surface resistance of the glass substrate is higher than the surface resistance of the glass substrate,as a grid celliThe potential of (a) is set to be,kis referred to askAnd (4) a boundary. From the above iteration steps, it can be seen that each iteration updates the electromagnetic field distribution and current distribution of each layer, rather than actively calculating the influence of the source layer on other layers. When the electromagnetic field distribution and the current distribution of each layer are calculated, determining the source item of the layer as an excitation source corresponding to an external circuitAnd source items for which other layers affect it. The influence of other layers on the electromagnetic field and the electromagnetic wave can be determined according to the attenuation rule of the electromagnetic field and the electromagnetic wave in the space. And meanwhile, in the iteration process, the range of the influence exerted on other layers by each source layer is adaptively adjusted according to the magnitude of the influence value of the dyadic Green function of each layer.
Particularly, when the voltage drop and the current distribution of a power supply layer of the integrated circuit are analyzed, the working frequency is low frequency, the direct current field model is adopted for analysis, no space coupling exists between the integrated circuit layers at the moment, only physical coupling exists, namely, the layers which are connected with each other through the through hole and the external circuit are mutually coupled, at the moment, the mutual influence layers between the integrated circuit layers are determined, and iteration is not needed to influence the influence rangeAnd (6) correcting.
The above description is only for the specific embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the appended claims.
Claims (10)
1. A coarse grain parallel iteration method for coupling dynamic application between layers of an integrated circuit is characterized by comprising the following steps:
step S100, dividing iterative computation of coupling between integrated circuit layers into parallel computation particles, combining the parallel computation particles into two-stage parallel computation coarse particles according to weighted CPU time of the parallel computation particles, wherein the first-stage parallel computation coarse particles are used for computing electromagnetic field and current distribution of each source layer, and the first-stage parallel computation coarse particles comprise the steps of updating source items acting on the source layers in a coupling mode by computing influences of other layers on the source layers and computing the electromagnetic field and current distribution of the source layers by utilizing two-dimensional finite elements; the second level of parallel computing coarse grains is used for computing the layer-layer mutual influence of the integrated circuit;
step S200, the large scale integrated circuit is summarizedN+1 layer, each layer numberedWhen considering the second aspect of LSImWhen the current source of the layer is in the current source of the layer, the layer is called the first layermA source layer provided with a secondmActive layer of source layerIs divided bymOthers of the source layerNLayer integrated circuits, i.e.WhereinThe 0 th layer is a bottom layer; setting the number of iterations;
Step S300, ifComputing the first based on the coarse grain of the first stage parallel computingmThe electromagnetic field and current distribution of the source layer are calculated, and the change amount of the electromagnetic field of the layer is calculateddE m And a firstlInfluence value of layer on dyadic Green function of layerG lm Maximum value ofG m,maxAnd minimum valueG m,minWherein 0 is less than or equal tom≤N;
Step S400, ifAnd isAnd (5) finishing the iteration, and outputting electromagnetic fields and current distributions of each layer, whereinThe iteration precision is preset;
step S500, ifMerging all the influence values of the dyadic Green function obtained by the parallel calculation of the coarse grains at the first stage, and obtaining the maximum value of the influence values of all the dyadic Green functions of the current iterationG maxAnd minimum valueG minCalculating the effective influence value of the dyadic Green functionHere, thethredsholdA discarding threshold value for the influence of a preset dyadic Green function;
step S600, ifCalculating satisfyOf the conditionG lm Middle distance layermNearest layerl near Will beIs updated to;
2. The method of claim 1, wherein in step S100, the parallel computing of the coarse grain is performedThe formula for calculating the weighted CPU time is as follows:in the formula:is as followsiThe weighted CPU times of the particles are calculated in parallel,is as followsiThe CPU time of a single calculation of a grain is calculated in parallel,is as followsiThe number of particle executions is calculated in parallel.
3. The integrated circuit interlayer coupling dynamic applied coarse grain parallel iteration method according to claim 2, wherein the first-stage parallel computation coarse grains and the second-stage parallel computation coarse grains are a set of computation tasks in different ranges, the first-stage parallel computation coarse grains comprise a plurality of second-stage parallel computation coarse grains and a computation task for computing electromagnetic fields and current distributions of the integrated circuit layer by using two-dimensional finite elements, parallel computation grains belonging to the same stage can be computed in parallel, the first-stage parallel computation grains can be allocated with a plurality of computation processes, and the second-stage parallel computation grains can be allocated with only one computation process.
