JP5066866B2 - Parallel processing method - Google Patents

Description

The present invention relates to a parallel processing method using a plurality of central processing units.
In recent years, semiconductor devices (LSIs) have been increased in scale and integration. Therefore, the processing time for converting layout data created based on logic circuit data into drawing data for drawing a mask or the like necessary for manufacturing a semiconductor device tends to be long. It is hoped to do.

  In general, the specifications of semiconductor devices (LSIs) are determined according to the application used, that is, mounting functions and operation speeds, and logic design and circuit design are performed using CAD (Computer Aided Design) devices based on the specifications. Done. Then, layout data that is graphic data of the semiconductor device in which elements are arranged based on the logic circuit data designed for the circuit is created. This layout data is converted into drawing data that can be input to a drawing apparatus (for example, an EB drawing exposure apparatus) that actually draws on a mask, reticle, or wafer in a computer for data conversion. By inputting this drawing data to the drawing apparatus, a mask or the like is drawn and used for manufacturing a semiconductor device.

  By the way, the processing time for converting from layout data to drawing data tends to become longer as the semiconductor device becomes larger and highly integrated, and this processing is performed by parallel processing using a plurality of central processing units (CPUs). It has also been proposed to reduce time.

  FIG. 7 is a flowchart showing a data conversion mode by such a computer. As shown in the figure, this computer inputs a GDSII format file as layout data (step 91), converts the file into an internal format, and sets a plurality of data areas to be processed in accordance with the conversion. The data is divided into divided data areas, and the divided data areas (JOBs) are allocated to the central processing units in the order of appearance (step 92). Specifically, for example, when the output after conversion is in the MEBES format, a plurality of vertical stripes obtained by dividing a data area are divided horizontally at equal intervals sequentially from the upper left side when viewed from the front of the data area. Is divided into divided data areas. In the drawing, as an example of such division, an example of division into three divided data areas “1” to “3” is illustrated.

  Then, the computer executes graphic processing such as logical product processing, logical sum processing, resizing processing, rectangular division processing, etc. in the divided data area by parallel processing by each central processing unit (step 93).

  Subsequently, the computer executes a registration process for integrating the divided data areas individually subjected to graphic processing by each central processing unit (step 94), and performs formatting into required drawing data (for example, MEBES format file) (step 95). ).

  By the way, in the case of graphic processing, this computer allocates divided divided data areas to each central processing unit in the order of appearance, and performs parallel processing by these central processing units. On the other hand, each divided data region forms a data region having a different specific gravity of the calculation amount, that is, a data region having a different calculation amount, in accordance with the line segment density or pattern density included in the divided data region. Therefore, if there is a data area in which the specific gravity of the calculation amount becomes heavy in the divided data area that appears in the final stage of parallel processing (for example, on the lower right side), the central processing unit of one central processing unit to which a particularly heavy data area is allocated. Even if the processing time becomes long and all the other central processing units have completed the graphic processing, the computer waits for completion of the one central processing unit. This is an inefficient state as parallel processing by a plurality of central processing units.

  FIG. 8 is an explanatory diagram illustrating a parallel processing mode of three divided data areas “1” to “3” by two central processing units as inefficient parallel processing by a plurality of central processing units. In the figure, the horizontal axis represents the processing time of each central processing unit. The divided data area “1” that appears first is a data area where the specific gravity of the calculation amount is light, and the divided data area “2” that appears next is a data area where the specific gravity of the calculation amount is slightly heavy. . Further, the divided data area “3” is a data area where the specific gravity of the calculation amount is heavy.

  Here, when processing by the two central processing units is started, a divided data area “1” is allocated to one central processing unit (hereinafter referred to as “CPU1”), and another central processing unit (hereinafter referred to as “CPU 1”). The divided data area “2” is allocated to the CPU 2 ”. Then, the last divided data area “3” is allocated to the CPU 1 that has completed the graphic processing of the divided data area “1” at time t11. Therefore, although the CPU 2 completes the graphic processing of the divided data area “2” at the time t12, the CPU 1 to which the divided data area “3” of the data area with a high specific gravity of the calculation amount is allocated has the graphic processing at the time t13. To complete. As a result, a waiting time of time T11 from time t12 to t13 occurs in the CPU 2, and it is confirmed that the parallel processing is in an inefficient state.

  For example, in the case of layout data in which the processing time of the entire parallel processing is a class of 2 to 3 hours, there is a case where a time difference of 20 to 30 minutes at the maximum occurs in the operation completion time between CPUs. Therefore, if such a time difference can be filled, a reduction in processing time of 5 to 15% per layout data is expected.

