Improved epitaxy process control in semiconductor manufacturing

Abstract

A method (60) for process control of epitaxy for semiconductor device fabrication is proposed. The method (60) comprises the following steps: obtaining (S1) measurements (15a-15c) of a semiconductor device processed according to a predefined parameterization of the epitaxy process (P), and determining (S2) an adapted parameterization of the epitaxy process for the subsequent semiconductor device using an R2R controller (13) based on the obtained measurements and a process model. The process model is configured to predict a first controlled variable (CV), a layer thickness (THK), as the first manipulated variable (MV), which is a time (TIME), and is additionally configured to predict a second controlled variable, a specific resistance (RES), based on a second manipulated variable, which is a gas flow (GAS).

Description

[0001] The invention relates to a control method of a run-to-run controller (R2R controller) for epitaxy, which executes a MIMO-R2R (multiple input multiple output) control, as well as a device, a computer program and a machine-readable storage medium which are configured to carry out the method. State of the art

[0002] It is known that R2R controllers receive measurements from processed wafers. These measurements are usually physical measurements on the wafer, but can also be virtual measurements. Physical measurements can also be referred to as product parameters. These are, for example, inline measurements that can be measured on the processed product, preferably using physical measuring devices. In contrast, there are process parameters (processing time, polishing pressure, exposure power, gas flow) that describe a manufacturing step (recipe) on the wafer.

[0003] Examples of measurement methods are: layer thickness measurement (THK), e.g. spectroscopic ellipsometry of several superimposed transparent layers, structure measurement using atomic force microscopy (AFM), e.g. trench depth, imaging techniques (CDSEM), e.g. geometric width, hole diameter, critical dimension, number and pattern of defects, (DFU), resistance measurement (RES), specific resistance.

[0004] Based on the measurements, a recipe for the processing step of at least one subsequent wafer is then proposed using the R2R controller in order to possibly readjust the processing step accordingly if the measurements were not within a predefined ideal range.

[0005] The processing step is controlled by batch control; that is, the recipe for the subsequent processing step for a wafer or batch (batch of wafers) is influenced by the R2R controller. A recipe characterizes the respective manufacturing or processing step, such as exposure, etching, deposition, or polishing. However, the R2R controller does not intervene during the processing of the wafer / batch. This type of control is also referred to as R2R process control. Run-to-run controllers for semiconductor manufacturing are well known; see, for example, Moyne, J. (2014). Run-to-Run Control in Semiconductor Manufacturing. In: Baillieul, J., Samad, T. (eds) Encyclopedia of Systems and Control. Springer, London. https: / / doi.org / 10.1007 / 978-1-4471-5102-9 255-1.

[0006] The R2R controller essentially serves to compensate for disturbances such as long-term drifting in production by making appropriate recipe adjustments in subsequent processing in order to keep relevant inline parameters at the desired target value.

[0007] Furthermore, it is known that R2R controllers can be used for process control in epitaxy. The R2R controller serves to control the product layer thickness, referred to as the controlled variable (CV) of the R2R controller, based on the adjustable processing time, referred to as the manipulated variable (MV) of the R2R controller. This means that known R2R controllers for epitaxy represent SISO R2R control (single input single output). This means that other recipe parameters, such as gas flows and lamp powers, are predefined in the recipe.

[0008] However, continuous and / or random disturbances, such as the aging of the EPL lamps over time, can cause not only the epitaxial process results such as layer thickness, but also other target parameters such as resistivity and the number of slip lines to drift, necessitating complex manual recipe corrections. Each of these properties has previously been adjusted using individual test wafers in time-consuming manual test procedures. In addition to the manual handling effort, this results in significant, costly test wafer consumption. For example, a lamp power can be manually adjusted in a few discrete steps, typically increasing monotonically over the life cycle without back-adjustment.

[0009] Therefore, an object of the invention is to provide a continuous, automatic control of these cost-intensive consumables with optimal utilization and minimal wear. Advantages of the invention

[0010] One advantage of the invention is the use of a common test wafer that undergoes all process and measurement steps, on which all three loops can be controlled together. This allows cost savings through a reduced number of wafers, as well as time savings through reduced test procedure effort. Furthermore, optimal utilization of the lamp power (minimal wear) can be achieved over the lamp's life cycle. Furthermore, the invention allows continuous tuning, for example, of the lamp power (only as much additional power as required at a minimum, discrete steps are used manually) and thus more precise control. Furthermore, the invention can perform automatic back-correction (power reduction, not provided for in the manual procedure). Disclosure of the invention

[0011] In a first aspect, the invention relates to a method for process control of an epitaxy for semiconductor component production. The process control is carried out by means of an R2R controller, which regulates process parameters of the epitaxy process so that wafers processed with these process parameters, in particular semiconductor components on this wafer, remain within a predetermined specification of the semiconductor components.

[0012] The method begins with obtaining measurements on one or a plurality of processed semiconductor components according to a predetermined process parameterization of the epitaxy process (P). The process parameterization can have been determined by the R2R controller in a previous step or specified in another way. The measurements each characterize a property of the wafer / semiconductor component that is influenced by the epitaxy or that is changed by the epitaxy. The measurements preferably include at least the controlled variables of the process model mentioned below.

[0013] This is followed by determining an adapted parameterization of the epitaxy process (P) for at least one, in particular immediately subsequent, semiconductor component by means of the R2R controller. The R2R controller determines the parameterization depending on the measurements obtained and a process model (P̂). The process model is set up to predict a first controlled variable (CV), a layer thickness (THK), depending on a first manipulated variable, which is a time. The time can be a process time of the epitaxy, i.e., as long as the wafer is under predefined process conditions (e.g., temperature and gas flow). In addition, the process model is set up to predict a second controlled variable, a specific resistance (RES), depending on a second manipulated variable, which is a gas flow (GAS). The manipulated variables are parameters of the epitaxy process that can be regulated by the R2R controller.After completing the epitaxy process, the controlled variables should meet the specifications of the semiconductor device. Note that the process model is configured to either predict the first controlled variable based on the first manipulated variable independently of the second manipulated variable, with the same analogous effect on the second controlled variable, or to predict both controlled variables based on both manipulated variables. It has been found that there is no significant cross-interaction between the two predictions, and they can therefore be modeled by independent causal relationships in the process model.

[0014] The proposed method extends the process model of known R2R controllers from SISO to MIMO, whereby a particularly precise control can be achieved precisely through the specific combination of the mentioned manipulated variables.

[0015] It is proposed that the second manipulated variable comprise a first and a second gas flow (GAS1, GAS2). The first gas flow can be a gas flow of, for example, 400 sccm (standard cubic centimeters) (MDOPANT) and the second gas flow can be a gas flow of, for example, 20 sccm (standard cubic centimeters) (MDOPANTAUX).

[0016] It is further proposed that the third manipulated variable comprises a first and a second lamp power (POWER1,POWER2), wherein the first lamp power (POWER1) can be a relative value (TOPPOWER) that relates to an absolute power, e.g. 70.8%, wherein the second lamp power (PWOER2) is a relative value (BOTPOWER), e.g. 17.2%.

[0017] Furthermore, it is proposed that the R2R controller be given a constraint regarding the third controlled variable and considered accordingly when solving its internal optimization problem. This constraint characterizes a maximum number of defects that may occur after executing the process according to the R2R control. The maximum number of defects is preferably greater than zero. Defects can be understood as any anomalies caused by the epitaxy process.

[0018] The advantage here is that the R2R controller can increase or decrease the lamp power based on this constraint. This constraint gives the R2R controller a new degree of freedom, allowing it to freely adjust the lamp power.

[0019] For example, the lamp power can be set to 60% at the beginning and slowly increased to a maximum power, e.g. 95% or 100%, via the R2R controller during the production of the semiconductor components in order to comply with this constraint, whereby a reduction of the lamp power can occur in between by the controller.

[0020] It is further proposed that the number of defects be a number of sliplines. The aforementioned constraint in this embodiment can be a maximum number of sliplines (TAR), e.g., TAR <= 1 or another value that is preferably smaller than the maximum number according to a specification for the semiconductor device.

[0021] Further advantageous embodiments of the first aspect of the invention are the subject of the dependent claims.

[0022] In further aspects, the invention relates to a device and a computer program, each of which is configured to carry out the above methods, and to a machine-readable storage medium on which this computer program is stored.

[0023] Embodiments of the invention are explained in more detail below with reference to the accompanying drawings. In the drawings: Fig. 1 schematically shows a known R2R controller; Fig. 2 schematically shows an R2R controller with integrated process model; Fig. 3 schematic process models for an R2R control of an epitaxy; Fig. 4 schematically shows a heuristic trim curve; Fig. 5 schematically shows an R2R control scheme 5x3; and Fig. 6 schematically shows a flow diagram of an embodiment of a control method with R2R control scheme 5x3. Description of the embodiments

[0024] Fig. 1 schematically shows an information flow representation 10 of an R2R controller 13, which is known from the prior art.

