Method for determining efficiently parameters in chemical-mechanical polishing (CMP)

Abstract

A method of determining efficiently parameters in chemical-mechanical polishing, especially a method includes a Neural-Taguchi method to seek the best parameter set to increase the quantity of output. This invention involves a time parameter to achieve end-point detection on line during the CMP procedure and also completes a maximum material removal rate (MRR) and a minimum within wafer non-uniformity (WIWNU) at the same time.

Description

BACKGROUND OF THE INVENTION

[0001] 1. Field of the Invention

[0002] The present invention relates a method of determining efficient parameters in chemical-mechanical polishing, wherein the method is based on imitating a virtual CMP process in neural network progress and then determining a parameter set inputted into an electronic polishing machine to make wafer polishing efficient.

[0003] 2. Description of Related Art

[0004] A chemical-mechanical polishing (CMP) is a process used to manufacture semiconductor wafers. Typically, the chemical-mechanical polishing involves rotating a thin, flat semiconductor wafer against a polishing pad, rotating the polishing pad against the wafer, or both. A chemical slurry containing a polishing agent such as alumina or silica, acts as an abrasive medium between the wafer and the pad and makes the wafer have an excellent planar surface after polishing. The CMP procedure includes a chemical reaction and a mechanical movement, therefore several coefficients such as the down force of the wafer, platen speed, solid content of the chemical slurry, and pH of the chemical slurry etc. significantly influence the efficiency of the polishing.

[0005] The conventional CMP does not provide a suitable method for detecting a planar end-point. For example, one conventional method requires a technician to remove a semiconductor wafer from the chemical mechanical polishing process, inspect the wafer for the desired end-point, and then return the wafer to the process if the desired end-point is not observed. This is obviously time-consuming and inefficient. How to control the polishing at an end-point precisely is a major problem in all kinds of wafer polishing.

[0006] Several types of end-point detection having no need to remove the semiconductor from the production line have been invented:

[0007] 1. Motor current method;

[0008] 2. Pad thermal image;

[0009] 3. Optic interferometer method;

[0010] 4. Acoustic method; and

[0011] 5. Small vibration method.

[0012] Although the above-mentioned detecting methods are based on different theories, those methods all need a precise close-loop to assist in controlling the wafer polishing. However, mature controlling techniques of close-loop are not common, so there is a requirement of an excellent controlling technique that can perfectly combine the CMP procedure and the end-point detection on line to increase the quantity of output.

[0013] To overcome the existing problem in chemical-mechanical polishing, the present invention tends to provide a method of determining efficient parameters in chemical-mechanical polishing, which achieve high efficiency in the CMP procedure.

[0014] Other objects, advantages and novel features of the invention will become more apparent from the following detailed description when taken in conjunction with the accompanying tables and figures.

SUMMARY OF THE INVENTION

[0015] The main objective of the invention is to provide a method of determining efficiently parameters in chemical-mechanical polishing, especially the parameters of this method involve a time parameter and the parameters are set at the best value.

[0016] Another objective of the invention is to provide a method of determining efficiently parameters in chemical-mechanical polishing, whereby the CMP procedure and the end-point detection are efficiently combined on line to increase the quantity of output and diminish the polishing time.

[0017] Still another objective of the invention is to provide a method of determining efficiently parameters in chemical-mechanical polishing, wherein the maximum material removal rate (MRR) and the minimum within wafer non-uniformity (WIWNU) are achieved in the CMP procedure.

BRIEF DESCRIPTION OF THE DRAWINGS

[0018] FIG. 1 is a response figure of material removal rate (MRR);

[0019] FIG. 2 is a response figure of standard deviation of t2(STDEV); and

[0020] FIG. 3 is an illustration of imitating virtual CMP process in neural network progress in this invention.

DETAILED DESCRIPTION OF PREFERRED EMBODIMENT

[0021] In this invention, an analyzing system named Neural-Taguchi experimental method is employed in a computer to determine a set of efficient parameters in a CMP process. The Neural-Taguchi method combines the neural networks with the Taguchi experimental method, a computer analysis system with a parameter setting system, and is used in modifying the essential parameters in the CMP procedure. Steps to determine the parameters of this invention are illustrated in accompaniment with experimental data in tables and figures.

[0022] 1. Using the Taguchi experiment to design and conduct the CMP procedure:

[0023] An oxide layer on the wafer is a removed target in this experiment because it is obtained easily. Four essential parameters are picked out as changeable parameters and adjusted during the experiment. The four parameters are:

[0024] A. Solid content;

[0025] B. Down force;

[0026] C. Back pressure;

[0027] D. Platen speed;

[0028] The Preston equation shows a relationship of removal rate with polishing pressure and polishing speed, and the Preston equation is shown as following: 1$\mathrm{Removal}\mathrm{rate}=\frac{\Delta H}{\Delta t}=\mathrm{kp}\times P\times \frac{\Delta s}{\Delta t}$

[0029] where ΔH: variation of surface thickness;

[0030] Δt: polishing time;

[0031] Kp: coefficient of polishing;

[0032] P: polishing pressure;

[0033] ΔS/ Δt: linear speed of the wafer comparative to polishing pad.

[0034] According to the Preston equation, the removal rate relates to the polishing pressure and polishing speed in a direct proportion. Furthermore, the polishing pressure is calculated from dividing the down pressure by the contact area of the polishing pad. When the polishing time changes, a uniformity of a wafer surface is different, and the contact area and the down pressure respectively change, too. Therefore, a material removal rate and a within wafer non-uniformity (WIWNU) are functions of the polishing time and the polishing time is picked out to be a fifth (E) elementary parameter of a CMP process. Therefor, all essential parameters are:

[0035] A. Solid content;

[0036] B. Down force;

[0037] C. Back pressure;

[0038] D. Platen speed; and

[0039] E. Polishing time.

[0040] In the design of experiment, an L25(5 ×5) orthogonal array as shown in Table 1 is adapted.

TABLE 1
L25 (5 × 5) orthogonal array in each trial in a CMP procedure.
	A	B	C	D	E
	Solid	Down	Back	Platen	Polish
Col.	L25 (55)	Content	Force	Pressure	Speed	Time
Row	A	B	C	D	E	(wt %)	(psi)	(psi)	(rpm)	(sec)
1	1	1	1	1	1	5%	4	0	20	40
2	1	2	2	2	2	5%	5	1	30	45
3	1	3	3	3	3	5%	6	2	40	50
4	1	4	4	4	4	5%	7	3	50	55
5	1	5	5	5	5	5%	8	3.5	60	60
6	2	1	4	2	3	10%	4	3	30	50
7	2	2	5	3	4	10%	5	3.5	40	55
8	2	3	1	4	5	10%	6	0	50	60
9	2	4	2	5	1	10%	7	1	60	40
10	2	5	3	1	2	10%	8	2	20	45
11	3	1	2	3	5	15%	4	1	40	60
12	3	2	3	4	1	15%	5	2	50	40
13	3	3	4	5	2	15%	6	3	60	45
14	3	4	5	1	3	15%	7	3.5	20	50
15	3	5	1	2	4	15%	8	0	30	55
16	4	1	5	4	2	20%	4	3.5	50	45
17	4	2	1	5	3	20%	5	0	60	50
18	4	3	2	1	4	20%	6	1	20	55
19	4	4	3	2	5	20%	7	2	30	60
20	4	5	4	3	1	20%	8	3	40	40
21	5	1	3	5	4	25%	4	2	60	55
22	5	2	4	1	5	25%	5	3	20	60
23	5	3	5	2	1	25%	6	3.5	30	40
24	5	4	1	3	2	25%	7	0	40	45
25	5	5	2	4	3	25%	8	1	50	50

[0041] This orthogonal array has 25 columns 5 rows, where 5 rows present 5 control variable from A to E. Five levels on each variable covering the domain of interest are set. The dependent variables to be observed are MRR and WIWNU, where the MRR is the bigger the better and the WIWNIJ is the smaller the better. The MRR and the WIWNU are calculated by the following methods and equations:

[0042] (a). Measure a thickness (t1) of the wafer at nine points by an AFT.

[0043] (b). Measure a thickness (t2) of the wafer after polishing in (T) minutes;

[0044] (c). Polishing speed =(t1−t2)/T, an average of polishing speeds in each point is the material removal rate (MRR).

[0045] (d). Within wafer non-uniformity (WIWNU) =(σ: standard deviation of t2)/(mean of t2)×100%

[0046] Since it only makes sense to compare WIWNUs after wafer have been polished and have similar thickness, we replace WIWNU with standard deviation of t2 in this invention.

[0047] 2. Signal/Noise ratio analysis:

[0048] Control factors and noise factors greatly influence the efficiency in the CMP procedure and the signal(S)/noise(N) ratio is an index to indicate the quality of output. If the S/N ratio is high, the efficiency of the procedure and the quality of output wafer are good.

[0049] As shown in table 2, the S/N ratio of the MRR and the standard deviation of t2 (STDEV (t2)) are calculated respectively based on the L25(55) orthogonal array, and then are used to make response tables (Tabs. 3 and 4) and response figures (FIGS. 1 and 2) with the five parameters. Lastly, analysis of variance (ANOVA) is used in estimating contributing degrees of each parameter.

TABLE 2
an experimental results and S/N ratio table
	L25 (55)	MRR
	A	B	C	D	E	(A/S)	(S/N)	σt2	(S/N)
1	1	1	1	1	1	9.14	19.22	137.540	17.23
2	1	2	2	2	2	14.5	23.23	109.591	19.20
3	1	3	3	3	3	16.34	24.26	99.567	20.04
4	1	4	4	4	4	20.31	26.15	128.673	17.81
5	1	5	5	5	5	20.55	26.26	98.534	20.13
6	2	1	4	2	3	16.09	24.13	51.999	25.68
7	2	2	5	3	4	23.45	27.4	152.866	16.31
8	2	3	1	4	5	28.92	29.22	298.643	10.50
9	2	4	2	5	1	34.64	30.79	88.683	21.04
10	2	5	3	1	2	22.51	27.05	112.108	19.01
11	3	1	2	3	5	19.44	25.77	92.238	20.70
12	3	2	3	4	1	27.68	28.84	114.126	18.85
13	3	3	4	5	2	36.89	31.34	114.363	18.83
14	3	4	5	1	3	22.11	26.89	55.401	25.13
15	3	5	1	2	4	31.84	30.06	110.232	19.15
16	4	1	5	4	2	20.44	26.21	100.058	19.99
17	4	2	1	5	3	29.22	29.31	169.638	15.41
18	4	3	2	1	4	18.37	25.28	167.419	15.52
19	4	4	3	2	5	28.72	29.16	114.870	18.80
20	4	5	4	3	1	39.28	31.88	69.075	23.21
21	5	1	3	5	4	23.25	27.33	113.494	18.90
22	5	2	4	1	5	17.1	24.66	46.615	26.63
23	5	3	5	2	1	25.74	28.21	52.903	25.53
24	5	4	1	3	2	34.59	30.78	139.578	17.10
25	5	5	2	4	3	44.11	32.89	206.076	13.72

[0050]

TABLE 3
a response table of material removal rate (MRR)
	A	B	C	D	E
	Solid	Down	Back	Platen	Polishing
Factor	Content	Force	Pressure	Speed	time
Level	(wt %)	(psi)	(psi)	(rpm)	(sec)
1	23.80363	24.53263	27.71867	24.62076	27.78998
2	27.71891	26.68942	27.59314	26.95966	27.72017
3	28.58194	27.66443	27.32932	28.02038	27.49836
4	28.37092	28.75611	27.63418	28.66399	27.24576
5	28.77449	29.60729	26.97457	28.9851	26.99562

[0051]

TABLE 4
a response table of the standard deviation of t2.
	A	B	C	D	E
	Solid	Down	Back	Platen	Polishing
Factor	Content	Force	Pressure	Speed	time
Level	(wt %)	(psi)	(psi)	(rpm)	(sec)
1	18.88245	20.50176	15.8791	20.7043	21.17418
2	18.50823	19.28189	18.03852	21.6729	18.82889
3	20.53432	18.08461	19.11871	19.4741	19.99524
4	18.58753	19.97647	22.43348	16.8772	17.54044
5	20.37666	19.04445	21.41936	20.6428	19.35043

[0052] The response tables show the effect of each factor at different levels and the response figures illustrate the trend of effect changes between levels. From the response tables and response plot, we can find the value of each factor that can maximize S/N ratio, i.e. the optimal value. Combining all the factors at the optimal level, we get the optimal parameters set with regard to the MRR and the WIWNU (in terms of σt2)

[0053] According to FIG. 1, A5B5C1D5E1 is an optimum parameter set of the MRR and according to FIG. 2, A3 B1C4D2E1 is an optimum parameter set of the standard deviation of t2. In FIG. 1, the back pressure (C) and the polishing time (E) are less influential than other parameters in MRR. But the WIWNU (in terms of the standard deviation of t2) is influenced by all five parameters to a certain extent.

[0054] 3.CMP process modeling:

[0055] Since CMP is a very complicated system, this invention tries to use highly non-linear neural network to stimulate it. The experimental results in Tab. 2 are used as training sample, and normalize the desired output between +0.8 ˜−0.8 for convenience of the neural network training. The neural network of this invention is shown in FIG. 3, which modifies the parameters and the coefficients automatically.

[0056] In the neural network training, the control variables in Tab. 1 are inputs in the training, the MRR and the standard deviation of t2(σt2) are the outputs. However, the MRR and the standard deviation of t2( σt2) are contrary in the request to achieve the optimum efficiency the MRR: the bigger the better; σt2:the smaller the better), therefore a hybird performance index is used here:

J=a×MRR (S/N)+b×σ12(S/N)

[0057] This performance index includes the MRR and the σt2 both, wherein an MRR(S/N) and an σt2(S/N) are optimum in a largest mode in questing the best parameters(S/N), therefore, a magnifying response is popular in this function and is a direction to adjust the parameters in this function. In order to make sure that the CMP works rapidly and the WIWNU remains in desired range, coefficients “a” and “b” are selected to be 0.6 and 0.4, respectively.

[0058] 4. CMP processes parameters optimization:

[0059] Since the values at different levels in orthogonal array are discrete, the optimal parameters set from S/N ratio analysis is still a relatively rough estimation. Additionally, the output of the network as shown in FIG. 3 is the MRR and σt2 combined performance index. Therefore, the optimal parameters set can not be obtained from S/N analysis directly. In order to acquire a more precise optimal parameters set, the neural network based implicit CMP model together with the parameters optimization method is adopted to fine-tune the parameters. The fine-tune steps are as follows:

[0060] (1) Put the local optimal parameters set from S/N ratio analysis into the neural network and calculate network output value (J)max.

[0061] (2) Calculate 2$\frac{\partial j}{\partial x_i}$

[0062] around (J)max, change x1 within allowable range to increase 3$\frac{\partial j}{\partial x_i}.$

[0063] If 4$\frac{\partial j}{\partial x_i}$

[0064] is positive, increase x1; if 5$\frac{\partial j}{\partial x_i}$

[0065] is negative, decrease x1. When x1 is not changeable, jump to step (4). ( x1 represents the network input)

[0066] (3) Use neural network to calculate J, If J>(J)max, set (J)max=J, and go back to step (2).

[0067] (4) The parameters set obtained from the above steps are the optimal parameters with regard to MRR and σt2

[0068] According to above steps, the best parameters set is:

[0069] Solid content: 20%;

[0070] Down force: 8.0 psi;

[0071] Back pressure: 3.2 psi;

[0072] Platen speed: 41 rpm; and

[0073] Polishing time: 42 sec

[0074] Wherein the 42 seconds is a optimal polishing time when other parameters are set as above i.e. the processes can be stopped after 42 seconds to achieve maximal removal rate and acceptable WIWNU.

[0075] In order to prove that the optimal parameter set in above-mentioned description really works, testing experiments were carried out at the National Nano Device Lab. (NDL) of Taiwan, R.O.C. The experimental data is shown in Tab. 5, wherein the optimal parameters set is in comparison with an original parameters set currently used at the NDL. Since the carrier speed and oscillation speed have little influence on the MRR and the WIWNU, the primitive settings of those two parameters by NDL are adopted.

TABLE 5
a comparison table of the parameters in best
mode and the original mode
		Original setting	Optimal setting
	Solid Content	15%	20%
	Down Force	7	psi	8	psi
	Back Pressure	3	psi	3.2	psi
	Platen Speed	20	rpm	41	rpm
	Carrier Speed	25	rpm	25	rpm
	Oscillation Speed	2	mm/sec	2	mm/sec
	Polishing Time	60	s	42	s
	Average of removal	1599.2	A°	1897.9	A°
	amount
	Average of thickness	3509.9	A°	3149.9	A°
	after polishing
	WIWNU of thickness	112.2	A°	118.0	A°
	after polishing
	Experimental MRR	26.65	A°/s	45.19	A°/s
Experimental WIWNU	3.1963%	3.7506%

[0076] When this optimal parameters set is implemented, the polishing processes could be completed in optimal time with the WIWNU staying in acceptable quality. In the experiment for verification, the proposed method offers a better polishing performance while reducing the polishing time by ⅓. The testing experiment proves that the optimal parameters set created from this invention exactly makes the CMP procedure efficient and achieves the end-point detection with no need to remove the wafer from the production lined. Therefore, manufacturers do not have to change equipment in hand or add extra instruments but have the precise CMP control. In the view of cost and efficiency, this invention is really practicable and economical to wafer manufacturers.

[0077] However, it is to be understood that the disclosure is illustrative only, and changes may be made in detail, especially in different materials which are needed to be removed from a surface of the wafer.

Claims

1. A method for determining efficient parameters in chemical-mechanical polishing processes comprising the steps of: a. using a Taguchi experiment to lead the CMP procedure and determine essential parameters to calculate a maximum material rate (MRR) and the within wafer non-uniformity (WIWNU), wherein four essential coefficients are picked out as changeable parameters and adjusted during the experiment and a polishing time is set to be a fifth parameter to complete a model, each parameter is tested under different levels to calculate the (MRR) and the (WIWNU); b. analyzing S/N ratios of the MRR and the WIWNU, wherein the S/N ratios of the MRR and the WIWNU in each trial are calculated and used to make response tables and response figures, and then analysis of variance (ANOVA) is employed in estimating contributing degrees of each parameter to decide a parameter set of the MRR and the WIWNU; c. establishing the CMP process with the parameter set of the MRR and the WIWNU, wherein a neural network simulates the CMP procedure and optimize the parameter automatically, data of step A are a training samples of the neural network, wherein a standard deviation of t2( σt2) after polishing and the MR are outputs and then define a performance index includes the MRR and the σt2, said performance index is optimum in its largest mode; and d. optimizing the parameters set of the CMP procedure, wherein the neural network based CMP model in step C is used in this step and has a fine-tuning function to modify each parameter of the parameter set to be the best ones, and then a optimum parameter set is obtained.

2. The method of determining efficiently parameters in chemical-mechanical polishing as claimed in claim 1, wherein the parameters in step A are: Solid content; Down force; Back pressure; Platen speed; and Polishing time, wherein the down force is the pressure (psi) to the wafer; the platen speed is the rotating speed (rpm) of a polishing platen; the solid content is a particle composition (%) of the slurry; the back pressure is the pressure (psi) of a polishing pad; and the polishing time is a deciding time (sec) to stop the CMP polishing.

3. The method of determining efficiently parameters in chemical-mechanical polishing as claimed in claim 1, wherein the fine-tuning function comprises the following steps: a. put the local optimal parameters set from S/N ratio into the neural network and calculate network output value (J)max. b. Calculate 6∂j∂xiaround (J)max, change x1 within allowable range to increase 7∂j∂xi.If 8∂j∂xiis positive, increase x1 and if 9∂j∂xiis negative, decrease x1. When x1 is not changeable, jump to step (4). (x1 represents the network input) c. Use neural network to calculate J. If J>(J)max, set (J)max=J, and go back to step (2). d. d. The parameters set obtained from the above steps are the optimal parameters with regard to MRR and σt2

4. The method of determining efficiently parameters in chemical-mechanical polishing as claimed in claim 1, wherein the performance index (J) in step C is:

5. The method of determining efficiently parameters in chemical-mechanical polishing as claimed in claim 4, wherein a coefficient a is 0.6 and a coefficient b is 0.4.

6. The method of determining efficiently parameters in chemical-mechanical polishing as claimed in claim 1,wherein the best parameter set is: solid content: 20%; down force: 8.1 psi; back pressure: 3.2 psi; platen speed: 41 rpm; and polishing time: 42 seconds;