TWI717727B - Method of placing macro cells - Google Patents

Abstract

A simulated-evolution-based macro refinement method includes evaluating a score of each placed macro cell to be refined; generating a random number; determining whether the score satisfies a predetermined condition; placing the macro cell into a queue if the score associated with the macro cell satisfies the predetermined condition; and sorting and placing macro cells of the queue according to scores of the macro cells in the queue.

Description

How to set up macro components

The present invention relates to macro settings, and particularly relates to a method of refinement based on simulated-evolution-based macros.

The setting of the macro has a significant effect on the wirelength and routability, and it also affects the placement of standard cells. Since the size of the macro is much larger than the standard component, the problems caused by the macro setting are far more complicated than the standard component setting. When many macros are pre-placed on the chip and cause obstacles, the problem becomes more complicated. When the macro cannot be set in the predetermined position, it will cause longer cable length and serious winding congestion. Therefore, the macro is usually set manually by an experienced engineer. When the number of macros in modern SoC systems becomes very large, relying on experienced engineers to perform large-scale macro settings will make the macro settings inefficient.

To solve this problem, someone previously proposed a simulated annealing algorithm. However, this algorithm takes a lot of time. Since traditional methods cannot effectively improve the setting of macros, it is urgent to propose a novel macro setting mechanism to effectively set macros.

In view of the foregoing, one of the objectives of the embodiments of the present invention is to propose a macro optimization method based on simulation evolution, which performs faster than traditional methods.

The macro optimization method based on simulated evolution proposed in the embodiment of the present invention includes the following steps. Evaluate to get the score of each macro component set, and generate random numbers. Decide whether the score meets the preset conditions. If the corresponding score of the macro component meets the preset conditions, the macro component is put into the queue. According to the scores of the macro components in the queue, sort and set the macro components in the column.

The first A shows a flowchart of a corner-stitching-based macro legalization method 100, and the first B shows a top view of a chip (or microchip) 101, the surface of which can be placed Macro components. The second A to the second C show an example of executing the method 100 of the macro set normalization of the first A. It should be noted that other macro normalization methods different from the macro normalization method 100 (Figure A) can also be used to initially set the macro on the chip 101.

In step 10, a placement region 102 is determined and configured on the chip 101 for setting a macro group containing a plurality of macro elements. In one embodiment, the setting area 102 is determined by using a recursive partition algorithm. Next, in step 11, a plurality of (rectangular) empty tiles 104 are generated on the wafer 101 according to the pre-placed component 103. As illustrated in FIG. 1B, the wafer 101 is provided with preset elements 103 (as shown by the slanted area), and includes a plurality of (rectangular) empty tiles 104 that are not covered by the preset elements 103. The horizontal boundary of the empty brick 104 is obtained by extending the horizontal boundary of the adjacent preset element 103.

Then, for each macro element to be set, steps 12-16 are performed. In step 12, a corner stitching method is used to count the number of bricks 104 in the setting area 102. For details, please refer to the "corner stitching: super-large integrated circuit layout tool" proposed by JK Ousterhout. "Corner Stitching: A Data-Structuring Technique for VLSI Layout Tools)", published in 1984 IEEE TCAD, Volume 3, No. 1, 87~100 page. As illustrated in the second diagram A, the tiles 104 in the setting area 102 are marked as 1, 2, 4, 7, 8, 9, and 11, respectively.

In step 13, trim (or cut) the area of the brick 104 beyond the boundary of the setting area 102, as illustrated in the second diagram B.

In step 14, as illustrated in Figure 2C, for each brick 104 (for example, brick 8), if the size of the macro element is not larger than the brick 104, place the macro element to be set (as shown in the dotted area) Each corner of the brick 104, and the corresponding packing cost is evaluated. For all bricks 104, evaluate the corresponding placement costs. In this way, the macro element is placed in the corresponding position with the smallest cost of placement, and the data of the position is updated (step 17).

If there is no brick 104 that can accommodate the macro element (step 105), the setting area 102 is increased in step 16, and steps 12 to 14 are repeated for the enlarged setting area 102.

The third figure shows a flowchart of a simulated-evolution-based macro refinement method 200 (hereinafter referred to as a macro optimization method) according to an embodiment of the present invention. The macro optimization method 200 can be performed after the macro normalization (for example, the corner stitched macro normalization method 100 shown in the first A).

In step 21, evaluate to obtain the score of each macro component set. Next, in step 22, the score is normalized to be within a preset range, for example, between 0 and 1, thereby generating a normal score.

In step 23, a random number is generated, and the generated random number is also within the same preset range, for example, between 0 and 1. Next, in step 24, it is determined whether the random number is greater than the generated random number (or, in a nutshell, it is determined whether the random number meets a preset condition). If the result of step 24 is affirmative, let the macro components enter the enqueue (step 25), and according to the size of the normal score, sort and set the macro components in the column (step 26); otherwise, the process Skip step 25 and step 26. Repeat steps 23~25 for all macro components set.

其中D代表(待優化的)巨集元件與所屬巨集群之重心(gravity)的距離；如果巨集群僅包含單一巨集元件，該巨集元件設置於重心處，則D的值為零；W代表待優化的巨集元件的相應線長(wirelength)；λ為使用者設定值；k為第三圖之流程的執行序。值得注意的是，於每一次執行後，重新決定巨集群的重心。 In this embodiment, the score function F _{i is used} to generate the aforementioned score, which can be expressed as follows:

Where D represents the distance between the macro element (to be optimized) and the gravity of the macro cluster to which it belongs; if the macro cluster contains only a single macro element and the macro element is set at the center of gravity, the value of D is zero; W Represents the corresponding wire length of the macro component to be optimized; λ is the value set by the user; k is the execution sequence of the process in the third figure. It is worth noting that after each execution, the center of gravity of the giant cluster is determined again.

According to one of the characteristics of this embodiment, the log function of the second term is used to adjust the score to make it larger, so that the preset condition of step 24 (for example, the score is greater than the random number generated) is more easily met. For the first few execution order values k, because the logarithmic function produces a very small value, the value of the second term becomes very large, so when the value of k is very small, the distance D has an influence on the value of the score. However, when the value of k increases, the logarithmic function produces a larger and stable value. Therefore, the value of the second term becomes very small, so that the distance D has no influence on the value of the score. In other words, the larger the execution order value k, the smaller the influence of the distance D on the score value.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the scope of the present invention; all other equivalent changes or modifications made without departing from the spirit of the invention should be included in the following Within the scope of patent application.

100 Large collection method based on corner stitching 11 Configuration setting area 12 Count the number of bricks 13 Trim the bricks 14 Evaluate placement costs 15 Determine whether the placement is feasible 16 Increase the installation area 17 Update information 101 Chip 102 Setting up area 103 Pre-installed components 104 Brick 200 Macro optimization method based on simulation evolution 21 Evaluate to get the score of each macro component set 22 Normalization score 23 Random numbers occurred 24 Determine whether the score is greater than the random number 25 Enter the queue 26 Sort and set up macro components

The first figure A shows the flow chart of the macro set normalization method based on corner stitching. Fig. 1B illustrates a top view of a wafer on which a macro element can be placed. The second A to the second C show an example of executing the macro set normalization method of the first A. The third figure shows a flowchart of a macro optimization method based on simulated evolution according to an embodiment of the present invention.

200 Macro optimization method based on simulation evolution 21 Evaluate to get the score of each macro component set 22 Normalization score 23 Random numbers occurred 24 Determine whether the score is greater than the random number 25 Enter the queue 26 Sort and set up macro components

Claims (9)

A method for setting macro components includes: determining and configuring a setting area on the chip; generating a plurality of bricks, which are not covered by preset components; performing corner stitching-based macro methodization for the macro group, the macro group containing plural numbers Macro element; evaluate to obtain the score of each macro element; normalize the score to be within a preset range, thereby generating a normal score; generate random numbers, which are located within the preset range with the same normal score; Determine whether the normal score is greater than the random number; if the normal score is greater than the random number, let the macro component enter the queue; and according to the score of the macro component in the queue, sort and set the macro in the queue Set components. For example, the method for setting macro elements in claim 1, wherein the corner stitching-based macro methodization includes the following steps: counting the number of the bricks in the setting area; trimming the area of the brick that exceeds the boundary of the setting area; If the size of the macro element is not larger than the brick, place the macro element in each corner of the brick; evaluate to obtain the corresponding placement cost of the macro element placed in the corners; and Place the cost to set up the macro component. For example, the method for setting a macro component in the request item 2 further includes: if the plurality of tiles cannot accommodate the macro component, increasing the setting area. For example, the method of setting a macro element in claim 1, wherein the setting area is determined by using a recursive segmentation algorithm. For example, the method for setting a macro element in claim 1, wherein the horizontal boundary of the brick is obtained by extending the horizontal boundary of the adjacent preset element. For example, the method for setting a macro element in claim 1, wherein the score is generated by a scoring function, and the parameters of the scoring function include the corresponding line length of the macro element, and the distance between the macro element and the center of gravity of the macro cluster, And optimize the execution order of the giant cluster. For example, the method for setting a macro element in the request item 6, wherein the scoring function includes an item for adjusting the score to make it larger, so that the normal score is more likely to be larger than the random number. For example, the method for setting a macro element in claim 6, wherein the center of gravity of the macro cluster needs to be re-determined after each execution. For example, the method for setting a macro element in claim 6, wherein the larger the value of the execution sequence, the smaller the influence of the distance on the score.

Images