[0025] The information flow diagram 10 shows a process flow 11 for manufacturing a semiconductor device. In a first step, one or more physical measurements 12 are performed on a wafer. A measurement 12 can be a physical measurement on the wafer. The measurement results MET are then forwarded to the R2R controller 13 and provided.

[0026] Depending on the measurement results MET, the R2R controller 13 can set a process step 14 for the wafer. In other words, the R2R controller 13 outputs a setting for the process step 14 depending on the measurement results MET, so that after executing the process step 14, the processed wafer exhibits product parameters as close as possible to a specified value range.

[0027] Optionally, after executing process step 14, one or more subsequent physical measurements 15 can be performed on the processed wafer according to the settings output by the R2R controller 13. The resulting post-MET measurement results can be provided to the R2R controller 13 as feedback, allowing it to make better settings for subsequent process steps 14 based on the subsequently obtained measurement results.

[0028] In summary, it can be said that the R2R process control according to Fig. 1 includes a pre-measurement, processing, and optionally a post-measurement. The R2R Controller 13 pursues one control objective: to maintain process parameters (controlled variables) within specification limits at the target value and automatically correct target deviations in a model-based, self-learning manner through targeted adjustment of manipulated variables, e.g., processing time, gas flows, etc.

[0029] The control law itself is typically a mathematical optimization problem with an objective function, possibly with constraints, using stored state variables and a system-theoretically based feedback principle (closed loop control).

[0030] R2R process control is preferably carried out with an integrated process model. Fig. Figure 2 shows an example of an R2R process control loop (20) with the aforementioned integrated process model P̂. For the process control, a controlled variable REF is specified. This is provided to the R2R controller R1, which determines a manipulated variable MV. A manufacturing process P is then executed according to the manipulated variable MV. During the execution of the manufacturing process and / or during subsequent measurement, a disturbance may occur, which in Fig. 2 is represented as the disturbance variable DIST. After a product has been processed according to the manufacturing process, it can be measured whether the variable CV to be controlled lies within a target range or specification range.

[0031] Parallel to the manufacturing process P, the process model P̂ predicts the controlled variable CV based on the manipulated variable MV. The process model P̂ is preferably a mathematical model that models the relationships between the manipulated variable MV and the controlled variable CV. Preferably, the process model P was created or parameterized based on experimental data. It is conceivable that the process model P is a simple model that reproduces phenomena, particularly the basic behavior of the process. Alternatively, the process model P can be a learned mathematical model, e.g., a Gaussian process or another machine learning system.

[0032] In the first information node 21, a difference between the prediction of the controlled variable CV by the process model P̂ and the (measured) controlled variable CV is determined. This difference can optionally be processed in a second controller R2. The second controller R2 can be a filter that performs smoothing and / or filters out noise. This advantageously minimizes the disturbance variable DIST in the signal, which is, for example, measurement noise. The second controller R2 can perform a moving average or exponentially smoothed weighting.

[0033] In the second information node 22, a difference between the reference variable REF and the signal from the first information node 21 or the processed signal of the second controller R2 is determined.

[0034] Depending on this difference, the first controller R1 will regulate the manipulated variable MV. For example, if this difference is 0, the controller is inactive; if the difference is a different value, the controller is active. Preferably, the first controller R1 is a dead-beat controller.

[0035] The process model P̂ can be a representation of the process P, often a phenomenological input-output model, which does not necessarily have to be globally valid, but describes a behavior of the process P at the operating point.

[0036] In one embodiment, thickness and specific resistance can be measured as controlled variables or product parameters after processing by process P. The first controller R1 pursues a control objective: to maintain process parameters (controlled variables CV) within specification limits at the target value. In this case, the first controller automatically, model-based, corrects target deviations based on the difference of the second information node 22 by specifically adjusting manipulated variables MV, e.g., processing time, gas flows, etc.

[0037] A first embodiment (30) of the process model P̂ in Fig. Figure 3 shows a known process model for epitaxial control. The process model P̂ according to the first embodiment (30) predicts a layer thickness (THK) as a function of time (TIME).

