CN1588380A - Non-linear planning layout method based minimum degree of freedom priority principle - Google Patents

Abstract

The present invention is non-linear planning configuration method based on maximum freedom priority principle, and belongs to the field of computer-aided integrated circuit design technology. Specially, the present invention features that the advantages of maximum freedom priority principle and those one the non-linear planning configuration method are combined for simultaneously optimizing in area and line length. The non-linear planning configuration method is first utilized to optimize total line length; and the maximum freedom priority principle is then utilized to optimize the configuration based on the length/width ratio and area utilization rate the user has given. The present invention has fast speed, high area utilization and optimized total line length.

Description

Non-linear planning layout method based on minimum degree of freedom priority principle

Technical field

Non-linear planning layout method based on minimum degree of freedom priority principle belongs to the integrated circuit CAD field, relates in particular to BBL (Building Block Layout) field.

Background technology

Physical Design is a ring main in VLSI (very large scale integrated circuit) (VLSI) design process.Xiang Guan Computer-aided Design Technology is called automatic graph generation therewith.The manufacturing process of integrated circuit enters into sub-micro (VDSM) by present deep-submicron (DSM), the design scale of integrated circuit is also just by ultra-large (VLSI), very extensive (ULSI) develops to G extensive (GSI), and the complicacy of circuit sharply increases the unprecedented attention that makes the reuse technology of hierarchy type circuit design and circuit module be subjected to academic and industrial community.The emerge in multitude of ip module has also proposed unprecedented challenge for IC layout design; The hierarchy type layout-design; the module reuse technology; the extensive application of ip module; SOC (system on a chip) especially digital-to-analogue is mixed the design of SOC (system on a chip); and mimic channel device level Butut problem etc.; these problems can be summed up as the floor planning and the location problem of integrated circuit macroblock, i.e. the Butut problem of Building Block Layout:BBL pattern.

The task of location problem is that to want the exact position of determination module on chip, its target be to guarantee to make that chip area is as far as possible little under the logical prerequisite of cloth.Placement algorithm can be divided into two big classes, random optimization method and Deterministic Methods, and the random optimization method obtains the result by search in the solution space of definition, and speed is slower.Deterministic Methods be more given inspire strategy, go to instruct layout by these strategies, common have a greedy algorithm.

For a long time, people build a wall building the riverbank with the stone with polygon plane, during flooring, always use restricted maximum resource at first, for example at first do from the bight of perform region, at first big or long piece of usable floor area or the like.Principle that Here it is so-called " minimum degree of freedom priority ".What " Jin Jiao, silver-colored limit, the careless belly " in the go said is exactly this principle.Determinacy floor planning and placement algorithm based on this principle aspect the speed then are being the fastest present placement algorithm, and layout result is suitable with artificial layout result, but line length aspect result is poor slightly.

The method of nonlinear programming is commonly used to solve some optimization problems, but often needs problem is similar to, and the effect that is directly used in location problem is also bad.

The present invention has studied nonlinear programming problem and minimum degree of freedom priority principle, and their advantage is combined, and finishes location problem efficiently according to the area utilization that the user is given.

Summary of the invention

The objective of the invention is to be to propose the non-linear planning layout method based on minimum degree of freedom priority principle, it combines the advantage of minimum degree of freedom priority principle and nonlinear programming approach, makes location problem more effective.Minimum degree of freedom priority principle is based on the experience in people's real work life, its main thought is exactly to handle earlier to take the big module of resource, handle then and take the little module of resource, in concrete the placement, adopt the strategy of " golden horn silver AgX limit grass belly ", consider earlier module is placed on the angle, consider then on the limit, in the middle of being only at last.Like this, the order of placement is gradually to the centre from four contentions.The method of nonlinear programming is a kind of Mathematical Method, and it can calculate the placement schemes that approximate model has minimum line length.

Be provided with the location problem of n module: rectangle module collection B={B ₁, B ₂, B ₃..., B _nIn parallel being placed in the rectangular coordinate system of each module, the length of each module i and wide (h _i, ω _i) all provide h wherein _iAnd w _iBe respectively B _iLength and wide.Target is to find a kind of Butut scheme, makes under the prerequisite of all modules under the placement used area and line length minimum.

The invention is characterized in: it contains following steps successively:

(1). computer initialization

Read in the length and width of all modules and the information of line between them;

Given area utilization and the length breadth ratio of input user;

Set following parameters:

The degree of freedom of layout white space:

The degree of freedom in bight is defined as P _c ^f, P _c ^f=1/2,

Degree of freedom on the limit is defined as P _s ^f, P _s ^f=3/4,

Middle degree of freedom is defined as P _h ^f, P _h ^f=1;

The degree of freedom of module is defined as R _i ^f:

R _i ^f＝{r ₁*(1-B _i/A _a)+r ₂*(1-max(w _i，h _i)/min(W，H))}；

Wherein: r ₁+ r ₂=1,

w _i, h _iBe respectively the wide and high of module i, B _iBe the area of module i,

A _aBe the area of placement space, W, H are respectively the wide and high of it;

The interconnection degree of freedom is defined as C _i ^f, i.e. the interconnection degree of freedom of module i,

t _i ^jBe module i and adjacent block j line number,

ω _i ^jBe module i and adjacent block j wire length, estimate with the half cycle regular way;

Initial solution degree of freedom S _i ^f, $S_i^f={|x_{\mathrm{icur}}-x_{\mathrm{iori}}|}^2+{|y_{\mathrm{icur}}-y_{\mathrm{iori}}|}^2$

x _Icur, y _IcurBe respectively the coordinate of the current place of module grid,

x _Iori, y _IofiBe respectively the coordinate of place grid in the module initial solution;

(2). the use approximate model is asked the initial layout result with minimum line length and is write down final coordinate of each circle, and it contains following steps successively:

(2.1). all rectangular modules are approximated to the equal circle of area with it, and all lines all send from the center of circle, and are not overlapping between each circle;

(2.2). use the length of approximate every the line of secondary line regular way and make total line length objective function W (x, y) minimum:

$W(x,y)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\ω}}_{\mathrm{ij}}\·\frac{1}{2}\left[\right({(x_i-x_j)}^2+{(y_i-y_j)}^2]$

Constraint condition: g (x, y)=(r _i+ r _j) ²-[(x _i-x _j) ²+ (y _i-y _j) ²]≤0

Wherein: n is a number of modules, (x _i, y _i) be the coordinate in i the module center of circle,

ω _IjBe module i, the line number between j

r _i, r _jBe respectively module i, the radius of j

(2.3). use the method for penalty function that problem is changed into no constrained objective function P (x, y, c _k)

$P(x,y,c_k)=W(x,y)+c_k\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}\mathrm{Max}(0,g(x,y))$

C wherein ₀=1, c _K+1=10c _k, k is the iterations counter, from 0 → ∞

(2.4). fix one of them module, use progressively quadratic programming to obtain initial solution with global optimum's line length by the function f mincon among the software package matlab, promptly convergent subsequence (x, y) ^kThe limit;

(2.5). the layout areas of initial solution is divided into m*m grid, and the grid at each place, the center of circle is write down in m=4～12;

(3). according to user given area utilization and length breadth ratio, calculate the length and width of layout areas, in conjunction with initial solution, use minimum degree of freedom priority principle to carry out layout, it contains successively and has the following steps:

(3.1). establish: the module collection of UBS for not placing,

PBS is the module collection that places,

PL is current placement situation;

(3.2). the module degree of freedom R of all modules by them _i ^fR is found in ordering _i ^fMinimum module attempts earlier placing this module, since 4 angles, has 8 kinds of placement schemes, for each placement schemes, after placing first module, adopts known greedy algorithm to place for the residue module;

(3.3). the described 8 kinds of schemes of calculation procedure (3.2) are calculated as follows total degree of freedom F separately _k, totally 8 values:

$F_k=\mathrm{\αP}+\mathrm{\β}{R}_{k^f}+\mathrm{\γ}{C_k}^F+\mathrm{\λ}{S_k}^f$

P is a layout white space degree of freedom

(3.4). choose wherein F _kMinimum placement schemes is as the placement schemes of the module of above-mentioned first placement;

(3.5). above-mentioned first module that has placed is removed from UBS, put into PBS;

(3.6). repeating step (3.2)～(3.6) are till the UBS sky;

(4). use the method for known simulated annealing to adjust layout, optimize total line length.

(5). with figure and two kinds of form output of file final layout result.

Non-linear planning layout algorithm based on minimum degree of freedom priority principle of the present invention has following advantage:

[1] uses minimum degree of freedom priority principle to finish layout, absorb the feature of the initial solution that the Mathematical Method of nonlinear programming obtains simultaneously, optimize area and line length simultaneously for given layout areas;

[2] algorithm has fast speeds, higher area utilization;

[3] have industrial application value, can be used for the integrated circuit (IC) design process: the location problem in module level floor planning/layout.

Experimental result is as follows:

The explanation of table one MCNC standard test circuit parameter

Circuit name	Number of modules	Line screen	The module total area (mm2)
				????Apte	????9	????97	46.561628
????Xerox	????10	????203	19.350296
				????Hp	????11	????83	8.830584
????Ami33	????33	????123	1.156449
				????Ami49	????49	????408	35.445424
????Playout.xlii	????62	????2056	88.177016

Table two experimental result

	??Apte(mm)	?Xerox(mm)	??Hp(mm)	??Ami33	??Ami49	??Playout
							Area utilization (%)	??96.56	?93.47	??92.22	??93.91	??94.96	??96.69
Total line length (mm)	??102.83	?521.30	??118.05	??45.30	??1021.23	??6523.65
							Time (s)	??0.08	?0.40	??0.10	??6.42	??53.07	??63.43

The hardware that the present invention uses is one and has disposed the CPU of P4 1.8G and the PC of 256 MB of memory; Use RedhatLinux7.2 operating system.

Description of drawings

Fig. 1. the program structure flow chartr of planing method of the present invention.

Fig. 2. use nonlinear programming approach to obtain the initial solution of example playout.

Fig. 3. 8 position candidate during first module placement.

Fig. 4. the initial solution degree of freedom of computing module.

Fig. 5. the position candidate when placing second module.

Fig. 6. the final layout result of example playout.

Embodiment

(1). read in all modules and link information between them, carry out initial work, memory module length and width and the information of line between them

(2). the use approximate model is asked the initial layout result with minimum line length and is write down final coordinate of each circle: all rectangular modules are approximated to the equal circle of area with it, think that all lines all send from the center of circle, can not be overlapping between the circle, use the length of approximate every the line of secondary line regular way, target is to make total wire length minimum.Formalized description is as follows:

$W(x,y)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\ω}}_{\mathrm{ij}}\·\frac{1}{2}\left[\right({(x_i-x_j)}^2+{(y_i-y_j)}^2]$

Constraint condition:

g(x，y)＝(r _i+r _j) ²-[(x _i-x _j) ²+(y _i-y _j) ²]≤0

Wherein:

W is total line length objective function

N is a number of modules

(x _i, y _i) be the coordinate in i the module center of circle

W _IjBe module i, the line number between j

r _iRadius for module i

Use the method (Penalty Function Method) of penalty function that problem is changed into no constrained objective function

$P(x,y,c_k)=W(x,y)+c_k\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}\mathrm{Max}(0,g(x,y))$

C wherein ₀=1, c _K+1=10c _k

In order to make P (x, y, c _k) obtain the overall situation and separate most, need fix some modules, establishing wherein has 1 module to fix.At this moment for each c _k, P (x, y, c _k) the matrix of coefficients of first order derivative be diagonal dominance battle array (diagonally dominant matrix), also be positive definite matrix (positively definite matrix), i.e. P (x, y, c _k) at each c _kThe place all exists and separates most, then just can obtain convergent subsequence (x, y) ^k, limit of a sequence is exactly separating most of problem.Concrete solution is many, as Newton method (Newton Method), and quasi-Newton method (Quasi-Newton Method), method of conjugate gradient (FR), progressively quadratic programming (SQP) uses progressively quadratic programming (SQP) in this method.

The layout areas of initial solution is divided into the m*m grid, writes down the grid at each place, the center of circle.

(3). in conjunction with initial solution, use minimum degree of freedom priority principle to carry out layout: minimum degree of freedom priority (LFF) principle is based on the experience of people's daily life and production, think and in layout, should handle the most difficult placement earlier, just take the maximum module of resource, the back is placed and is taken the few module of resource.Algorithm has further developed this thought, and degree of freedom is quantized.Be the definition and the quantitative formula of each degree of freedom below:

Layout white space degree of freedom: the degree of freedom of the white space of placement.Be placed on the angle, can only move to both direction; Be placed on the limit, then have three directions to move; In the middle of only being placed on, just can move up and down.Therefore, the degree of freedom in definition bight is minimum, is P _c ^f=2/4=1/2; Secondly, be P on the limit _s ^f=3/4; The centre then is P _h ^f=4/4=1

The module degree of freedom: size and shape by module are determined, suppose B _iBe the area of module i, w _iBe the wide of it, h _iBe its height, A _aBe the area of placement space, W is the wide of it, and H is its height.Then the degree of freedom of this module can be defined as:

R _i ^f＝{r ₁*(1-B _i/A _a)+r ₂*(1-max(w _i，h _i)/min(W，H))}；

Wherein: r ₁+ r ₂=1.

Interconnection degree of freedom: be used for reflecting the line relation between the module.Suppose module B _iBe the module that will place, the interconnection degree of freedom be exactly its on the position that will place with the line of adjacent block and, the degree of freedom between this module just can be defined as so:

ω wherein _IjBe module i and adjacent block j wire length, estimate t with the half cycle regular way _IjBe module i and adjacent block j line number

The initial solution degree of freedom: it is poor to be used for the relative position of current placement of representation module and initial solution:

S _i ^f＝|x _icur-x _iori| ²+|y _icur-y _iori| ²

Wherein, x _Icur, y _IcurBe the coordinate of the current place of module grid, x _Iori, y _IoriIt is the coordinate of place grid in the module initial solution

Be detailed flow process below:

UBS: the module collection of Fang Zhiing not

PBS: the module collection that has placed

PL: current placement situation

(3.1). modules all among the UBS is attempted putting into successively each angle of layout areas, and whether detection has with current placement and conflicts, if not conflict (just do not overlap with existing the placement, also do not exceed the border of layout areas) just counts in the legal placement schemes

(3.2). the pseudo-placement down worked as front module, is labeled as dummy node, for remaining module, places them with greedy algorithm, till can't putting down new module.

(3.3). calculate current degree of freedom:

$F_k=\mathrm{\αP}+\mathrm{\β}{R}_{k^f}+\mathrm{\γ}{C_k}^f+\mathrm{\λ}{S_k}^f$

Wherein: α, beta, gamma, λ are the weight coefficient of each degree of freedom

(3.4). upgrade PL, among the placement schemes adding PBS that has minimum degree of freedom, among UBS, remove this module.

(3.5). repeat (3.1)～(3.4) till the UBS sky.

(4). use the method for simulated annealing to adjust final layout: the upset that module that will cloth is good is not stopped in level and vertical direction, improve the line length of layout.

(5). two kinds of form output of figure and file result.

Location problem is exactly to finish the placement problem of module according to user given area length breadth ratio and area utilization, and satisfy other constraint, layout wire length etc. for example, the FB(flow block) of this layout method such as Fig. 1, below we illustrate method of the present invention with the put procedure of MCNC standard instance playout in conjunction with Fig. 1.It has the following steps successively:

Concrete implementation step is as follows:

(1) reads input file, the length and width of logging modle and the link information of intermodule.Program at first is the circuit description document of reading in example playout: playout.yal, extract module wherein size and between the line relation information and store in the related data structure

(2) model according to step (2) uses the method for nonlinear programming to obtain having the initial solution of optimum line length, each module is approximated to and the equal-sized circle of its area, the multiterminal gauze of intermodule is splitted into 2 end line nets, with the length of every line of secondary line regular way estimation, point-to-point transmission wire length secondary line regular way estimation formulas is as follows:

$W=\sqrt{{(x_i-x_j)}^2+{(y_i-y_j)}^2}---\left(8\right)$

(x wherein _i, y _i) (x _j, y _j) be the coordinate of two end points of gauze

Find the solution the initial solution problem with minimum overall line length and just change into a nonlinear programming problem, target is institute's wired network length summation, sees formula (1), and constraint is that any two circles are not overlapping, sees formula (2).Convert it into and do not have the constraint nonlinear programming problem, see formula (3), using progressively, quadratic programming solves the initial solution with optimum overall line length, this step can use mathematical tool matlab to realize, function f mincon wherein can use progressively that quadratic programming solves the initial solution with global optimum's line length.The coordinate of the initial solution directly module coordinate in the scheme of final layout is associated, so must connect the contact of setting up between them with a kind of relative coordinate, layout areas is divided into m*m grid, the mesh coordinate at each place, the module center of circle is noted, and the suggestion span that obtains m from experimental data is 4～12.At this example of playout initial solution is divided into the zone of 4*4, writes down the zone at the place, the center of circle of each module, as shown in Figure 2, wherein the initial solution coordinate of No. 29 modules in the upper left corner is (1,1)

(3) according to user given area utilization and length breadth ratio, calculate the length and width of layout areas, for example the user wishes to obtain 95.7% area utilization and the length breadth ratio of layout areas is 1.5: 1, and the area of all modules is 88.177016mm among the example playout ², then the area of layout areas should be 88.177016/0.957=92.151366, and the length of all layouts should be: 11.757mm, wide should be: 7.838mm.

Then with the module degree of freedom ordering of module according to them, can calculate the module degree of freedom minimum of module 29 according to formula (4), attempt earlier placing it, because the spatial degrees of freedom minimum at angle, module 29 is attempted since 4 angles, consider that module can horizontal and vertically place, it has 8 kinds of placement schemes, sees Fig. 3, for each placement schemes, after placing first module, all adopt greedy algorithm to place for remaining module.

Use formula (5), (6) calculate the interconnection degree of freedom and the initial solution degree of freedom of current placement schemes.For first module that will place, wherein for there not being module, all interconnection degree of freedom are 0, for the module of back, use half cycle regular way (frame residence half of minimum rectangle girth) to calculate the module around front module.Wherein the computation process at Fig. 4 initial solution degree of freedom is as follows: module 29 initial place grids are (1,1), and current place grid is (2,4), and according to formula (6), its initial solution degree of freedom is: | (2-1) | ²+ | (4-1) | ²=10.

Use formula (7) to calculate total degree of freedom F at last _i, for module 29 8 F are just arranged like this _iValue, example playout one has 62 modules, a total 62*8 candidate scheme and a F _iValue is chosen wherein F _iMinimum placement schemes.

Module 29 can occupy angle 1 after placing and finishing, and bears 6 two the new angles in angle 5 and angle simultaneously.New angle and the angle that is not occupied all are the position candidate of remaining module, suppose that module 55 is the modules that next will place, then it has ten candidate schemes, see Fig. 5, with module deletion from the placement module tabulation of having put well, place all modules that finish according to the sequential loop of (3.1)～(3.4).

(4) will place the module that finishes and use the method for simulated annealing to carry out level and vertical upset, if after finding certain module level or flip vertical, total line length descends, then with this module upset.The process that the repetition that does not stop is such is till total line length of layout no longer descends.The total line length that can further optimize distribution like this.

(5) two kinds of form output of graphic file final layout result.

Claims (1)

1. based on the non-linear planning layout method of minimum degree of freedom priority principle, it is characterized in that it contains following steps successively:

(1). computer initialization

Read in the length and width of all modules and the information of line between them;

Given area utilization and the length breadth ratio of input user;

Set following parameters:

The degree of freedom of layout white space:

The degree of freedom in bight is defined as P _c ^f, P _c ^f=1/2

Degree of freedom on the limit is defined as P _s ^f, P _s ^f=3/4

Middle degree of freedom is defined as P _h ^f, P _h ^f=1

The degree of freedom of module is defined as R _i ^f:

R _i ^f＝{r ₁*(1-B _i/A _a)+r ₂*(1-max(w _i，h _i)/min(W，H))}；

Wherein: r ₁+ r ₂=1,

w _i, h _iBe respectively the wide and high of module i, B _iBe the area of module i,

A _aBe the area of placement space, W, H are respectively the wide and high of it;

The interconnection degree of freedom is defined as C _i ^f, i.e. the interconnection degree of freedom of module i,

t _IjBe module i and adjacent block j line number,

ω _IjBe module i and adjacent block j wire length, estimate with the half cycle regular way;

Initial solution degree of freedom S _i ^f, $S_i^f={|x_{\mathrm{icur}}-x_{\mathrm{iori}}|}^2+{|y_{\mathrm{icur}}-y_{\mathrm{iori}}|}^2$

x _Icur, y _IcurBe respectively the coordinate of the current place of module grid,

x _Iori, y _IoriBe respectively the coordinate of place grid in the module initial solution;

(2). the use approximate model is asked the initial layout result with minimum line length and is write down final coordinate of each circle, and it contains following steps successively:

(2.1). all rectangular modules are approximated to the equal circle of area with it, and all lines all send from the center of circle, and are not overlapping between each circle;

(2.2). use the length of approximate every the line of secondary line regular way and make total line length objective function W (x, y) minimum:

$W(x,y)=\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}{\mathrm{\ω}}_{\mathrm{ij}}\·\frac{1}{2}\left[({(x_i-x_j)}^2+{(y_i-y_j)}^2)\right]$

Constraint condition: g (x, y)=(r _i+ r _j) ²-[(x _i-x _j) ²+ (y _i-y _j) ²]≤0

Wherein: n is a number of modules, (x _i, y _i) be the coordinate in i the module center of circle,

ω _{I, j}Be module i, the line number between j

r _i, r _jBe respectively module i, the radius of j

(2.3). use the method for penalty function that problem is changed into no constrained objective function P (x, y, c _k)

$P(x,y,c_k)=W(x,y)+c_k\underset{i=1}{\overset{n}{\mathrm{\Σ}}}\underset{j=1}{\overset{n}{\mathrm{\Σ}}}\mathrm{Max}(0,g(x,y))$

C wherein ₀=1, c _K+1=10c _k, k is the iterations counter, from 0 → ∞

(2.4). fix one of them module, use progressively quadratic programming to obtain initial solution with global optimum's line length by the function f mincon among the software package matlab, promptly convergent subsequence (x, y) ^kThe limit;

(2.5). the layout areas of initial solution is divided into m*m grid, and the grid at each place, the center of circle is write down in m=4 ~ 12;

(3). according to user given area utilization and length breadth ratio, calculate the length and width of layout areas, in conjunction with initial solution, use minimum degree of freedom priority principle to carry out layout, it contains successively and has the following steps:

(3.1). establish: the module collection of UBS for not placing,

PBS is the module collection that places,

PL is current placement situation;

(3.2). the module degree of freedom R of all modules by them _i ^fR is found in ordering _i ^fMinimum module attempts earlier placing this module, since 4 angles, has 8 kinds of placement schemes, for each placement schemes, after placing first module, adopts known greedy algorithm to place for the residue module;

(3.3). the described 8 kinds of schemes of calculation procedure (3.2) are calculated as follows total degree of freedom F separately _k, totally 8 values:

F _k＝αP+βR _kf+γC _k ^f+λS _k ^f

P is a layout white space degree of freedom

(3.4). choose wherein F _kMinimum placement schemes is as the placement schemes of the module of above-mentioned first placement;

(3.5). above-mentioned first module that has placed is removed from UBS, put into PBS;

(3.6). repeating step (3.2)～(3.6) are till the UBS sky;

(4). use the method for known simulated annealing to adjust layout, optimize total line length.

(5). with figure and two kinds of form output of file final layout result.

Images