CN102490120B - Adaptive inverse control system for chemical-mechanical polishing machine - Google Patents
Adaptive inverse control system for chemical-mechanical polishing machine Download PDFInfo
- Publication number
- CN102490120B CN102490120B CN 201110388573 CN201110388573A CN102490120B CN 102490120 B CN102490120 B CN 102490120B CN 201110388573 CN201110388573 CN 201110388573 CN 201110388573 A CN201110388573 A CN 201110388573A CN 102490120 B CN102490120 B CN 102490120B
- Authority
- CN
- China
- Prior art keywords
- head chamber
- rubbing head
- control
- amount
- pressure amount
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Images
Abstract
The invention discloses an adaptive inverse control system for a chemical-mechanical polishing machine, which comprises a target pressure input device used for inputting target pressure of a polishing head chamber to a compound control device, a neural network identification device used for identifying a control model of a system based on the input pressure of the polishing head chamber input by the compound control device and the pressure output by the polishing head chamber, and the compound control device used for outputting the input pressure of the polishing head chamber based on the target pressure, the pressure output by the polishing head chamber and the control model. Each compound control unit comprises a neural network controller used for calculating neural network controlled quantity, a proportional integral controller used for generating proportional integral controlled quantity and a weighting device used for generating the input pressure of the polishing head chamber for the neural network controlled quantity and the proportional integral controlled quantity with the subsection variable-parameter control strategy. The adaptive inverse control system for the chemical-mechanical polishing machine is capable of reducing coupling among areas of the polishing head chamber through dynamic on-line decoupling of neurons.
Description
Technical field
The present invention relates to the chemical-mechanical polishing mathing equipment technical field, particularly a kind of Adaptive inverse control system for chemical-mechanical polishing mathing.
Background technology
In IC (Integrated Circuit, integrated circuit) manufacturing technology, along with improving constantly of properties of product, more and more higher to the requirement of surface quality.Silicon chip is as the basic material of IC chip, and its surface roughness and surface smoothness become one of key factor that influences integrated circuit etching live width.Polishing is the important means of surperficial planarization process.CMP (Chemical mechanical polishing, chemical-mechanical polishing mathing) technology the most extensively adopts the leveling technology, occupies important position in the IC manufacturing technology.CMP is the combination technique of mechanical skiving and chemical attack, throws in copper/throwing silicon field at chemical-mechanical polishing mathing, and rubbing head clamping silicon chip will wait to throw the polishing disk that copper layer/silicon face is pressed to rotation.Realize that by the polishing pad friction on the polishing disk and polishing fluid corrosion copper/silicon removes fast and effectively.Wherein, rubbing head can be by the overall dynamic adjustments of each annulus chamber pressure realization of the control silicon chip back side to silicon wafer polishing pressure.
The rubbing head internal chamber is equivalent to a plurality of airtight chamber, and wherein, a plurality of airtight chamber can expand and shrink.Can push or not have extruding mutually between the chamber.When chemical-mechanical polishing mathing polishes, intercouple between each chamber.Wherein, the coupling between the chamber comprises volume coupling and two kinds of situations of source of the gas input coupling.
(1) volume coupling: the extruding between each chamber or contraction cause the variation of cavity volume.
(2) source of the gas input coupling: cause source of the gas moment air supply pressure fluctuation when each chamber pressurizes simultaneously.
The coupling of rubbing head chamber will cause the pressure upheaval of chemically mechanical polishing CMP system, and then reduces the quality of finish of wafer.
Summary of the invention
Purpose of the present invention is intended to solve at least above-mentioned technological deficiency, proposes a kind of Adaptive inverse control system for chemical-mechanical polishing mathing especially, and this system can reduce the respectively coupling between the district of rubbing head chamber by neuronic dynamic online decoupling zero.
For achieving the above object, embodiments of the invention provide a kind of Adaptive inverse control system for chemical-mechanical polishing mathing, comprise: goal pressure amount input unit, composite control apparatus, neural network identification device, wherein, described goal pressure amount input unit is used for to the goal pressure amount of the rubbing head chamber of described composite control apparatus input chemical-mechanical polishing mathing; Described neural network identification device is used for the control model according to the described Adaptive inverse control system for chemical-mechanical polishing mathing of the input pressure amount of rubbing head chamber and the identification of rubbing head chamber output pressure amount; Be used for according to described goal pressure amount with described composite control apparatus, described rubbing head chamber output pressure amount and described control model calculate the input pressure amount of described rubbing head chamber, wherein, described composite control apparatus comprises at least one group of compound control module, wherein, every group of described compound control module is corresponding continuous with each district of the rubbing head chamber of described chemical-mechanical polishing mathing, every group of described compound control module comprises: nerve network controller, described nerve network controller receive from the described goal pressure amount of described goal pressure amount input unit and described control model and generate the ANN Control amount; Pi controller, described proportional plus integral control generates the proportional plus integral control amount according to the amount of pressure of described goal pressure amount, the output of described rubbing head chamber; And weighter, described weighter links to each other with described pi controller with described nerve network controller respectively, be used for described ANN Control amount and described proportional plus integral control amount are generated the input pressure amount of described rubbing head chamber with segmentation VARIABLE PARAMETER PID CONTROL strategy, wherein
u＝ku
n+u
p，
Wherein, u is input pressure amount, the u of described rubbing head chamber
nBe described ANN Control amount, u
nBe described proportional plus integral control amount, k is the segmentation variable element.