[0038] In the following, an extension of the previous SISO-R2R control according to the first embodiment (30) will be carried out in order to achieve MIMO-R2R (multiple input multiple output) controls.

[0039] In a second embodiment (31) of the process model, a novel, first EPI control variant is proposed, which is a 2x2-MIMO process model P̂. The 2x2-MIMO process model P̂ of the second embodiment (31) differs from the first embodiment (30) in that a gas flow GAS is additionally used as the manipulated variable MV and a specific resistance RES is used as an additional controlled variable CV. Accordingly, the time control (TIME) based on layer thickness (THK) is extended by an additional loop of gas flow (GAS) based on specific resistance (RES).

[0040] In general, the loops can be modeled independently or dependently in the process model.

[0041] In a third embodiment (32) of the process model, a novel, second EPI control variant is proposed, which is a 3x2-MIMO process model P̂. The 3x2-MIMO process model P̂ of the third embodiment (32) differs from the 2x2-MIMO process model P̂ of the second embodiment (31) in that a further gas flow GAS2 is used as an additional manipulated variable MV and that a dynamically calculated constraint is specified for the first controller R1, which defines a relationship between the gas flow GAS1 and the further gas flow G2: a * GAS1 + b * GAS2 = c, where the parameters a, b, c are real numbers. The parameters a, b, c can be derived empirically in advance from recipe settings and the corresponding measurements (DOE series, Design Of Experiments, experimental planning).The dynamically calculated constraint allows the further gas flow parameter (GAS2) to be set depending on the first gas flow parameter (GAS1) according to a * GAS1 + b * GAS2 = c.

[0042] It should be noted that the R2R controller can receive constraints and, when controlling, performs an optimization preferably with respect to a quadratic cost function or quality criterion taking into account one or more constraints.

[0043] Preferably, a control loop for the gas flows GAS1 and GAS2 is defined or modeled for the specific resistance RES. This allows stabilization of the specific resistance, whereby control can be achieved by the gas flow parameter GAS1 (e.g., MDOPANT) or an additional linearly dependent gas flow parameter GAS2 (e.g., MDOPANTAUX) as an adjustable recipe parameter.

[0044] In a fourth embodiment (33) of the process model, a novel, third EPI control variant is proposed, which is a 3×3-MIMO process model P̂. The 3×3-MIMO process model P̂ of the fourth embodiment (32) differs from the 2×2-MIMO process model P̂ of the second embodiment (31) in that, in addition to the control loop

[0045] Control variable MV of the gas flow GAS to the controlled variable CV of the specific resistance RES, an additional control loop is defined, which defines as control variable MV a lamp power POWER to the controlled variable CV of a number of sliplines DFU.

[0046] Preferably, a constraint is set for the first controller R1 for the further control loop, which adjusts the lamp power POWER to the number of sliplines DFU. To minimize the number of sliplines, continuous control of the lamp power (POWER) is used. The constraint can then be a maximum number of sliplines, e.g., TAR < 1.

[0047] In a fifth embodiment (34) of the process model, a novel, fourth EPI control variant is proposed, which is a 5×3 MIMO process model P̂. The 5×3 MIMO process model P̂ of the fifth embodiment (32) differs from the 3×2 MIMO process model of the third embodiment (32) in that the manipulated variable MV, a lamp power POWER, is given by two variables POWER1 and POWER2 and is defined as a control loop for the number of slip lines DFU. For POWER1 / POWER2, ​​TOPPOWER / BOTPOWER, for example, is used as an adjustable recipe parameter.

[0048] Furthermore, as in the fourth embodiment (33), the constraint of a maximum number (TAR) of sliplines DFU is used for the first controller R1. Additionally, a dynamically calculated constraint d * POWER1 + e * POWER2 = f, where parameters d, e, f are real numbers, can be used. The parameters d, e, f are typically specified as system settings, e.g., d=1, e=-4, and f=0.5.

[0049] In Fig. Figure 4 shows a heuristic adjustment curve for the POWER / slipline relationship. The manipulated variable MV is plotted on the x-axis, in this case the lamp power, for example. The controlled variable CV, in this case the number of sliplines, is plotted on the y-axis. TAR is the target value of the controlled variable, in this case the maximum number of sliplines. TAR is typically 1, but other values ​​> 0 are also conceivable. The advantage of using TAR < 1 as a constraint is that it enables backward correction of the controller. This means that the controller can increase or, advantageously, reduce the lamp power.

[0050] For the relationship between manipulated variable MV (e.g. POWER1 / POWER2 =POWERRATIO) and controlled variable CV (number of slip lines), a straight line 40 with a negative increase, ie MV increase reduces CV, has been found.

[0051] It is conceivable that the manipulated variable can only be set within a predefined range: Mvmin < MV < MVmax.

[0052] Fig. 5 schematically shows an R2R control scheme 5x3 according to the fifth embodiment 34 of the process model.

[0053] The R2R Controller 13 differs from Fig. 1 in that it now receives more measurements as input and can control a plurality of manipulated variables MV based on these measurements.

[0054] The process flow in Fig. 5 differs from the process flow of Fig. 1 is that the POST-MET step 15 comprises a number of different measurements. The controlled variables CV are recorded in the individual measurements 15a-15c.

[0055] Measurements 15a-15c can be carried out as in Fig. 5 are performed one after the other. Alternatively, parallel measurements are possible.

[0056] Fig.6 schematically shows a flowchart 40 of an embodiment of a control method of the R2R controller 13 based on the above process models 30-34.

[0057] The method begins with providing or receiving (S1) measurements (15a-15c) of a semiconductor component processed according to a predetermined parameterization of the epitaxy process (P).

[0058] In the subsequent step S2, an adapted parameterization of the epitaxial process (14) for the subsequent semiconductor component is determined (S2) using an R2R controller (13) depending on the obtained measurements and a process model (P̂). The process model is one of the process models 30-34.

[0059] In the subsequent step S3, the process is controlled according to the parameterization from step S2. Optionally, measurements 15 can be performed in step S3.

[0060] Optionally, steps S1-S3 may be performed again by providing the measurements from step S3 to step S1. QUOTES CONTAINED IN THE DESCRIPTION

[0000] This list of documents submitted by the applicant was generated automatically and is included solely for the convenience of the reader. This list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions. Cited non-patent literature

[0000] Moyne, J. (2014). Run-to-Run Control in Semiconductor Manufacturing. In: Baillieul, J., Samad, T. (eds) Encyclopedia of Systems and Control. Springer, London. https: / / doi.org / 10.1007 / 978-1-4471-5102-9 255-1

[0005]

Claims

[1] Method (60) for process control of an epitaxy for semiconductor device production comprising the following steps: Obtaining (S1) measurements (15a-15c) of a semiconductor component processed according to a predetermined parameterization of the epitaxial process (P); Determining (S2) an adapted parameterization of the epitaxy process (P,14) for a subsequent semiconductor component by means of an R2R controller (13) depending on the measurements obtained and a process model (P̂), wherein the process model is configured to predict a first controlled variable (CV), which is a layer thickness (THK), depending on a first manipulated variable (MV), which is a time (TIME), characterized by , that the process model is additionally configured to predict a second controlled variable of a specific resistance (RES) depending on a second manipulated variable, which is a gas flow (GAS). [2] Method according to claim 1, characterized bythat the second manipulated variable comprises a first and a second gas flow (GAS1,GAS2), wherein the R2R controller (13) takes into account a first secondary condition which characterizes a ratio of the first gas flow to the second gas flow. [3] Method according to one of the preceding claims, characterized by that the process model is additionally set up to predict a third controlled variable of a number of defects (DFU) depending on a third manipulated variable, which is a lamp power (POWER). [4] Method according to claim 3, characterized byin that the third manipulated variable comprises a first and a second lamp power (POWER1,POWER2), wherein the first lamp power (POWER1) is a relative first power (TOPPOWER), wherein the second lamp power (PWOER2) is a second relative power (BOTPOWER), wherein the R2R controller (13) takes into account a second secondary condition which characterizes a ratio of the first lamp power to the second lamp power. [5] Method according to claim 3 or 4, characterized by that the R2R controller (13) receives a third constraint with respect to the third controlled variable, wherein the third constraint characterizes a maximum number of defects that may occur after execution of the process. [6] Method according to one of the preceding claims, characterized by that the process model (P̂) is a parameterizable model, whereby the parameterization is predetermined, in particular determined on the basis of experiments. [7] Device which is arranged to carry out the method according to one of the preceding claims. [8] A computer program comprising instructions which, when executed by a computer, cause the computer to carry out the method according to any one of claims 1 to 6. [9] A machine-readable storage medium on which the computer program according to claim 8 is stored.

Images