  An object of the present invention is to convert a divided data area obtained by dividing a plurality of data areas based on layout data of a semiconductor device into drawing data by efficiently performing graphic processing by parallel processing by a plurality of central processing units. It is to provide a parallel processing method that can be used.

In order to solve the above problems, the invention according to claim 1 and claim 2 divides a data area based on layout data of a semiconductor device into a plurality of divided data areas, and performs parallel processing by a plurality of central processing units. In the parallel processing method in which the plurality of divided data areas are subjected to graphic processing and converted into drawing data, the graphic information contained in each of the divided data areas is extracted, and the amount of calculation for each of the divided data areas a calculating step of calculating a specific gravity of, based on the specific gravity of the calculated amount of each divided data area the calculated, Ru and a processing order determination step of allocating said plurality of central processing units.

According to the above SL constituent, when parallel processing by the plurality of central processing units, each central processing unit, the divided data area is allocated on the basis of the specific gravity of the calculation amount is graphic processing. Therefore, the divided data area processed by each central processing unit at the final stage is reduced in the specific gravity of the calculation amount, so that the graphic processing of another central processing unit is completed during the graphic processing of one central processing unit. Even if there is no divided data area allocated to the other central processing unit, the waiting time generated in the other central processing unit is reduced. Accordingly, the plurality of divided divided data areas can be efficiently subjected to graphic processing and converted into drawing data by parallel processing by a plurality of central processing units.

In particular, according to claim 1 invention comprises a comparison step of the specific gravity of the calculated amount of the divided data area is pre-Symbol calculated with a predetermined determination threshold magnitude comparison, the processing order determining step, the predetermined The gist is to allocate only the divided data areas having the specific gravity of the calculation amount larger than the determination threshold to the plurality of central processing units in the order of the specific gravity of the calculation amount.

  According to this configuration, in the parallel processing by the plurality of central processing units, each central processing unit includes only the divided data areas having the specific gravity of the calculation amount larger than the predetermined determination threshold, in the order of the large calculation amount specific gravity. Allocated by Therefore, for the divided data area having a specific gravity of the calculation amount smaller than the predetermined determination threshold, that is, the divided data area having a small influence on the waiting time generated in other central processing units, the calculation amount has a higher specific gravity in order. Allocation to each central processing unit is omitted. For this reason, in order to assign to each central processing unit, for example, it is not necessary to rearrange all the divided data areas in the order of the specific gravity of the calculation amount, and an increase in calculation load can be avoided.

In particular, the invention described in claim 2, prior Symbol calculation step, in parallel the in graphics processing from one direction of the divided data areas by each central processing unit, contains the other direction in the divided data area The gist of extracting the graphic information is to calculate the specific gravity of the calculation amount of each divided data area.

According to the configuration, in parallel with the graphic processing from one direction of each divided data area by each central processing unit, the graphic information included in each divided data area is extracted from the other direction, and each divided data area is extracted. By calculating the specific gravity of the calculation amount of the data area, an increase in the processing time as a whole can be suppressed even if the calculation process is added.
According to a third aspect of the present invention, in the parallel processing method according to the first aspect, the calculation step is performed in the other direction in parallel with the graphic processing from one direction of the divided data areas by the central processing units. The gist of extracting the graphic information included in each of the divided data areas is to calculate the specific gravity of the calculation amount of each of the divided data areas.
According to the configuration, in parallel with the graphic processing from one direction of each divided data area by each central processing unit, the graphic information included in each divided data area is extracted from the other direction, and each divided data area is extracted. By calculating the specific gravity of the calculation amount of the data area, an increase in the processing time as a whole can be suppressed even if the calculation process is added.

The invention according to claim 4 is the parallel processing method according to any one of claims 1 to 3, wherein the graphic information is at least one of a line segment density and a pattern density. To do.
According to a fifth aspect of the present invention, a data area based on layout data of a semiconductor device is divided into a plurality of divided data areas in a lattice shape, and the plurality of divided data divided by parallel processing by a plurality of central processing units. In a parallel processing method of performing graphic processing on an area and converting it into drawing data, a calculation step of extracting graphic information included in each divided data area and calculating a specific gravity of a calculation amount of each divided data area; and A comparison stage that compares the calculated specific gravity of each divided data area with a predetermined determination threshold, and a divided data area having a specific gravity greater than the predetermined determination threshold from the center of the data area. given when held within a radius in the order of the divided data areas arranged spirally on one side rotational direction from the center of the data area, the plurality of central processing units And summarized in that and a processing order determination step of allocating.

According to the same configuration, when a divided data area having a specific gravity with a calculation amount larger than the predetermined determination threshold is within a predetermined radius from the center of the data area, each central processing unit includes the data The divided data areas arranged in a spiral shape in the one-side rotation direction from the center of the area are allocated in the order, and graphic processing is performed. In general, in the case of a semiconductor device having an arithmetic processing function block, the arithmetic processing function block has a high layout scale and pattern density, while an external input / output functional block has a low layout scale and pattern density. Yes. Accordingly, when a divided data area having a specific gravity with a calculation amount larger than the predetermined determination threshold is within a predetermined radius from the center of the data area, the arithmetic processing function block is located at the center of the data area. It is presumed that the layout is arranged. At this time, each of the central processing units is assigned in the order of the divided data areas arranged in a spiral shape in the one-side rotation direction from the center of the data area, so that the order of the divided data areas having a large specific gravity of the calculation amount is approximately. Is allocated and processed in graphics. Therefore, the divided data area processed by each central processing unit at the final stage is reduced in the specific gravity of the calculation amount, so that the graphic processing of another central processing unit is completed during the graphic processing of one central processing unit. Even if there is no divided data area allocated to the other central processing unit, the waiting time generated in the other central processing unit is reduced. Accordingly, the plurality of divided divided data areas can be efficiently subjected to graphic processing and converted into drawing data by parallel processing by a plurality of central processing units.

  In the present invention, parallel data can be converted into drawing data by efficiently performing graphic processing on a divided data area formed by dividing a plurality of data areas based on layout data of a semiconductor device by parallel processing by a plurality of central processing units. A processing method can be provided.

(First embodiment)
Hereinafter, a first embodiment of the present invention will be described with reference to the drawings.
FIG. 1 is an explanatory diagram schematically showing a parallel processing method according to the present embodiment. As shown in the figure, this parallel processing system includes a plurality of (for example, four) processors 1 to 4 as central processing units and a parallel processing computer 11 that controls these processors 1 to 4 in an integrated manner. Composed. This system may be a so-called multi-CPU machine in which the processors 1 to 4 and the parallel processing computer 11 are mounted on one computer. For example, a plurality of computers (4 computers) in a LAN environment may be used. May be a machine. Although not shown in FIG. 1, in any configuration, a storage area (ROM area) for storing a program related to the processing of the processors 1 to 4 and the parallel processing computer 11 and a processing result are temporarily stored. There is a working area (memory) for storing.

  The parallel processing computer 11 inputs a GDSII format file as layout data, converts the file into an internal format, and divides the data area 12 to be processed into a plurality of divided data areas in accordance with the conversion. To do. Here, the plurality of divided data areas are vertical stripes L1 to L8 obtained by dividing the data area 12 into a plurality of parts. These vertical stripes L1 to L8 are arranged in order from the left to the right in FIG.

  Next, the parallel processing computer 11 performs a graphic inspection for extracting graphic information for each of the vertical stripes L1 to L8. Specifically, the parallel processing computer 11 uses the intermediate data (line segment information obtained by the slit method or the raster scan method) of the graphic processing of each vertical stripe L1 to L8 so that the burden of information extraction is reduced as much as possible. Extracted as graphic information of the vertical stripes L1 to L8.

Subsequently, the parallel processing computer 11 calculates the specific gravity of the calculation amount of each vertical stripe L1 to L8 based on the graphic information of each vertical stripe L1 to L8 (calculation stage).
The parallel processing computer 11 compares the specific gravity of the calculation amount of each vertical stripe L1 to L8 with a predetermined determination threshold (comparison stage). Then, the parallel processing computer 11 selects only the vertical stripes L1 to L8 having a specific gravity greater than the predetermined determination threshold in the order of the vertical stripes L1 to L8 having the large specific gravity. To 4 (processing order determination stage). That is, the parallel processing computer 11 determines the priority in units of the vertical stripes L1 to L8, and the processors 1 to 4 use the vertical stripes L1 to L8 as one processing unit.

  In FIG. 1, for two vertical stripes L4 and L7 having a specific gravity of a calculation amount larger than the predetermined determination threshold, the order of the vertical stripes L7, L4 to L1 and L2 having a large specific gravity of the calculation amount (circled numbers). An example in which the processors 1 to 4 are allocated in ascending order) is shown. Note that the remaining four vertical stripes L3, L5, L6, and L8 having a specific gravity of a calculation amount smaller than the predetermined determination threshold are not rearranged in the order of the specific gravity of the calculation amount. For example, the codes are allocated to the processors 1 to 4 as they are in ascending order. This is to avoid an increase in calculation load due to, for example, rearranging all the vertical stripes L1 to L8 in the descending order of the specific gravity of the calculation amount.

  Then, the processors 1 to 4 to which the vertical stripes L1 to L8 are allocated in the above-described manner perform graphic processing of the corresponding vertical stripes L1 to L8 in parallel with each other, for example, logical product processing, logical sum processing, resizing processing, rectangular Perform graphic operations such as split processing.

  As described above, when the graphic processing of all the vertical stripes L1 to L8 is completed, the parallel processing computer 11 integrates the vertical stripes L1 to L8 individually processed by the processors 1 to 4 to obtain required drawing data ( For example, format to MEBES format file).

  FIG. 2 is a flowchart summarizing such a parallel processing method. As shown in the figure, this system inputs a GDSII format file as layout data (step 11), converts the file into an internal format, and sets a plurality of data areas 12 to be processed in accordance with the conversion. The vertical stripes L1 to L8 are divided (step 12).

  Subsequently, this system performs a graphic inspection for extracting graphic information for each of the vertical stripes L1 to L8 (step 13), and based on the graphic information of each vertical stripe L1 to L8, the vertical stripes L1 to L8. The specific gravity of the calculation amount of L8 is calculated, and further assigned to the plurality of processors 1 to 4 in the order of the vertical stripes L1 to L8 having a larger specific gravity of the calculation amount (step 14). At this time, the flow order of step 12 and step 13 may be switched as necessary to perform the process.

Then, this system executes graphic processing of the vertical stripes L1 to L8 by parallel processing by the processors 1 to 4 (step 15).
Subsequently, the system executes a registration process for integrating the divided data areas individually subjected to graphic processing by the processors 1 to 4 (step 16), and performs formatting into necessary drawing data (for example, MEBES format file) ( Step 17).

Next, the operation of the present embodiment will be described with a part of the configuration simplified for convenience.
FIG. 3 is an explanatory diagram for explaining a mode in which three divided data areas “1” to “3” as a plurality of vertical stripes are processed in parallel by two CPUs 1 and 2 as a plurality of processors. In the figure, the horizontal axis represents the processing time of each processor. Further, the divided data area “1” is a data area where the specific gravity of the calculation amount is slightly heavy, and the divided data area “2” is a data area where the specific gravity of the calculation amount is light. Further, the divided data area “3” is a data area where the specific gravity of the calculation amount is heavy.

  Here, when the processing by this system is started, the divided data area “3” having a heavy (heaviest) specific gravity of the calculation amount is allocated to the CPU 1 and the specific gravity of the calculation amount is slightly heavy (next heavy) to the CPU 2. ) The divided data area “1” is allocated. Then, the CPU 2 that has completed the graphic processing of the divided data area “1” at the time t1 is assigned the divided data area “3” that has a light specific gravity (lightest). Then, CPU1 and CPU2 complete their graphic processing at approximately the same time. Thereby, the waiting time which generate | occur | produces in each of CPU1 and CPU2 is suppressed.

As described above in detail, according to the present embodiment, the following effects can be obtained.
(1) In this embodiment, in parallel processing by a plurality of processors 1 to 4, each processor 1 to 4 is assigned in the order of vertical stripes (divided data areas) having a large specific gravity, and graphic processing is performed. . In other words, each vertical stripe is ordered in advance according to the specific gravity of the calculation amount, and each processor 1 to 4 performs the graphic processing in advance from the vertical stripe with the large specific gravity of the calculation amount, thereby performing the processing. In the final stage, vertical stripes with a small specific gravity can be concentrated to perform graphic processing. Then, the specific gravity of the amount of calculation of the vertical stripe processed by each processor 1-4 in the final stage is reduced, so that the graphic processing of the other processors 1-4 is completed during the graphic processing of one processor 1-4, and new processing is performed. Even if there is no vertical stripe allocated to the other processor, the waiting time generated in the other processors 1 to 4 is reduced, and the graphic processing of these processors 1 to 4 can be completed almost simultaneously. Therefore, the plurality of divided vertical stripes can be efficiently converted into drawing data by performing graphic processing efficiently by parallel processing by the plurality of processors 1 to 4.

  (2) In addition, since the processing time for converting layout data to drawing data can be shortened, for example, user inspection of pre-drawing data that has become essential in foundry business (COT business), etc. Timely supply to the user by shortening the processing time can be realized. In addition, shortening the processing time and speedily confirming the inspection results can contribute to shortening the delivery time of the reticle.

  (3) In the present embodiment, in parallel processing by a plurality of processors 1 to 4, each processor 1 to 4 has a large specific gravity only for vertical stripes having a specific gravity greater than a predetermined determination threshold. Allocated in order. Therefore, for vertical stripes having a specific gravity of a calculation amount smaller than the predetermined determination threshold, that is, vertical stripes having a small influence on the waiting time generated in the other processors 1 to 4, in the order of the large calculation amount specific gravity. Allocation to each of the processors 1 to 4 is omitted. For this reason, in order to allocate to each of the processors 1 to 4, for example, it is not necessary to rearrange all the vertical stripes in the descending order of the specific gravity of the calculation amount, and an increase in calculation load can be avoided.

(4) In the present embodiment, the divided data area that is an allocation unit (processing unit) to each of the processors 1 to 4 can be formed into a vertical stripe having an extremely simple shape.
(Second Embodiment)
Hereinafter, a second embodiment of the present invention will be described with reference to the drawings. Note that the second embodiment has a configuration that is changed to perform graphic inspection by the parallel processing computer 11 in parallel with the parallel processing by each of the processors 1 to 4, and therefore the same parts as the first embodiment. Will not be described in detail.

  FIG. 4 is an explanatory diagram schematically showing the parallel processing method according to the present embodiment. As shown in the figure, the parallel processing computer 11 inputs a GDSII format file as layout data, converts the file into an internal format, and sets a plurality of data areas 12 to be processed in accordance with the conversion. Divide into divided data areas. Here, the plurality of divided data areas are grid-like areas 13 formed by dividing the vertical stripes L1 to L8 obtained by dividing the data area 12 into a plurality of parts at intervals at equal intervals.

  Next, the parallel processing computer 11 assigns the processors 1 to 4 as they are in the order from the upper left area 13-1 in FIG. 4 to the lower side, and in the order from the lower right area 13-2 in FIG. A graphic inspection for extracting graphic information is performed for each area 13. Specifically, the parallel processing computer 11 is configured so that the accuracy of information extraction is increased so that the intermediate data (graphic segment density (number of line segments, etc.) information obtained by the slit method or the raster scan method) The pattern density (number of patterns, pattern occupancy) in the area is extracted as graphic information of the area 13. Then, the parallel processing computer 11 calculates the specific gravity of the calculation amount of each area 13 based on the graphic information of each area 13 (calculation stage). Then, the parallel processing computer 11 assigns the plurality of processors 1 to 4 in the order of the region 13 where the specific gravity of the calculation amount is large (processing order determination stage). At this time, since the accuracy of information extraction in each region 13 is high, the processing order can be adjusted more finely.

  On the other hand, the processors 1 to 4 to which the area 13 is allocated in the above-described manner perform graphic processing on the corresponding area 13 in parallel with each other. That is, the parallel processing computer 11 performs graphic inspection and processing order determination in parallel with the parallel processing by the processors 1 to 4. In FIG. 4, the area 13 that is assigned to the processors 1 to 4 as they are at the initial stage of the processing and has been processed by the processors 1 to 4 is shown with a mesh pattern. Using the processing time of these areas 13, the parallel processing computer 11 performs graphic inspection and processing order determination.

  In addition, since the areas 13 are initially allocated to the processors 1 to 4 in the order in which the data areas 12 are arranged, the allocation of the areas 13 in the descending order of the calculation amount relative to the processors 1 to 4 It may start when the area 13 processed by the parallel processing computer 11 from the reverse direction catches up with the area 13 processed from one to four by the one direction, or the area processed by the parallel processing computer 11 from the reverse direction. You may start when 13 reaches a certain number.

  In FIG. 4, an example in which the regions 13 a, 13 b, 13 c, and 13 d having a large specific gravity in the calculation amount are allocated to the processors 1 to 4 in the order (in ascending order of circled numbers) in the middle of the processing of each processor 1 to 4. Show.

  As described above, when the graphic processing of all the regions 13 is completed, the parallel processing computer 11 integrates the regions 13 individually processed by the processors 1 to 4 into the required drawing data (for example, MEBES format file). And formatting.

As described above in detail, according to the present embodiment, in addition to the effects (1) and (2) in the first embodiment, the following effects can be obtained.
(1) In this embodiment, in parallel with the graphic processing from one direction of each area 13 by each of the processors 1 to 4, graphic information contained in each area 13 is extracted from the other direction, and each area 13 is extracted. By calculating the specific gravity of the calculation amount, even if the calculation process is added, an increase in the processing time as a whole can be suppressed. This is particularly effective when various data created in the graphic information extraction stage can be used only by determining the processing order.

(2) In the present embodiment, the divided data area that is an allocation unit (processing unit) to each of the processors 1 to 4 can be a lattice-shaped area 13 having a very simple shape.
(Third embodiment)
Hereinafter, a third embodiment of the present invention will be described with reference to the drawings. The third embodiment is changed so that the order of the areas 13 to be allocated to the respective processors 1 to 4 is determined by determining the degree of concentration of the area 13 having a large calculation weight specific gravity at the center of the data area 12. Therefore, detailed description of the same parts as those in the first and second embodiments is omitted.

  FIG. 5 is an explanatory diagram schematically showing the parallel processing method according to the present embodiment. As shown in the figure, the parallel processing computer 11 inputs a GDSII format file as layout data, converts the file into an internal format, and sets a plurality of data areas 12 to be processed in accordance with the conversion. Divide into divided data areas. Here, the plurality of divided data areas are grid-like areas 13 formed by dividing the vertical stripes L1 to L8 obtained by dividing the data area 12 into a plurality of parts at intervals at equal intervals.

  Next, the parallel processing computer 11 performs a graphic inspection for extracting graphic information for each region 13. Various data created in this graphic information extraction stage is stored in the memory and can be used in subsequent graphic processing. The increase in the computation load, that is, the increase in the processing time as a whole is suppressed.

  Subsequently, the parallel processing computer 11 calculates the specific gravity of the calculation amount of each area 13 based on the graphic information of each area 13 (calculation stage). The parallel processing computer 11 compares the specific gravity of the calculation amount of each region 13 with a predetermined determination threshold (comparison stage), and extracts the region 13 having the specific gravity of the calculation amount larger than the predetermined determination threshold. .

  Next, the parallel processing computer 11 determines whether or not the area 13 having a large specific gravity of the extracted calculation amount is within an area of, for example, a radius of 60% from the center of the data area 12. Then, the parallel processing computer 11 rotates one side from the center of the data area 12 when the region 13 having a large specific gravity of the extracted calculation amount is within an area of 60% radius from the center of the data area 12. The plurality of processors 1 to 4 are allocated in the order of the regions 13 arranged in a spiral shape in the direction (counterclockwise rotation direction indicated by the broken line in FIG. 5) (ascending order of the circled numbers) (processing order determination stage). In the case of a system LSI, the arithmetic processing function block (CPU section) has a high layout scale and pattern density, while the external input / output function block (I / O section) has a low layout scale and pattern density. It depends on what is obtained as a result. That is, when the area 13 having a large specific gravity of the calculation amount is within the radius of 60% from the center of the data area 12, the parallel processing computer 11 has the arithmetic processing function block at the center of the data area 12. It is determined that the layout is arranged, and the priority order of each region 13 is determined in the above-described manner based on the center of the data area 12.

Then, the processors 1 to 4 to which the area 13 is allocated in the above-described manner perform graphic processing on the corresponding area 13 in parallel with each other.
As described above, when the graphic processing of all the regions 13 is completed, the parallel processing computer 11 integrates the regions 13 individually processed by the processors 1 to 4 into the required drawing data (for example, MEBES format file). And formatting.

  FIG. 6 is a flowchart collectively showing such a parallel processing method. As shown in the figure, this system inputs a GDSII format file as layout data (step 21), converts the file into an internal format, and sets a plurality of data areas 12 to be processed in accordance with the conversion. (Step 22).

  Subsequently, this system performs a graphic inspection for extracting the graphic information for each region 13 (step 23), and calculates the specific gravity of the calculation amount of each region 13 based on the graphic information of each region 13. Further, it is determined whether or not the region 13 having a larger specific gravity of the calculation amount is within the area of, for example, a radius of 60% from the center of the data area 12 (step 24).

  In this system, when it is determined that the area 13 having a large specific gravity of the calculation amount is within an area of a radius of 60% from the center of the data area 12, the one-side rotation direction from the center of the data area 12 is determined. Are allocated to the processors 1 to 4 in the order of the regions 13 arranged in a spiral pattern (step 25). On the other hand, if it is determined that they do not fall within an area of a radius of 60% from the center of the data area 12, they are set in advance. The data areas 12 are allocated to the processors 1 to 4 in the arrangement order of the data areas 12, for example, in the order from the upper left area 13-1 in FIG.

And this system performs the graphic process of the area | region 13 by the parallel processing by each processor 1-4 (step 26).
Subsequently, the system executes a registration process for integrating the divided data areas individually subjected to graphic processing by the processors 1 to 4 (step 27), and performs formatting into required drawing data (for example, MEBES format file) ( Step 28).

As described above in detail, according to the present embodiment, the following effects can be obtained in addition to the effect (2) in the first embodiment.
(1) In the present embodiment, when the area 13 having a specific gravity with a calculation amount larger than the predetermined determination threshold is within the area of a radius of 60% from the center of the data area 12, that is, the arithmetic processing function block When it is determined that the layout is arranged at the center of the area 12, the processors 1 to 4 are assigned to the processors 1 to 4 in the order of the regions 13 arranged in a spiral shape from the center of the data area 12 in the one side rotation direction. As a result, the processors 1 to 4 are allocated in the order of the region 13 having a large specific gravity of the calculation amount, and are subjected to graphic processing. Accordingly, the area 13 processed in the final stage by each of the processors 1 to 4 is reduced in the specific gravity of the calculation amount, so that the graphic processing of the other processors 1 to 4 is completed during the graphic processing of one processor 1 to 4. Even if there is no newly allocated area 13, the waiting time generated in the other processors 1 to 4 is reduced, and the graphic processing of these processors 1 to 4 can be completed almost simultaneously. Accordingly, the plurality of divided areas 13 can be efficiently subjected to graphic processing by parallel processing by a plurality of processors 1 to 4 and converted into drawing data.

  (2) In the present embodiment, the respective areas 13 are allocated to the processors 1 to 4 in the order of the large specific gravity of the calculation amount without strictly rearranging the respective areas 13 based on the specific gravity of the calculation amount of each area 13. be able to. For this reason, the calculation load for rearrangement can be reduced.

In addition, you may change the said embodiment as follows.
In the first embodiment, the divided data area that is an allocation unit (processing unit) to each of the processors 1 to 4 may be the area 13. In this case, the parallel processing computer 11 calculates the specific gravity of the calculation amount of each area 13 based on the graphic information of each area 13. Then, the parallel processing computer 11 assigns the plurality of processors 1 to 4 in the order of the region 13 where the specific gravity of the calculation amount is large. In addition, if various data created in the graphic information extraction stage of the area 13 are stored in the memory and used in subsequent graphic processing, even if the graphic information extraction processing is added, the entire data accompanying this will be The increase in the computation load, that is, the increase in the processing time as a whole is suppressed.

  In the second embodiment, if various data created in the graphic information extraction stage of the region 13 are stored in a memory and used in subsequent graphic processing, the graphic information extraction processing is added. However, an increase in the calculation load as a whole, that is, an increase in the processing time as a whole is suppressed.

  In the second embodiment, the specific gravity of the calculation amount of each region 13 is compared with a predetermined determination threshold value, and only the region 13 having the specific gravity of the calculation amount larger than the predetermined determination threshold value is calculated. May be assigned to the plurality of processors 1 to 4 in descending order of specific gravity.

  In the third embodiment, when the region 13 with a large specific gravity of the extracted calculation amount is within the area of a radius of 60% from the center of the data area 12, the other side from the center of the data area 12 The plurality of processors 1 to 4 may be allocated in the order of the regions 13 arranged in a spiral shape in the rotation direction (clockwise rotation in FIG. 5).

In the third embodiment, the radius relating to the determination of the degree of concentration of the area 13 having a large calculation amount specific gravity at the center of the data area 12 is not limited to 60%.
In the third embodiment, when the CPU unit can obtain information that is within 60% of the chip in advance by some means in the design layout stage, the data area can be obtained without determining the specific gravity of the calculation capability. The graphic processing in parallel may be started in the order of the regions 13 arranged in a spiral shape from the center of 12 in the direction of one side rotation.

  In each of the above embodiments, a characteristic parameter may be extracted as graphic information when calculating the specific gravity of the calculation amount of each divided data area. Specifically, the number of repeated arrangements of the same figure, the occupation rate of the correction prohibition frame, optimized arrangement information unique to the program, and the like.

In each of the above embodiments, the number of processors related to parallel processing may be two to three, or may be five or more.
Next, the technical idea that can be grasped from the above embodiment and other examples will be described below.
(Appendix 1)
A data area based on layout data of a semiconductor device is divided into a plurality of divided data areas, and the divided divided data areas are subjected to graphic processing by parallel processing by a plurality of central processing units, and converted into drawing data. In the parallel processing method,
Calculating the specific gravity of the calculation amount of each divided data area by extracting graphic information included in each divided data area;
A parallel processing method comprising: a processing order determination step for allocating to the plurality of central processing units based on the calculated specific gravity of each divided data area.
(Appendix 2)
In the parallel processing method according to attachment 1 ,
The parallel processing method according to claim 1, wherein the graphic information is at least one of a line segment number and a pattern density.
(Appendix 3)
In the parallel processing method according to attachment 1 or 2 ,
A comparison step of comparing the specific gravity of the calculated amount of each divided data area with a predetermined determination threshold;
In the processing order determination step, only the divided data areas having a specific gravity of a calculation amount larger than the predetermined determination threshold are allocated to the plurality of central processing devices in the order of the high specific gravity of the calculation amount. Parallel processing method.
(Appendix 4)
In the parallel processing method according to any one of appendices 1 to 3 ,
The parallel processing method, wherein the plurality of divided data areas are vertical stripes formed by dividing the data area into a plurality of parts.
(Appendix 5)
In the parallel processing method according to any one of appendices 1 to 3 ,
The parallel processing method, wherein the plurality of divided data areas are data areas obtained by dividing vertical stripes formed by dividing the data area into a plurality of parts at equal intervals.
(Appendix 6)
In the parallel processing method according to any one of appendices 1 to 3 ,
A parallel processing method, wherein graphic information contained in each of the extracted divided data areas is utilized in graphic processing of each central processing unit to which the divided data areas are allocated.
(Appendix 7)
In the parallel processing method according to any one of appendices 1 to 3 ,
In the calculation step, the graphic information included in each divided data area is extracted from the other direction in parallel with the graphic processing from one direction of each divided data area by each central processing unit, and each divided data area is extracted. A parallel processing method characterized by calculating a specific gravity of a calculation amount of a region.
(Appendix 8)
The data area based on the layout data of the semiconductor device is divided into a plurality of divided data areas in a lattice shape, and the divided divided data areas are subjected to graphic processing by parallel processing by a plurality of central processing units, and drawing data In the parallel processing method of converting to
Calculating the specific gravity of the calculation amount of each divided data area by extracting graphic information included in each divided data area;
A comparison step of comparing the calculated specific gravity of each divided data area with a predetermined determination threshold;
When a divided data area having a specific gravity with a calculation amount larger than the predetermined determination threshold is within a predetermined radius from the center of the data area, the data areas are arranged in a spiral from the center of the data area to one side rotation direction. A parallel processing method comprising: a processing order determination step for allocating to the plurality of central processing units in the order of the divided data areas.

Explanatory drawing which shows 1st Embodiment which concerns on this invention. The flowchart which shows the process aspect of 1st Embodiment. Explanatory drawing which shows operation | movement of 1st Embodiment. Explanatory drawing which shows 2nd Embodiment which concerns on this invention. Explanatory drawing which shows 3rd Embodiment which concerns on this invention. The flowchart which shows the process aspect of 3rd Embodiment. The flowchart which shows the process aspect of a prior art form. Explanatory drawing which shows operation | movement of a prior art form.

Explanation of symbols

1-4 Processor 11 Parallel processing computer 12 Data area 13 Area L1-L8 Vertical stripe

Claims (5)

A data area based on layout data of a semiconductor device is divided into a plurality of divided data areas, and the divided divided data areas are subjected to graphic processing by parallel processing by a plurality of central processing units, and converted into drawing data. In the parallel processing method,
Calculating the specific gravity of the calculation amount of each divided data area by extracting graphic information included in each divided data area;
A processing order determination step for allocating to the plurality of central processing units based on the specific gravity of the calculated amount of each divided data area ;
A comparison step of comparing the specific gravity of the calculated amount of each divided data area with a predetermined determination threshold,
In the processing order determination step, only the divided data areas having a specific gravity of a calculation amount larger than the predetermined determination threshold are allocated to the plurality of central processing devices in the order of the high specific gravity of the calculation amount. Parallel processing method.   A data area based on layout data of a semiconductor device is divided into a plurality of divided data areas, and the divided divided data areas are subjected to graphic processing by parallel processing by a plurality of central processing units, and converted into drawing data. In the parallel processing method,
  Calculating the specific gravity of the calculation amount of each divided data area by extracting graphic information included in each divided data area;
  A processing order determination step for allocating to the plurality of central processing units based on the specific gravity of the calculated amount of each divided data area;
  In the calculation step, the graphic information included in each divided data area is extracted from the other direction in parallel with the graphic processing from one direction of each divided data area by each central processing unit, and each divided data area is extracted. A parallel processing method characterized by calculating a specific gravity of a calculation amount of a region. The parallel processing method according to claim 1 ,
In the calculation step, the graphic information included in each divided data area is extracted from the other direction in parallel with the graphic processing from one direction of each divided data area by each central processing unit, and each divided data area is extracted. A parallel processing method characterized by calculating a specific gravity of a calculation amount of a region. In the parallel processing method according to any one of claims 1 to 3 ,
The parallel processing method, wherein the graphic information is at least one of a line segment density and a pattern density. The data area based on the layout data of the semiconductor device is divided into a plurality of divided data areas in a lattice shape, and the divided divided data areas are subjected to graphic processing by parallel processing by a plurality of central processing units, and drawing data In the parallel processing method of converting to
Calculating the specific gravity of the calculation amount of each divided data area by extracting graphic information included in each divided data area;
A comparison step of comparing the calculated specific gravity of each divided data area with a predetermined determination threshold;
When a divided data area having a specific gravity with a calculation amount larger than the predetermined determination threshold is within a predetermined radius from the center of the data area, the data areas are arranged in a spiral from the center of the data area to one side rotation direction. A parallel processing method comprising: a processing order determination step for allocating to the plurality of central processing units in the order of the divided data areas .

Images