The Adaptive inverse control system that is used for machine glazed finish according to the embodiment of the invention by neuronic online real-time dynamic decoupling characteristic, can obtain the inversion model of object, and the coupling between each district of elimination rubbing head chamber.In addition, adopt segmentation VARIABLE PARAMETER PID CONTROL strategy to reduce because the adverse effect and the system's upheaval that bring forbidden in the contrary control of initial time Model Distinguish.
In one embodiment of the invention, described segmentation variable element k is:
Wherein, a
0Initial value, t for k
0For time end point, the Δ k of initial time section is that the variable quantity of described segmentation variable element k, variable quantity, the n+2 that Δ t is adjacent two periods are the number of fragments of described segmentation variable element k.
In one embodiment of the invention, the quantity of described compound control module equates with the quantity of the rubbing head chamber of described chemical-mechanical polishing mathing.
In one embodiment of the invention, when the quantity of described rubbing head chamber is two or more, described Adaptive inverse control system for chemical-mechanical polishing mathing also comprises: neuron decoupling compensation device, described neuron decoupling compensation device links to each other with the rubbing head chamber of described composite control apparatus and described chemical-mechanical polishing mathing respectively, be used for the input pressure amount of the described rubbing head chamber of described composite control apparatus output is revised and generated the input pressure amount of revised rubbing head chamber, and the rubbing head chamber that the input pressure amount of described revised rubbing head chamber is exported to described chemical-mechanical polishing mathing.
In one embodiment of the invention, described neuron decoupling compensation device comprises a plurality of neuron decoupling compensations unit, the input of each described neuron decoupling compensation unit links to each other with the output of the weighter of a plurality of described compound control modules, the output of each described neuron decoupling compensation unit links to each other with a district of described rubbing head chamber, be used for the input pressure amount of described rubbing head chamber is carried out the input pressure amount that decoupling zero adjustment generates one tunnel correction back rubbing head chamber, and the described input pressure amount of revising back rubbing head chamber exported to a district of the rubbing head chamber of described chemical-mechanical polishing mathing.
In one embodiment of the invention, each described neuron decoupling compensation unit is weighted calculating to generate the input pressure amount of one tunnel revised rubbing head chamber to a plurality of input pressure amounts of the rubbing head chamber of described chemical-mechanical polishing mathing with different neuron weights.
In one embodiment of the invention, described neuron decoupling compensation unit equates with the quantity of described compound control module.
The aspect that the present invention adds and advantage part in the following description provide, and part will become obviously from the following description, or recognize by practice of the present invention.
Description of drawings
Above-mentioned and/or the additional aspect of the present invention and advantage are from obviously and easily understanding becoming the description of embodiment below in conjunction with accompanying drawing, wherein:
Fig. 1 is the internal structure schematic diagram of rubbing head;
The pressure response curve in each district when Fig. 2 is two districts pressurization 0.5psi;
The pressure response curve in each district when Fig. 3 is two districts pressurization 1.5psi;
Fig. 4 is DRNN neural network structure figure;
Fig. 5 is the schematic diagram that is used for the Adaptive inverse control system of chemical-mechanical polishing mathing according to an embodiment of the invention;
Fig. 6 is the schematic diagram that is used for the Adaptive inverse control system of chemical-mechanical polishing mathing in accordance with another embodiment of the present invention;
Fig. 7 be compound control module structured flowchart;
Fig. 8 is rubbing head multi-region pressure system closed loop decoupling zero control chart;
Fig. 9 (a) is not for adopting segmentation VARIABLE PARAMETER PID CONTROL policy response curve map;
Fig. 9 (b) is for adopting segmentation VARIABLE PARAMETER PID CONTROL policy response curve map;
Figure 10 (a) is that two districts join PID and Adaptive inverse control pressure-responsive comparison diagram when applying 0.5psi surely;
Figure 10 (b) is that two districts join PID and Adaptive inverse control pressure-responsive comparison diagram when applying 1psi surely;
Figure 10 (c) is that two districts join PID and Adaptive inverse control pressure-responsive comparison diagram when applying 1.5psi surely;
Figure 11 (a) is for deciding parameter PID multi-region pressure response curve; And
Figure 11 (b) is neural network decoupling Adaptive inverse control multi-region pressure response curve.
The specific embodiment
Describe embodiments of the invention below in detail, the example of described embodiment is shown in the drawings, and wherein identical or similar label is represented identical or similar elements or the element with identical or similar functions from start to finish.Be exemplary below by the embodiment that is described with reference to the drawings, only be used for explaining the present invention, and can not be interpreted as limitation of the present invention.
With reference to following description and accompanying drawing, these and other aspects of embodiments of the invention will be known.These describe and accompanying drawing in, some specific implementations in the embodiments of the invention are specifically disclosed, represent to implement some modes of the principle of embodiments of the invention, still should be appreciated that the scope of embodiments of the invention is not limited.On the contrary, embodiments of the invention comprise spirit and interior all changes, modification and the equivalent of intension scope that falls into institute's additional claims.
Fig. 1 shows the internal structure of rubbing head chamber.The rubbing head chamber can be divided into a plurality of districts to form a plurality of chambers.Between the chamber in a plurality of districts and with silicon chip 11 contact portions be flexible sheet 12.Arranged outside at chamber has retaining ring 13.As shown in Figure 1, the rubbing head chamber comprises 4 districts, and Pr is the rubbing head chamber pressure, and P1 is rubbing head chamber central area (district) pressure, and P2 is ring central area (two districts) pressure, and P3 is rubbing head outer rim (three districts) pressure.When the chamber the blowing pressure rose, cavity volume will correspondingly increase, simultaneously will mutual extrusion between the chamber, and chamber pressure changes, and namely has coupling phenomenon between the chamber.When increasing along with pressure differential between the chamber, degree of coupling will be more serious also.Fig. 2 and Fig. 3 are the pressure history that two districts respectively distinguish chamber when pressurizeing 0.5psi and 1.5psi (not pressurizeing in He San district, a district) respectively.As can be seen from Figures 2 and 3, He San district, a district is pushed in volumetric expansion rapidly when pressurize in two districts, He San district, a district is because thereby extruding makes volume reducing pressure rise, yet because the effect of electric Proportion valve, He San district, Dang Yi district pressure was greater than 0 o'clock, through after a while, its pressure will finally be adjusted to 0.Along with pressure reduction between the chamber increases, degree of coupling strengthens.
The Adaptive inverse control system that is used for chemical-mechanical polishing mathing that the embodiment of the invention provides can carry out decoupling zero to the coupling between the chamber.Because CMP multi-region control pressurer system is multi-input multi-output system, can utilize the height of neutral net non-linear, high learning ability and the mapping ability of nonlinear system carried out decoupling zero to the coupling between the chamber.When training and learn, neuroid itself does not also know that it is for decoupling zero or control, just according to the requirement of object function finish control object in interior system from being input to the mapping of output.Therefore as long as comprise the requirement of decoupling zero control in the training sample, neuroid will progressively be adjusted the connection weights according to learning algorithm, makes system realize decoupling zero control.
DRNN (Diagonal Recurrent Neural Network, Diagonal Recurrent Neural Network) neutral net is the neutral net with feedback, the dynamic characteristic of the reflection system that this network is can be more direct more lively.The DRNN neutral net is on the basis of BP network basic structure, makes its function that possesses the mapping behavioral characteristics by the storage internal state, thereby makes system have the ability that adapts to time-varying characteristics.
Based on the generalized ensemble positive model identifier on-line identification generalized object of DRNN, give feedforward controller with reciprocal directly tax of the model that recognizes, constituted a pseudo-unit link thereby connect with generalized object, make that output quantity can the real-time tracking input quantity.Closed loop feedback PI (Proportional Integral, proportional integral) controller can improve the system pressure fluctuation problem that some uncertain factors that the contrary control system of open loop is difficult to overcome cause, and has also remedied initial time neural network identification model inaccuracy on the other hand and has caused the defective that contrary control is forbidden.
As shown in Figure 4, the DRNN neutral net has three layers, and hidden layer is for returning layer.Wherein, the algorithm of DRNN neutral net is:
DRNN input layer: I
1(k), I
2(k), I
3(k);
DRNN returns layer input:
DRNN returns layer output: X
j(k)=f (S
j(k));
The DRNN output layer:
In following formula, I
1(k), I
2(k) and I
3(k) be respectively neuronic input quantity, X
j(k) for returning j neuronic output quantity of layer, S
j(k) be j recurrent nerve unit input summation, f (*) is the S function, y
m(k) and O (k) be respectively the output of DRNN network, W
DAnd W
OBe the weight vector of network recurrence layer and output layer, W
IBe the weight vector of network input layer, j=1,2 ..., 7, k is the iteration step of network.
Utilize non-linear ability and the neuronic online real-time dynamic decoupling characteristic of DRNN neutral net, can eliminate the coupling between the chamber.
As shown in Figure 5, the Adaptive inverse control system 100 that is used for chemical-mechanical polishing mathing CMP based on the DRNN network of the embodiment of the invention comprises goal pressure input unit 110, composite control apparatus 120 and neural network identification device 130.Wherein, goal pressure input unit 110 is used for to the goal pressure amount of composite control apparatus 120 input rubbing head chambers.Neural network identification device 130 is according to the control model of the Adaptive inverse control system 100 that is used for chemical-mechanical polishing mathing of the input pressure amount of the rubbing head chamber of composite control apparatus 120 output and the rubbing head chamber output pressure amount identification embodiment of the invention.Composite control apparatus 120 is used for according to the amount of pressure of goal pressure amount, the output of rubbing head chamber and the control model that 130 identifications of neural network identification device obtain and the input pressure amount of exporting the rubbing head chamber.Wherein, composite control apparatus 120 comprises at least one group of compound control module 121, and wherein, every group of compound control module 121 is corresponding continuous with each district of rubbing head chamber.
In one embodiment of the invention, neural network identification device 130 be the positive model identifier (Neural Network Identification, NNI)
As shown in Figure 7, every group of compound control module 121 comprise nerve network controller (Neural Network Control, NNC) 1211, pi controller 1212 and weighter 1213.Wherein, nerve network controller 1211 receives the goal pressure amount r from goal pressure amount input unit 110, and calculates ANN Control amount u according to goal pressure amount r with by the control model that 130 identifications of neural network identification device obtain
nPi controller 1212 generates proportional plus integral control amount u according to the amount of pressure y of goal pressure amount r, the output of rubbing head chamber
pWeighter 1213 links to each other with pi controller 1212 with nerve network controller 1211 respectively, with ANN Control amount u
nWith proportional plus integral control amount u
pAdjust according to default segmentation variable element strategy, generate the input pressure amount u of rubbing head chamber.Wherein, u=ku
n+ u
p
But because initial time inversion model and inaccuracy, thereby can cause ANN Control amount u
nAmplitude of fluctuation excessive, thereby the pressure oscillation of initiating system.Even regulating proportionality coefficient and the integral coefficient of pi controller 1212 all can't compensate owing to the inaccurate pressure oscillation that brings of inversion model identification.At the problems referred to above, the present invention proposes and based on response time of proportioning valve the initial press process is adopted segmentation VARIABLE PARAMETER PID CONTROL strategy.The main thought of segmentation VARIABLE PARAMETER PID CONTROL strategy is that dynamic control initially adds nerve network controller 1211 contrary control output controlled quentity controlled variable u
nAccurate inadequately, reduced u among the input pressure amount u of rubbing head chamber
nProportion.Thereby can be As time goes on, contrary control Model Distinguish is progressively accurate, and u among the u
nProportion also progressively amplify.
In one embodiment of the invention, u among the u
nProportion k is piecewise function:
Wherein, a
0Be [0, t
0) constantly initial value, t
0Be the time end point of initial time section, unit is second, and Δ k is that the variable quantity of segmentation variable element k, converted quantity, the n+2 that Δ t is adjacent two periods are the number of fragments of segmentation variable element k.Wherein, n=1,2,3 ....
In an example of the present invention since the response time of the electric Proportion valve of rubbing head chamber be 0.1 second, so Δ t be generally 0.1 second the integer time doubly, namely Δ t=0.1*n (n=1,2,3 ... .).The value of segmentation variable element k is as follows:
Need to prove that the value of above-mentioned segmentation variable element k only for exemplary purposes.The number of fragments of segmentation variable element k and every section value all can arrange separately.By adopting segmentation VARIABLE PARAMETER PID CONTROL strategy to reduce because the adverse effect and the system's upheaval that bring forbidden in the contrary control of initial time Model Distinguish, thereby make The whole control system tend towards stability.
After contrary control Model Distinguish is progressively accurate, nerve network controller 1211 will weaken the effect of pi controller 1212, make the output controlled quentity controlled variable mainly by u
nBear, and then the input of the output real-time tracking of the system of realization, realize the tracking control of system.
In one embodiment of the invention, the quantity of compound control module 121 equates with the number of partitions of rubbing head chamber.Particularly, when the rubbing head chamber had only a district, compound control module 121 was one, and there is not coupling phenomenon in chamber.When the rubbing head chamber comprises two or more subregions, compound control module 121 is two or more of equal number.At this moment, can there be coupling between the chamber of a plurality of subregions.
When the quantity of rubbing head chamber was two or more, as shown in Figure 6, the Adaptive inverse control system that is used for chemical-mechanical polishing mathing of the embodiment of the invention also comprised neuron decoupling compensation device 140.Neuron decoupling compensation device 140 links to each other with rubbing head chamber 1 with composite control apparatus 120 respectively, input pressure amount for the rubbing head chamber that composite control apparatus 120 is exported is revised, generate the input pressure amount of revised rubbing head chamber, and the input pressure amount of revised rubbing head chamber is outputed to rubbing head chamber 1.
Neuron decoupling compensation device 140 comprises a plurality of neuron decoupling compensations unit 141, wherein, the input of each neuron decoupling compensation unit 141 links to each other with the output of the weighter 1213 of a plurality of compound control modules 121, and one of the output of each neuron decoupling compensation unit 141 and rubbing head chamber 1 links to each other.Each neuron decoupling compensation unit 141 can carry out the decoupling zero adjustment to the input pressure amount u by the rubbing head chamber of compound control module 121 outputs, generates the input pressure amount u* of one tunnel revised rubbing head chamber.Then, each neuron decoupling compensation unit 141 can output to the input pressure amount of revising back rubbing head chamber a district of rubbing head chamber.In one embodiment of the invention, neuron decoupling compensation unit 141 links to each other with the quantity of compound control module 121, thereby can revise accordingly the amount of pressure of each compound control module 121 output.
The 141 pairs of a plurality of input pressure amounts by the rubbing head chamber of a plurality of compound control module 141 outputs in each neuron decoupling compensation unit are weighted with different neuron weights, thereby generate the input pressure amount of one tunnel revised rubbing head chamber, and this amount of pressure is applied to a district of rubbing head chamber.
Following Yi San district rubbing head chamber is described the Adaptive inverse control system that is used for chemical-mechanical polishing mathing that the embodiment of the invention provides for example.
As shown in Figure 8, it is three districts that the rubbing head chamber is divided into, and is respectively a district, He San district, two districts, and wherein, the rubbing head chamber output pressure amount in a district is y1, and the rubbing head chamber output pressure amount in two districts is y2, and the rubbing head chamber output pressure amount in three districts is y3.Composite control apparatus 120 comprises the first compound control module, the second compound control module and the 3rd compound control module.
The first compound control module comprises first nerves network controller NNC1, the first pi controller P1 and first weighter.First nerves network controller NNC1 calculates first nerves network control amount u according to the control model of the system of the first goal pressure amount r1 that imports from the goal pressure input unit and 130 identifications of neural network identification device
N1The first pi controller P1 is that y1 calculates the first proportional plus integral control amount u according to the first goal pressure amount r1 that imports from the goal pressure input unit, the rubbing head chamber output pressure amount in a district
P1First weighter is to first nerves network control amount u
N1With the first proportional plus integral control amount u
P1Generate the input pressure amount u1 of the first rubbing head chamber with segmentation VARIABLE PARAMETER PID CONTROL strategy.Wherein, u1=ku
N1+ u
P1
The second compound control module comprises nervus opticus network controller NNC2, the second pi controller P2 and second weighter.Nervus opticus network controller NNC2 calculates nervus opticus network control amount u according to the control model of the system of the second goal pressure amount r2 that imports from goal pressure input unit 110 and 130 identifications of neural network identification device
N2The second pi controller P2 is that y2 calculates the second proportional plus integral control amount u according to the second goal pressure amount r2 that imports from goal pressure input unit 110, the rubbing head chamber output pressure amount in two districts
P2Second weighter is to nervus opticus network control amount u
N2With the second proportional plus integral control amount u
P2Generate the input pressure amount u2 of the second rubbing head chamber with segmentation VARIABLE PARAMETER PID CONTROL strategy.Wherein, u2=ku
N2+ u
P2
The 3rd compound control module comprises third nerve network controller NNC3, the 3rd pi controller P3 and the 3rd weighter.Third nerve network controller NNC3 calculates third nerve network control amount u according to the control model of the system of the 3rd goal pressure amount r3 that imports from goal pressure input unit 110 and 130 identifications of neural network identification device
N3The 3rd pi controller P3 is that y3 calculates the 3rd proportional plus integral control amount u according to the rubbing head chamber output pressure amount in the 3rd goal pressure amount r3 He San district that imports from goal pressure input unit 110
P3The 3rd weighter is to third nerve network control amount u
N3With the 3rd proportional plus integral control amount u
P3Generate the input pressure amount u3 of the 3rd rubbing head chamber with segmentation VARIABLE PARAMETER PID CONTROL strategy.Wherein, u3=ku
N3+ u
P3
Because the rubbing head chamber comprises three subregions, there is coupling between the chamber of subregion, therefore at the back of composite control apparatus 120 series connection neuron decoupling compensation device 140.Particularly, neuron decoupling compensation device 140 comprises peripheral sensory neuron decoupling compensation unit, nervus opticus unit decoupling compensation unit and third nerve unit decoupling compensation unit, and wherein, first is three input neurons to third nerve unit decoupling compensation unit.
Particularly, peripheral sensory neuron decoupling compensation unit is with neuron weights ω
11, ω
12And ω
13Input pressure amount to first to the 3rd rubbing head chamber is weighted calculating, generates the input pressure amount u*1 of the revised first rubbing head chamber, and u*1 is exported to the electric Proportion valve in a district of rubbing head chamber 1.Nervus opticus unit decoupling compensation unit is with neuron weights ω
21, ω
22And ω
23Input pressure amount to first to the 3rd rubbing head chamber is weighted calculating, generates the input pressure amount u*2 of the revised second rubbing head chamber, and u*2 is exported to the electric Proportion valve in two districts of rubbing head chamber 1.Third nerve unit decoupling compensation unit is with neuron weights ω
31, ω
32And ω
33Input pressure amount to first to the 3rd rubbing head chamber is weighted calculating, generates the input pressure amount u*3 of revised the 3rd rubbing head chamber, and u*3 is exported to the electric Proportion valve in three districts of rubbing head chamber 1.Online correction by the neuron weights can realize the online decoupling zero to being coupled between the chamber.
Because two districts of rubbing head chamber are between the He San district, a district, it is bigger influenced by the adjacent area when therefore pressurizeing.Below respectively with 0.5psi, 1psi, the 1.5psi force value is advanced and is pressurizeed in two districts.At first to whether adopting segmentation VARIABLE PARAMETER PID CONTROL strategy to contrast.Fig. 9 a shows k=1 when not adopting segmentation VARIABLE PARAMETER PID CONTROL strategy, the step response curve of gained when pressurizeing at two district 1psi.Fig. 9 b shows the response curve when adopting segmentation VARIABLE PARAMETER PID CONTROL strategy.
From Fig. 9 a and Fig. 9 b as can be seen, when not adopting the segmentation VARIABLE PARAMETER PID CONTROL, because existing, the identification of initial time inversion model influences whole pressure process than mistake, so that the concussion of initiating system, and proportional plus integral control is difficult to correct and compensate the instability relatively thereby system that makes becomes.Unsettled system can cause unstability and the inexactness of online nerve network recognition, thereby has constituted vicious circle.When adopting segmentation VARIABLE PARAMETER PID CONTROL strategy, initial time is because therefore known contrary the inaccurate of identification model of controlling can reduce the proportion of ANN Control amount in the master control amount to retrain the adverse effect that it brings.At this moment, mainly by the main effect of proportional plus integral control amount performance.Along with the stable rising of system pressure, the training time of neutral net lengthens, and will progressively approach object model.At this moment, amplify the ANN Control amount at the proportion of master control amount, until k=1 thereupon.
When system reached steady-state value, ANN Control amount (contrary controlled quentity controlled variable) accounted for 96.8% of master control amount.In addition, for the recessed phenomenon that occurs in the initial pressure ascent stage be since with the three districts pressure decline of two districts couplings, and then cause that two district's volumes increases cause.
Below in conjunction with Figure 10 a to Figure 10 c to adopting segmentation VARIABLE PARAMETER PID CONTROL strategy two districts applied 0.5psi, 1psi and the 1.5psi situation being carried out applied voltage test respectively.Wherein, it is as shown in table 1 that the step response of the two districts pressurization that test obtains is adjusted the time, the adjustment time for remain on ± the interior minimum time Ts of 5% allowable error scope.
Table 1
{ the multi-region of 1.5psi} is carried out applied voltage test simultaneously for 0.5psi, 1psi to adopt segmentation VARIABLE PARAMETER PID CONTROL strategy respectively a district, He San district, two districts to be exerted pressure.Figure 11 a and Figure 11 b show the contrast step response curve contrast before and after the decoupling zero respectively, and table 2 is the adjustment time of step response.
Table 2
From above-mentioned empirical curve and response time as can be seen, decide parameter PID control and have capability of fast response preferably, less overshoot and stability with respect to traditional based on the contrary control of the neural neural network identification of DRNN.Can reduce the influence of coupling between each district of rubbing head chamber simultaneously by neuronic dynamic online decoupling zero, make the operation that each loop can be relatively independent, improve the dynamic characteristic of system.
The Adaptive inverse control system that is used for chemical-mechanical polishing mathing that the embodiment of the invention provides utilizes DRNN Neural Network Online identification objects model capability and neuron decoupling zero control technology.Particularly, utilize the non-linear approximation capability of neutral net and learning ability can be online the accurate inversion model of identification original system in real time, thereby change when can be used for, decoupling zero non-linear and the characteristic unknown object are controlled, avoid off-line to find the solution the inversion model of Complex Nonlinear System, thereby widened the scope of application of method of inverse to a certain extent.In addition, segmentation VARIABLE PARAMETER PID CONTROL strategy has reduced owing to the inaccurate adverse effect of bringing of contrary control Model Distinguish in system's initial press process, makes The whole control system relatively stable.Adopt the Adaptive inverse control system that is used for chemical-mechanical polishing mathing of the embodiment of the invention not only to make control system have on-line identification preferably and decoupling zero ability, and it is fast to make that by the compound control strategy of segmentation variable element system has response speed, overshoot is few, characteristics such as good and good stability of robustness.In addition, the Adaptive inverse control system that is used for chemical-mechanical polishing mathing that the embodiment of the invention provides decides parameter PID than routine and has very strong adaptive ability, has certain construction value and reference significance.
Describe and to be understood that in the flow chart or in this any process of otherwise describing or method, expression comprises module, fragment or the part of code of the executable instruction of the step that one or more is used to realize specific logical function or process, and the scope of preferred embodiment of the present invention comprises other realization, wherein can be not according to order shown or that discuss, comprise according to related function by the mode of basic while or by opposite order, carry out function, this should be understood by the embodiments of the invention person of ordinary skill in the field.
In flow chart the expression or in this logic of otherwise describing and/or step, for example, can be considered to the sequencing tabulation for the executable instruction that realizes logic function, may be embodied in any computer-readable medium, use for instruction execution system, device or equipment (as the computer based system, comprise that the system of processor or other can be from the systems of instruction execution system, device or equipment instruction fetch and execution command), or use in conjunction with these instruction execution systems, device or equipment.With regard to this specification, " computer-readable medium " can be anyly can comprise, storage, communication, propagation or transmission procedure be for instruction execution system, device or equipment or the device that uses in conjunction with these instruction execution systems, device or equipment.The example more specifically of computer-readable medium (non-exhaustive list) comprises following: the electrical connection section (electronic installation) with one or more wirings, portable computer diskette box (magnetic device), random-access memory (ram), read-only storage (ROM), can wipe and to edit read-only storage (EPROM or flash memory), fiber device, and portable optic disk read-only storage (CDROM).In addition, computer-readable medium even can be paper or other the suitable media that to print described program thereon, because can be for example by paper or other media be carried out optical scanner, then edit, decipher or handle to obtain described program in the electronics mode with other suitable methods in case of necessity, then it is stored in the computer storage.
Should be appreciated that each several part of the present invention can realize with hardware, software, firmware or their combination.In the above-described embodiment, a plurality of steps or method can realize with being stored in the memory and by software or firmware that suitable instruction execution system is carried out.For example, if realize with hardware, the same in another embodiment, in the available following technology well known in the art each or their combination realize: have for the discrete logic of data-signal being realized the logic gates of logic function, special IC with suitable combinational logic gate circuit, programmable gate array (PGA), field programmable gate array (FPGA) etc.
Those skilled in the art are appreciated that and realize that all or part of step that above-described embodiment method is carried is to instruct relevant hardware to finish by program, described program can be stored in a kind of computer-readable recording medium, this program comprises one of step or its combination of method embodiment when carrying out.
In addition, each functional unit in each embodiment of the present invention can be integrated in the processing module, also can be that the independent physics in each unit exists, and also can be integrated in the module two or more unit.Above-mentioned integrated module both can adopt the form of hardware to realize, also can adopt the form of software function module to realize.If described integrated module realizes with the form of software function module and during as independently production marketing or use, also can be stored in the computer read/write memory medium.
The above-mentioned storage medium of mentioning can be read-only storage, disk or CD etc.
In the description of this specification, concrete feature, structure, material or characteristics that the description of reference term " embodiment ", " some embodiment ", " example ", " concrete example " or " some examples " etc. means in conjunction with this embodiment or example description are contained at least one embodiment of the present invention or the example.In this manual, the schematic statement to above-mentioned term not necessarily refers to identical embodiment or example.And concrete feature, structure, material or the characteristics of description can be with the suitable manner combination in any one or more embodiment or example.
Although illustrated and described embodiments of the invention, for the ordinary skill in the art, be appreciated that without departing from the principles and spirit of the present invention and can carry out multiple variation, modification, replacement and modification to these embodiment that scope of the present invention is by claims and be equal to and limit.
Claims (7)
1. an Adaptive inverse control system that is used for chemical-mechanical polishing mathing is characterized in that, comprising: goal pressure amount input unit, composite control apparatus, neural network identification device, wherein,
Described goal pressure amount input unit is used for to the goal pressure amount of the rubbing head chamber of described composite control apparatus input chemical-mechanical polishing mathing;
Described neural network identification device is used for the control model according to the described Adaptive inverse control system for chemical-mechanical polishing mathing of the input pressure amount of rubbing head chamber and the identification of rubbing head chamber output pressure amount; With
Described composite control apparatus is used for calculating according to described goal pressure amount, described rubbing head chamber output pressure amount and described control model the input pressure amount of described rubbing head chamber, wherein, described composite control apparatus comprises at least one group of compound control module, wherein, every group of described compound control module is corresponding continuous with each district of the rubbing head chamber of described chemical-mechanical polishing mathing, and every group of described compound control module comprises:
Nerve network controller, described nerve network controller receive from the described goal pressure amount of described goal pressure amount input unit and described control model and generate the ANN Control amount;
Pi controller, described proportional plus integral control generates the proportional plus integral control amount according to the amount of pressure of described goal pressure amount, the output of described rubbing head chamber; With
Weighter, described weighter links to each other with described pi controller with described nerve network controller respectively, be used for described ANN Control amount and described proportional plus integral control amount are generated the input pressure amount of described rubbing head chamber with segmentation VARIABLE PARAMETER PID CONTROL strategy, wherein
u=ku
n+u
p，
Wherein, u is input pressure amount, the u of described rubbing head chamber
nBe described ANN Control amount, u
pBe described proportional plus integral control amount, k is the segmentation variable element.
2. Adaptive inverse control as claimed in claim 1 system is characterized in that described segmentation variable element k is:
Wherein, a
0Initial value, t for k
0For time end point, the Δ k of initial time section is that variable quantity, the Δ t of described segmentation variable element k is the variable quantity of adjacent two periods, and the number of fragments of described segmentation variable element k is n+2.
3. Adaptive inverse control as claimed in claim 1 system is characterized in that the quantity of described compound control module equates with the quantity of the rubbing head chamber of described chemical-mechanical polishing mathing.
4. Adaptive inverse control as claimed in claim 1 system, it is characterized in that, when the quantity of described rubbing head chamber is two or more, described Adaptive inverse control system for chemical-mechanical polishing mathing also comprises: neuron decoupling compensation device, rubbing head chamber with described composite control apparatus and described chemical-mechanical polishing mathing links to each other respectively, be used for the input pressure amount of the described rubbing head chamber of described composite control apparatus output is revised and generated the input pressure amount of revised rubbing head chamber, and the rubbing head chamber that the input pressure amount of described revised rubbing head chamber is exported to described chemical-mechanical polishing mathing.
5. Adaptive inverse control as claimed in claim 4 system, it is characterized in that, described neuron decoupling compensation device comprises a plurality of neuron decoupling compensations unit, the input of each described neuron decoupling compensation unit links to each other with the output of the weighter of a plurality of described compound control modules, the output of each described neuron decoupling compensation unit links to each other with a district of described rubbing head chamber, be used for the input pressure amount of described rubbing head chamber is carried out the input pressure amount that decoupling zero adjustment generates one tunnel correction back rubbing head chamber, and the described input pressure amount of revising back rubbing head chamber exported to a district of the rubbing head chamber of described chemical-mechanical polishing mathing.
6. Adaptive inverse control as claimed in claim 5 system, it is characterized in that each described neuron decoupling compensation unit is weighted calculating to generate the input pressure amount of one tunnel revised rubbing head chamber to a plurality of input pressure amounts of the rubbing head chamber of described chemical-mechanical polishing mathing with different neuron weights.