4. The integrated circuit interlayer coupling dynamic applied coarse grain parallel iterative method of claim 1, wherein said step S300 comprises:
step S310, calculating coarse grains based on the second-stage parallel calculation, and calculating the second stage by utilizing a dyadic Green functionlLayer to layermInfluence of the Source layer, isG lm Wherein, in the step (A),(ii) a All will beG lm Overlapping to obtain the other layer pairs within the range of the action layermSum of influence of source layers:(ii) a Finishing the second-stage parallel calculation of coarse grains;
step S320, ifiter>0, willG m As a firstmAdditional sources of source layers added to the secondmIn the source item of the source layer;
step S330, for the secondmThe source layer applies two-dimensional finite element to calculate the electromagnetic field distribution so as to update the electromagnetic field and current distribution of the layer, and the change quantity of the electromagnetic field of the layer is calculateddE m ;
Step S340, according tolInfluence value of layer on dyadic Green function of layerG lm CalculatingG lm Maximum value ofG m,maxAnd minimum valueG m,min。
5. The method of claim 4, wherein the step of performing parallel coarse grain computation at the second stage comprises:
step S311, calculating the electric field generated by the point current source at the field point, wherein the electric field expression generated by the point current source at the field point is a special analytic expression formed according to the layered special structure of the integrated circuit, and the current sources of the multilayer integrated circuit are layered, that is, the current density distributed on each metal layer of the integrated circuit layout with the complex shape is only equal to that of each metal layer of the integrated circuit layout with the complex shapexAndyis related tozIndependently, the current density distribution is onlyx, yA function of (a);
step S312, taking an electric field expression generated by the point current source at the field point as an integrand of two-dimensional gaussian integration, and calculating a field generated by the surface current source of the simple-shaped polygon at the same position based on a field linear superposition principle, including: the field generated by the current source in the two-dimensional plane S at any point in space can be calculated by the two-dimensional gaussian integral:
wherein the content of the first and second substances,at any point in space for the current source within the two-dimensional plane S (x,y,z) The field that is generated is,is an arbitrary position within the two-dimensional surface S: (u,v) At any point in space (a)x,y,z) The expression of the field that is generated,representing a gaussian integration point corresponding to a two-dimensional gaussian integration in the two-dimensional plane S,p,qrespectively representu,vIn the first directionpA first, aqThe number of the Gaussian integration points is equal to the total number of the points,is the weight factor corresponding to the gaussian integral point;
step S313, calculating fields generated by the current on the simple-shaped polygon at different positions of other layers of the integrated circuit, and determining the fields generated by the current on the simple-shaped polygon divided on the layout of other layers of the integrated circuit based on the linear superposition principle of the fields;
step S314, determining the second step based on the linear superposition principle of the fieldlIs layered onmThe electric field generated by the source layer.
7. A coarse grain parallel iteration device for coupling dynamic application between integrated circuit layers is characterized by comprising a parallel calculation grain division module, a parallel calculation coarse grain division module and a parallel calculation coarse grain operation module;
the parallel computing particle dividing module is used for dividing iterative computing of interlayer coupling of the integrated circuit into parallel computing particles;
the parallel computing coarse grain dividing module is used for acquiring weighted CPU time of each parallel computing grain and total CPU time of an iteration method for coupling real-time updating between layers of an integrated circuit based on one-time complete serial iterative computation, and combining the parallel computing grains into two-stage parallel computing coarse grains according to the ratio of the weighted CPU time to the total CPU time; the first stage of parallel coarse grain calculation is to calculate the electromagnetic field and current distribution of each source layer, and the first stage of parallel coarse grain calculation comprises the steps of calculating the influence of other layers on the source layer to update source items acting on the source layer in a coupling mode and calculating the electromagnetic field and current distribution of the source layer by utilizing two-dimensional finite elements; the second level of parallel computing coarse grains is used for computing the layer-layer mutual influence of the integrated circuit;
the parallel computing coarse grain operation module is used for randomly disordering the sequences of all computing tasks executed by the parallel computing coarse grains at the same level in the process of executing the parallel computing coarse grains to form a new computing task sequence, and dynamically distributing the new computing task sequence to different computing processes to complete the parallel computing of the computing tasks.
8. The integrated circuit interlayer coupling dynamic applied coarse grain parallel iteration device according to claim 7, wherein the first-stage parallel computation coarse grains and the second-stage parallel computation coarse grains are a set of computation tasks in different ranges, the first-stage parallel computation coarse grains comprise a plurality of second-stage parallel computation coarse grains and a computation task for computing electromagnetic fields and current distributions of the integrated circuit layer by using two-dimensional finite elements, parallel computation grains belonging to the same stage can be computed in parallel, the first-stage parallel computation grains can be allocated with a plurality of computation processes, and the second-stage parallel computation grains can be allocated with only one computation process.
9. The integrated circuit interlayer coupling dynamic applied coarse grain parallel iteration device of claim 7, wherein the specific step of the second stage parallel computation coarse grain comprises:
step 1, calculating an electric field generated by a point current source at a field point, wherein an electric field expression generated by the point current source at the field point is a special analytical expression formed according to a layered special structure of an integrated circuit, and the current sources of the multilayer integrated circuit are distributed in a layered manner, namely the current density distributed on each metal layer of the integrated circuit layout with the complex shape is only equal to that of each metal layer of the integrated circuit layout with the complex shapexAndyis related tozIndependently, the current density distribution is onlyx, yA function of (a);
step 2, taking an electric field expression generated by the point current source at the field point as an integrand function of two-dimensional Gaussian integration, and calculating fields generated by the surface current source of the simple-shaped polygon at the same position based on a linear superposition principle of the fields, wherein the field expression comprises the following steps: the field generated by the current source in the two-dimensional plane S at any point in space can be calculated by the two-dimensional gaussian integral:
wherein the content of the first and second substances,at any point in space for the current source within the two-dimensional plane S (x,y,z) The field that is generated is,is an arbitrary position within the two-dimensional surface S: (u,v) At any point in space (a)x,y,z) The expression of the field that is generated,representing a gaussian integration point corresponding to a two-dimensional gaussian integration in the two-dimensional plane S,p,qrespectively representu,vIn the first directionpA first, aqThe number of the Gaussian integration points is equal to the total number of the points,is the weight factor corresponding to the gaussian integral point;
step 3, calculating fields generated by the current on the simple-shaped polygon at different positions of other layers of the integrated circuit, and determining the fields generated by the current on the simple-shaped polygon divided on the layout of other layers of the integrated circuit based on the linear superposition principle of the fields;
step 4, determining the first step based on the linear superposition principle of the fieldlIs layered onmThe electric field generated by the source layer.
10. The apparatus of claim 7, wherein the formula for computing the weighted CPU time of the parallel computation grain is:in the formula:is as followsiThe weighted CPU times of the particles are calculated in parallel,is as followsiThe CPU time of a single calculation of a grain is calculated in parallel,is as followsiThe number of particle executions is calculated in parallel.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110425242.5A CN112989756B (en) | 2021-04-20 | 2021-04-20 | Coarse grain parallel iteration method and device for dynamically applying coupling between layers of integrated circuit |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110425242.5A CN112989756B (en) | 2021-04-20 | 2021-04-20 | Coarse grain parallel iteration method and device for dynamically applying coupling between layers of integrated circuit |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112989756A true CN112989756A (en) | 2021-06-18 |
CN112989756B CN112989756B (en) | 2021-08-10 |
Family
ID=76341333
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110425242.5A Active CN112989756B (en) | 2021-04-20 | 2021-04-20 | Coarse grain parallel iteration method and device for dynamically applying coupling between layers of integrated circuit |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112989756B (en) |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106095574A (en) * | 2016-06-13 | 2016-11-09 | 北京唯智佳辰科技发展有限责任公司 | Magnanimity calculates coarse granule Parallel Implementation and calculates task stochastic and dynamic distribution method |
US20180275300A1 (en) * | 2017-03-24 | 2018-09-27 | Volkan AKCELIK | Full Wavefield Inversion with Reflected Seismic Data Starting from a Poor Velocity Model |
CN111814332A (en) * | 2020-07-08 | 2020-10-23 | 上海雪湖科技有限公司 | PML boundary three-dimensional seismic wave propagation simulation method based on FPGA |
CN111881641A (en) * | 2020-06-08 | 2020-11-03 | 北京智芯仿真科技有限公司 | Multi-process self-adaptive distribution multi-layer VLSI field coupling method |
CN112287625A (en) * | 2020-12-21 | 2021-01-29 | 北京智芯仿真科技有限公司 | Parallel analysis method and system for direct current voltage drop of system-level integrated circuit |
-
2021
- 2021-04-20 CN CN202110425242.5A patent/CN112989756B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106095574A (en) * | 2016-06-13 | 2016-11-09 | 北京唯智佳辰科技发展有限责任公司 | Magnanimity calculates coarse granule Parallel Implementation and calculates task stochastic and dynamic distribution method |
US20180275300A1 (en) * | 2017-03-24 | 2018-09-27 | Volkan AKCELIK | Full Wavefield Inversion with Reflected Seismic Data Starting from a Poor Velocity Model |
CN111881641A (en) * | 2020-06-08 | 2020-11-03 | 北京智芯仿真科技有限公司 | Multi-process self-adaptive distribution multi-layer VLSI field coupling method |
CN111814332A (en) * | 2020-07-08 | 2020-10-23 | 上海雪湖科技有限公司 | PML boundary three-dimensional seismic wave propagation simulation method based on FPGA |
CN112287625A (en) * | 2020-12-21 | 2021-01-29 | 北京智芯仿真科技有限公司 | Parallel analysis method and system for direct current voltage drop of system-level integrated circuit |
Non-Patent Citations (1)
Title |
---|
卢章疑等: "并行空间自适应多重网格集成电路热分析方法", 《计算机辅助设计与图形学学报》 * |
Also Published As
Publication number | Publication date |
---|---|
CN112989756B (en) | 2021-08-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112989677B (en) | Iteration method and device for accumulation calculation of interlayer coupling part of large-scale integrated circuit | |
CN112836466B (en) | Two-dimensional fast iteration method and device for three-dimensional large-scale integrated circuit electromagnetic response | |
US8423328B2 (en) | Method of distributing a random variable using statistically correct spatial interpolation continuously with spatially inhomogeneous statistical correlation versus distance, standard deviation, and mean | |
US10121279B1 (en) | Systems and methods for generating a mesh | |
TWI421703B (en) | Simulation techniques | |
US20090171636A1 (en) | High-speed operation method for coupled equations based on finite element method and boundary element method | |
CN112989678B (en) | Coarse grain parallel iteration method and device for integrated circuit interlayer coupling part accumulation | |
Nayak et al. | Fractal dimension-based generalized box-counting technique with application to grayscale images | |
CN112989756B (en) | Coarse grain parallel iteration method and device for dynamically applying coupling between layers of integrated circuit | |
CN112818585B (en) | Method and device for dividing iterative computation parallel particles of integrated circuit interlayer coupling | |
CN112232001B (en) | Self-adaptive determination method and system for ultra-wideband resonance response of integrated circuit | |
CN112818633B (en) | Iterative method and device for integrated circuit current distribution dynamically applied by interlayer coupling | |
CN112989676B (en) | Iterative method and device for instantly updating current distribution of integrated circuit by interlayer coupling | |
CN112989675B (en) | Coarse grain parallel iteration method and device for integrated circuit interlayer coupling instant update | |
CN116776695A (en) | One-dimensional electromagnetic calculation method, system and equipment based on ultra-high-order finite element technology | |
CN112989735B (en) | Iterative solution method and device for interlayer coupling of multilayer very large scale integrated circuit | |
CN112818584B (en) | Space electromagnetic radiation computing system and method for integrated circuit | |
Lavaei et al. | Topology optimization of continuum structures using element free Galerkin method on irregular nodal distribution | |
US10810339B1 (en) | Determination of dimensional changes of features across mask pattern simulation fields | |
CN113449838B (en) | Biological particle cluster construction method based on BCCA optimization model | |
CN116644646B (en) | Method, device, equipment and storage medium for mesoscopic reconstruction of rock | |
Delgado et al. | Fast monostatic RCS computation using the near-field sparse approximate inverse and the multilevel fast multipole algorithm | |
CN107562690A (en) | A kind of AIM PO mixed methods based on GPU | |
Mondal et al. | Efficient hierarchical discretization of off-chip power delivery network geometries for 2.5 D electrical analysis | |
Tsamis et al. | The Marching Shape 3D: Extensions of the Ice Ray Shape Grammar |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |