JP2842324B2 - Circuit division improvement device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a device for improving circuit division, and more particularly to a device for sequentially improving circuit division in layout design of a large-scale integrated circuit of a standard cell system and a gate array system.

[0002]

2. Description of the Related Art Conventionally, in circuit division (assuming that a circuit is divided into two), an initial division is created by an appropriate method, and then one element is divided into
It has been performed by sequentially repeating the operation of moving from the partial circuit to which it belongs at the time to the other (literature: "VLSI Computer CAD" edited by Hiroshi Yamada, Sangyo Tosho, 1983).

Here, the simplest method of creating the initial division is a method of randomly allocating each element to a partial circuit, and a circuit division improvement process after the initial division is created is essential.
Another typical technique for creating the initial division is a “clustering method”, in which two elements having a strong connection relationship (a large number of commonly connected nets) are merged into one element. The whole process of combining into elements is 2
It repeats until there are two elements, and then determines the two initial partial circuits by checking the history of the merging process and determining which element (element representative) each element is finally merged with. When this clustering method is used,
Generally, local improvement is performed by circuit division improvement processing.

FIG. 5 shows a prior art concerning a circuit division improving measure.
This will be described with reference to FIG.

FIG. 5 shows one bisection of a circuit composed of six elements. The four elements on the left side and the two elements on the right side of the division boundary indicated by broken lines respectively constitute partial circuits. As can be seen from the figure, in this division, a total of four nets, net 1, net 2, net 4, and net 5, intersect the boundary. That is, the number of nets (referred to as "cut number") extending between partial circuits is four. Here, if the element A is moved from the right partial circuit to the left partial circuit, the net crossing the boundary becomes the net 4
And the net 5 in total, and the number of cuts can be reduced by 2 compared to the divided state before the element movement.

In this way, focusing on the fact that the number of cuts changes due to the movement of the element, in the division improvement processing, first, the amount of reduction (referred to as "gain") in the number of cuts relating to the operation of moving each element is determined. For example, as described above, when the element A is moved in FIG. 5, the number of cuts changes from 4 to 2, and the movement gain of the element A becomes 2.

[0007] When the gain of the moving operation for each element is obtained, a policy is based on the principle that "an element having a positive gain and the largest gain is found, the element is moved, and the gain of each element is recalculated." Incremental improvements are made.
At this time, it is usual that the element once moved is removed from the movement target and re-movement is suppressed. In many cases, a restriction (for example, the difference in the number of elements is within a predetermined constant) is imposed on the variation in the number of elements of each partial circuit. to continue.

[0008]

As described above, in the conventional circuit division improving device, whether or not to move an element is determined by the gain of each element, and the priority of the element to be moved is determined. The conventional definition of the gain for the element c is expressed by the following equation.

[0009]

(Equation 1)

Here, N (c) is a set of nets connected to element c, and F (n) is a set of elements connected to net n.
The number of elements belonging to the same partial circuit as the element c, and T (n) is the number of elements connected to the net n that do not belong to the same partial circuit as the element c. As is clear from the equation (1), the gain of the element is determined by the net connected to the element, and other nets are not considered at all. That is, only the neighboring elements are considered in the conventional gain calculation.

FIG. 6 clearly shows the adverse effect caused by such a gain calculation method. In the division shown in Fig.│, the gains of all the elements on the right side are negative, and it is determined that the number of cuts cannot be further reduced by the conventional method.
The division (the number of cuts is 2) in the figure is not an optimal division but only one local optimal solution. It is clear that the optimal solution is a division (the number of cuts is 1) that cuts only the net 6 or only the net 7.

In order to solve the above problem while using the conventional gain calculation method, for example, the element B in FIG. 6 is moved to the left partial circuit, and the number of cuts which is the evaluation function value is once reduced from 2 to 3. After that, the element C must be further moved to the left partial circuit. However, the number of movement operations for increasing the evaluation function value is much larger than the number of movement operations for decreasing the evaluation function value, and as a result, such a method (probably performs element selection stochastically to reduce the number of applications) (If any) causes a significant increase in processing time.

[0013]

SUMMARY OF THE INVENTION A circuit division improving apparatus according to the present invention performs an operation for moving an element included in a circuit from a partial circuit to which the element belongs to another partial circuit for a divided circuit. In a circuit division improving device that reduces an evaluation function value related to a connection relationship between partial circuits by sequentially repeating, an adjacent element selection unit that enumerates elements having a connection relationship with a given element with respect to a given element. , For each of the listed elements, the amount of decrease (gain) in the evaluation function value when the element is moved to another partial circuit
Read-ahead gain calculating means for taking into account elements which do not have a direct connection to the element, moving element selecting means for selecting an element having a positive and maximum reduction amount (gain), Element moving means for moving the selected element to another partial circuit.

Further, the circuit division improving apparatus according to the present invention is arranged such that, for each of the listed elements, the reduction amount (gain) of the evaluation function value when the element is moved to another partial circuit is determined by the element. And non-look-ahead gain calculating means for calculating taking into account only elements having a direct connection with the moving element selecting means based on the gain obtained by the non-look-ahead gain calculating means. Only when an element cannot be selected, the gain is recalculated and output to the moving element selecting means.

[0015]

DESCRIPTION OF THE PREFERRED EMBODIMENTS As described above, in the conventional gain calculation method, only neighboring elements (elements connected to a net directly connected to a certain element) are involved in the gain calculation.

The present invention solves the problem by using a gain calculation method in consideration of an element having no direct connection to each element as shown below (see FIG. 4).
That is, for each element i, the element j adjacent thereto is examined, and if the element j has been moved (the position is fixed), the direct action of the element j on the element i is estimated. If present, element k (≠ i) adjacent to element j
Is calculated, the indirect effect of the element k on the element i is estimated, and the gain of the element i is calculated based on those effects. Expressed correctly by a mathematical expression, the gain of the element i is as follows.

[0017]

(Equation 2)

Here, side (n) is the position (left or right) of the element n, and c (m, n) is the connection strength between the elements m and n (the number of nets connecting m and n). As can be seen from equation (2), the gain calculation takes into account not only the element (j) having an indirect connection with the element i, but also the element (k) having an indirect connection.

The following describes that the problem of the conventional method shown in FIG. 6 can be solved by the present invention.

The element A in FIG. 6 is excluded because it is not to be moved (since it is not divided into two when moved). When the gains of the remaining four elements B, C, D, and E are calculated based on equation (2), the gains are -3/4, 7/10,
Element C having a positive maximum value of -3/4 and -1/2
Moves to the left, and the divided state of FIG. 7 is obtained (the number of cuts is 6, and the position of the element C is fixed). Subsequently,
When the gain of each element in FIG. 7 is calculated, the elements A, B, D, and E excluding the element C which has been excluded from the movement due to the movement.
Are -6/5, 3, 1, and 1/2, respectively, so that the element B moves to the partial circuit on the left side, and enters the division state of FIG. 8 (the number of cuts is 1).

The above description has shown that the present invention can provide the ability to effectively escape a local optimal solution.

Next, the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing one embodiment of the present invention. As shown in FIG. 1, the present invention comprises an adjacent element selecting means 1, a prefetch gain calculating means 2, a moving element selecting means 3, and an element moving means 4.

The adjacent element selecting means 1 obtains, for a given element, a set of elements adjacent to the given element.

The prefetch gain calculating means 2 calculates the gain of each element while using the adjacent element selecting means 1. That is, for each element i, the element adjacent to the element j is obtained by calling the adjacent element selecting means 1 and calling the adjacent element selecting means for each j. k is obtained, and the gain of element i is calculated based on equation (2) (each element j
Is the sum of the gains for element i).

The moving element selecting means 3 selects the element having the largest positive gain. If there is no such element, the division improvement processing ends.

The element moving means 4 inverts the position (left or right) of the selected element with respect to the selected element, activates the look-ahead gain calculating means 2 and calculates the gain of each element in the state after the movement. Recalculate.

Next, a second embodiment of the present invention will be described.

In a general circuit, a net connecting elements is not limited to a two-terminal net. That is, equation (2)
Cannot define the connection strength c (m, n) referred to in the above. Therefore, as a second embodiment of the present invention, a method for applying Equation (2) to a circuit having a multi-terminal net will be described.

FIG. 2 shows a block diagram of this embodiment.

In the second embodiment, before using the circuit division improving device shown in the first embodiment, a given circuit is subjected to "multi-terminal net conversion means 5" described below. A circuit consisting only of a two-terminal net is generated.

That is, each multi-terminal net existing in a given circuit is converted into a set of two-terminal nets.
In the case of an n-terminal net, this is broken down into n (n-1) / 2 possible element pairs, giving a 2 / n connection strength to each pair. For example, a net connected to three elements a, b, and c is decomposed into three two-terminal nets between a and b, between b and c, and between a and c, and the connection strength of each two-terminal net is 2
/ 3.

Next, a third embodiment of the present invention using a conventional gain calculation method will be described.

FIG. 3 is a block diagram showing a third embodiment.

The non-look-ahead gain calculating means 6 performs a gain calculation for each element by using the conventional method (Equation (1)) in consideration of only the element adjacent thereto.

The adjacent element selecting means 1 and the prefetch gain calculating means 2 are the same as in the first embodiment.

As described in the first embodiment, the moving element selecting means 3 selects the element having the largest positive gain, but among the elements calculated by the non-look-ahead gain calculating means 6, If there is no element, the prefetch gain calculating means 2 is started, and the gain of each element is recalculated based on the equation (2), and the element having the largest positive gain is selected.
If an element cannot be selected at this point, the division improvement processing is terminated.

The element moving means 4 inverts the position (left side or right side) of the element selected by the non-look-ahead gain calculating means 6 or the pre-ahead gain calculating means 2, and Recalculate the gain of each element in the state after starting and moving.

In the non-look-ahead gain calculating means 6, since each element is referred to only the element adjacent thereto, the processing time is somewhat shorter than in the pre-fetch gain calculating means 2. Further, when a local optimal solution as illustrated in FIG. 6 is encountered, the non-look-ahead gain calculating means 2 is activated and recovery is achieved, so that the optimality of the division result is improved as compared with the conventional method.

[0040]

As described above, according to the present invention,
The gain calculation in consideration of the influence of the element having no direct connection and the movement of the element in accordance with the gain calculation become possible. As illustrated with respect to the division in FIG. It can be performed without using the inefficient measures for escaping from the locally optimized solution that has been conventionally adopted.

[Brief description of the drawings]

FIG. 1 is a block diagram of a first embodiment of the present invention.

FIG. 2 is a block diagram of a second embodiment of the present invention.

FIG. 3 is a block diagram of a third embodiment of the present invention.

FIG. 4 is an explanatory diagram of a gain calculation method according to the present invention.

FIG. 5 is an explanatory view of a conventional method.

FIG. 6 is an explanatory diagram of a problem of the conventional method.

FIG. 7 is an explanatory diagram of a problem solving according to the present invention.

FIG. 8 is an explanatory diagram of the problem solving according to the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Neighboring element selection means 2 Look-ahead gain calculation means 3 Moving element selection means 4 Element movement means 5 Multi-terminal net conversion means 6 Non-look-ahead gain calculation means 10 Division boundary 11 Elements 12 Connection relation

Claims (3)

(57) [Claims] 1. An operation of sequentially moving an element included in the circuit from a partial circuit to which the element belongs to another partial circuit with respect to the two divided circuits,
In a circuit division improving apparatus for reducing an evaluation function value relating to a connection relationship between partial circuits, an adjacent element selection means for enumerating elements having a connection relationship with a given element, and for each of the enumerated elements On the other hand, a look-ahead gain calculating means for calculating a reduction amount (gain) of the evaluation function value when the element is moved to another partial circuit in consideration of an element having no direct connection to the element. And a moving element selecting means for selecting an element having the largest amount of reduction (gain) and a moving element for moving the selected element to another partial circuit. apparatus. 2. For each of the listed elements, the amount of decrease (gain) of the evaluation function value when the element is moved to another partial circuit is determined for only the elements directly connected to the element. Non-look-ahead gain calculating means for calculating in consideration of, the look-ahead gain calculating means only when the moving element selecting means cannot select an appropriate element based on the gain obtained by the non-look-ahead gain calculating means 2. The circuit division improving device according to claim 1, wherein the gain is recalculated and output to the moving element selecting means. 3. A method according to claim 2, wherein said look-ahead gain calculating means includes:
For each element that has been connected, the element, directly connected to the element
Connected with the element having the direct connection
Specially, it calculates the look-ahead gain based on the number of connected elements.
2. The circuit division improving device according to claim 1, wherein: