Patent Grant
6893882
(1) Field of the Invention
The present invention relates to the fabrication of integrated circuit devices, and more particularly, to a method of monitoring tool performance using a multivariate advanced process control system in the fabrication of integrated circuit devices.
(2) Description of the Prior Art
Run by Run (RBR) feedback control schemes have been used widely to control and monitor the semiconductor manufacturing processes. Statistical process control is combined with feedback control and uses data from past runs to adjust the recipe for the next run. This method offers the potential for reducing process variability caused by equipment aging, chemical depletion, or fluctuation in ambient conditions with a minimal cost. Most literature in the RBR control, however, is only applicable to the manufacturing process with a single product by a single tool. For a foundry industry, due to different customer requirements, one of the most popular production styles is multiple products by multiple tools. For example, if a process runs eight products by five tools, then there are forty combinations which will lead to a very complex situation using conventional RBR control schemes. If separate feedback schemes are set up for each combination, besides affecting the efficiency of the RBR control schemes, this method also ignores the useful information between the tools and products. For example, under a specific tool, an individual control scheme does feedback only on a specific product and it loses the opportunity to adjust for other products produced by the same tool. In addition, since customer orders may not be continuous and scheduled tool set-up may change from time to time, the feedback scheme for any production lot cannot be successfully applied to the next production lot. Therefore, it is desired to provide a multiple run by run control process.
A number of patents address process control issues. For example, U.S. Pat. No. 5,943,550 to Fulford, Jr. et al teaches a method of controlling transistor drive current across a wafer by varying light exposure based on parameters such as gate oxide thickness, RTA temperature, polyetch bias, etc. U.S. Pat. No. 6,148,239 to Funk et al shows a process control system using feed forward control to label materials in groups having common properties based on process variables, tool variables, and material variables. U.S. Pat. No. 6,248,602 to Bode et al discloses a method to store control input parameters such as for overlay control in a photolithography process, process a lot of devices, collect and analyze data for the lot, and perform automated rework based on the data analysis. U.S. Pat. No. 6,245,581 to Bonser et al teaches a method of measuring critical dimension, analyzing the measurement, and performing a secondary etch process based on the analysis.
Accordingly, it is a primary object of the invention to provide a multiple run by run process control method.
Another object of the invention is to provide a multiple run by run process control method that works across tools and products.
Yet another object is to provide a multiple run by run process to detect tool aging.
Yet another object is to provide a multiple run by run process to detect photolithography tool aging.
A further object is to provide a multiple run by run process to detect photolithography tool aging based on exposure dose.
A still further object is to provide a multiple run by run process to update a recipe based on past data.
Another object of the invention is to provide a multiple run by run process to update a recipe for a particular tool and product based on tool decay and variation shift of product.
In accordance with the objects of the invention, a multiple run by run process is achieved. A plurality of tools and a plurality of products to be run on the tools are provided. Tool effects and product effects on a parameter are identified for each tool and each product. A desired recipe is calculated for each product on each tool based on the tool effects and product effects identified. Thereafter, the plurality of products is run on the plurality of tools. The desired recipe is updated after each run of each tool. Tool aging is calculated after each run of each tool based on the desired recipe used.
In the accompanying drawings forming a material part of this description, there is shown:
The process of the present invention provides a multiple run by run process control method. The multiple run by run (MRBR) process control method of the present invention can be used for any process in semiconductor manufacturing. The photolithography process will be used to demonstrate the MRBR scheme of the present invention.
The quality characteristic of the photolithography process is critical dimension (CD), shown in FIG. 2. According to physical knowledge, CD is a function of the input recipe and the exposure dose (E). The increment or decrement of the exposure dose will lead to a proportional change in CD. That is,
delta CD=β×delta E (1)
where β is the slope. Based on this relationship, the MRBR controller is to adjust the exposure dose setting such that CD will meet a specific target.
The exposure dose depends strongly on the combination of products and tools. Thus, we try to express each exposure dose by the product effect and the tool effect. Let E(i,j) denote the exposure does for the i-th product by the j-th tool. Then,
E(i,j)=P(i)×T(j)+C(i,j) (2)
where P(i) denotes the effect of the i-th product, T(j) denotes the effect of the j-th tool and C(i,j) denotes the correction for the exposure dose which can be regarded as the error term of equation (2).
The MRBR control scheme of the present invention consists of two phases, as illustrated in FIG. 3. The first phase uses information from past data to estimate P(i) and T(j) for i=1, . . . , m and j=1, . . . , n (31). This phase takes place off-line. The second phase takes place on-line and uses this prediction model to implement a hierarchical feedback control scheme. Phase II includes adjusting the recipe by an Exponential Weighted Moving Average (EWMA) controller 34 and detecting the changes on tool effect and product effect using integrated statistical process control 33.
log E(i,j)=log P(i)+log T(j)+ε(i,j) (3)
where ε(i,j) is the energy setting for product i in tool j.
In step 44, the past recipe table is to be updated by applying equation (2) for the known cells (step 43). In practice, some of the E(i,j) terms are not available.
For this kind of missing data, an expectation maximization (EM) algorithm can be used to estimate the exposure doses for all empty cells. For example, let P(i) and T(j) denote the estimates of P(i) and T(j) for all i=1, . . . , m and j=1, . . . , n, respectively. The two-way analysis of variance (ANOVA) method is used to estimate these effects. Column means and row means are calculated using the following equations:
where k is the iteration number and M is the stop condition.
Next, the recipes for the missing cells are calculated as follows:
E(I,J)[k+1]=M×(N×E(.,J)[k]−E(I,J)[k])+N×(M×E(I,)[k]−E(I,J)[k])−(N×M×E(.,.)−E(I,J)[k])/((N−1)(M−1)) (6)
The stop condition is estimated by the following:
M=Σ(E(I,J)[k+1]−E(I,J)[k])2 (7)
The error term can be estimated by
C(i,j)=E(i,j)−P(i)×T(j) (8)
for all i=1, . . . , m and j=1, . . . , n.
Phase II of the MRBR control process has two levels. Level 1, illustrated in
Et−1(i,j)=P(i)×T(j)+Ct−1(i,j) (9)
Assume that CDt(i,j) denotes the critical dimension at the end of the t-th run. For the (t−1)-th run, Ct(i,j) is updated by the following EWMA controller, as shown in FIG. 2:
Ct(i,j)=Ct−1(i,j)+w(i,j)×β(i)(CDt(i,j)−τ(i)) (10)
where w(i,j) is the discount factor of the EWMA controller and the new exposure dose setting is obtained by:
Et(i,j)=P(i)×T(j)+Ct(i,j) (11)
Note that C0(i,j) is given by Phase I.
Level 2, detecting aging of a tool, is illustrated in FIG. 7. In practice, there are two major situations that will cause changes in exposure dose: 1) tool (machine) aging and 2) incoming variation shift of product. If the j-th tool is aging, then the effect of T(j) is getting larger and larger. According to equation (2), we will find that the exposure dose setting is gradually increasing for most products. To overcome this difficulty, we use the following approach. Let Q(j) be the indicator of the aging of the j-th tool, It is adjusted run by run for the same tool. For instance, the current product i is the product by tool j at the t-th run. After it is done, the new exposure dose setting Et(i,j) is obtained by equation (11) of level 1, then
Qnew(j)=Qold(j)+log(Et(i,j)/Et−1(i,j)) (12)
regardless of what product is being run now. As shown in
The MRBR process of the present invention has been implemented.
The multiple run by run process control system of the present invention reduces the process parameters from m×n to m+n. An adequate recipe is provided for the initial run. The exposure dose, or other recipe, is adjusted for each product by tools run to run using a varied weighted EWMA controller, even if the product was not run recently, and integrating information about tool aging.
While the invention has been particularly shown and described with reference to the preferred embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made without departing from the spirit and scope of the invention.
U.S. patent application Ser. No. 10/749,698, to the same inventors, filed on Dec. 31, 2003.
| Number | Name | Date | Kind |
|---|---|---|---|
| 5943550 | Fulford, Jr. et al. | Aug 1999 | A |
| 6148239 | Funk et al. | Nov 2000 | A |
| 6245581 | Bonser et al. | Jun 2001 | B1 |
| 6248602 | Bode et al. | Jun 2001 | B1 |
| 6303395 | Nulman | Oct 2001 | B1 |
| 6405144 | Toprac et al. | Jun 2002 | B1 |
| Number | Date | Country | |
|---|---|---|---|
| 20040110310 A1 | Jun 2004 | US |