7. Adaptive inverse control as claimed in claim 5 system is characterized in that described neuron decoupling compensation unit equates with the quantity of described compound control module.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 201110388573 CN102490120B (en) | 2011-11-29 | 2011-11-29 | Adaptive inverse control system for chemical-mechanical polishing machine |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN 201110388573 CN102490120B (en) | 2011-11-29 | 2011-11-29 | Adaptive inverse control system for chemical-mechanical polishing machine |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102490120A CN102490120A (en) | 2012-06-13 |
| CN102490120B true CN102490120B (en) | 2013-10-09 |
Family
ID=46182002
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN 201110388573 Expired - Fee Related CN102490120B (en) | 2011-11-29 | 2011-11-29 | Adaptive inverse control system for chemical-mechanical polishing machine |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102490120B (en) |
Families Citing this family (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103345159B (en) * | 2013-07-03 | 2015-08-05 | 江苏大学 | Based on the hybrid electric vehicle BSG system control method of neural Network Adaptive Inversion |
| CN105773397B (en) * | 2016-03-09 | 2017-09-29 | 天津华海清科机电科技有限公司 | Chemically-mechanicapolish polish multi-region pressure On-line Control algorithm |
| US10387298B2 (en) | 2017-04-04 | 2019-08-20 | Hailo Technologies Ltd | Artificial neural network incorporating emphasis and focus techniques |
| US11238334B2 (en) | 2017-04-04 | 2022-02-01 | Hailo Technologies Ltd. | System and method of input alignment for efficient vector operations in an artificial neural network |
| US11615297B2 (en) | 2017-04-04 | 2023-03-28 | Hailo Technologies Ltd. | Structured weight based sparsity in an artificial neural network compiler |
| US11551028B2 (en) | 2017-04-04 | 2023-01-10 | Hailo Technologies Ltd. | Structured weight based sparsity in an artificial neural network |
| US11544545B2 (en) | 2017-04-04 | 2023-01-03 | Hailo Technologies Ltd. | Structured activation based sparsity in an artificial neural network |
| CN110684547A (en) * | 2019-10-22 | 2020-01-14 | 中国计量大学 | Optimized control method for biomass pyrolysis carbonization kiln |
| US11237894B1 (en) | 2020-09-29 | 2022-02-01 | Hailo Technologies Ltd. | Layer control unit instruction addressing safety mechanism in an artificial neural network processor |
| US11263077B1 (en) | 2020-09-29 | 2022-03-01 | Hailo Technologies Ltd. | Neural network intermediate results safety mechanism in an artificial neural network processor |
| US11221929B1 (en) | 2020-09-29 | 2022-01-11 | Hailo Technologies Ltd. | Data stream fault detection mechanism in an artificial neural network processor |
| US11811421B2 (en) | 2020-09-29 | 2023-11-07 | Hailo Technologies Ltd. | Weights safety mechanism in an artificial neural network processor |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102133729A (en) * | 2011-01-06 | 2011-07-27 | 清华大学 | Pressure control system for CMP (Chemical Mechanical Polishing) head |
| CN102205522A (en) * | 2011-05-24 | 2011-10-05 | 清华大学 | Gas path positive pressure system used for chemico-mechanical polishing and chemico-mechanical polishing equipment |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR100979071B1 (en) * | 2002-02-22 | 2010-08-31 | 에이저 시스템즈 인크 | Chemical mechanical polishing of dual orientation polycrystalline materials |
| US6926584B2 (en) * | 2002-10-09 | 2005-08-09 | Taiwan Semiconductor Manufacturing Co., Ltd. | Dual mode hybrid control and method for CMP slurry |
-
2011
- 2011-11-29 CN CN 201110388573 patent/CN102490120B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102133729A (en) * | 2011-01-06 | 2011-07-27 | 清华大学 | Pressure control system for CMP (Chemical Mechanical Polishing) head |
| CN102205522A (en) * | 2011-05-24 | 2011-10-05 | 清华大学 | Gas path positive pressure system used for chemico-mechanical polishing and chemico-mechanical polishing equipment |
Non-Patent Citations (2)
| Title |
|---|
| 低压力铜化学机械抛光集散控制系统;门延武等;《计算机集成制造系统》;20101231;第16卷(第12期);第2653-2659页 * |
| 门延武等.低压力铜化学机械抛光集散控制系统.《计算机集成制造系统》.2010,第16卷(第12期),第2653-2659页. |
Also Published As
| Publication number | Publication date |
|---|---|
| CN102490120A (en) | 2012-06-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102490120B (en) | Adaptive inverse control system for chemical-mechanical polishing machine | |
| Hassani et al. | A survey on hysteresis modeling, identification and control | |
| Guo et al. | Parametric adaptive control of single-rod electrohydraulic system with block-strict-feedback model | |
| CN109839823B (en) | Asynchronous hysteresis compensation-linear quadratic form H of piezoelectric deformable mirror∞Control method and system | |
| US20110276150A1 (en) | Neural network optimizing sliding mode controller | |
| CN103324093B (en) | A kind of multi-model Adaptive Control system and control method thereof | |
| CN102998973A (en) | Multi-model self-adaptive controller of nonlinear system and control method | |
| Khalil et al. | Implementation of speed controller for rotary hydraulic motor based on LS-SVM | |
| CN108181818A (en) | Containing not modeling the dynamic electro-hydraulic position servo system Robust Adaptive Control method of friction | |
| Matias et al. | Using a neural network for estimating plant gradients in real-time optimization with modifier adaptation | |
| Jami'In et al. | An adaptive predictive control based on a quasi-ARX neural network model | |
| Peng et al. | Novel models for one-sided hysteresis of piezoelectric actuators | |
| CN108959787B (en) | Thermal deformation prediction method and system of macro-macro dual-drive system considering actual working conditions | |
| Zagorowska et al. | Online feedback optimization of compressor stations with model adaptation using Gaussian process regression | |
| CN107490962B (en) | Data-driven optimal control method for servo system | |
| Salim et al. | Position control of pneumatic actuator using an enhancement of NPID controller based on the characteristic of rate variation nonlinear gain | |
| CN110393954B (en) | Thickener online control method based on reinforcement learning | |
| CN113043266A (en) | Adaptive force tracking control method based on iterative learning | |
| CN109324503B (en) | Multilayer neural network motor system control method based on robust integration | |
| Bahita et al. | Neural stable adaptive control for a class of nonlinear systems without use of a supervisory term in the control law | |
| CN115130771A (en) | Torque prediction method of hydraulic retarder based on long-term and short-term memory network LSTM | |
| CN101640969B (en) | Matching method and plasma device applying same | |
| Lin et al. | A hybrid model combining simulation and optimization in chemical mechanical polishing process | |
| Sun et al. | Model reference adaptive control with adjustable gain for piezoelectric actuator | |
| Gao et al. | Beam-pumping unit energy-saving control system based on support vector machine |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20131009 Termination date: 20211129 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |