Method for monitoring amount of discharge of processing liquid, and substrate processing apparatus

Abstract

[Problem] To automatically and accurately monitor flow rate without requiring a highly skilled operator. [Solution] A monitoring method for monitoring discharge of processing liquid from a nozzle in a processing liquid supply mechanism which includes a nozzle, a processing liquid line through which the processing liquid supplied from a processing liquid supply source flows to the nozzle, a flow rate sensor and a flow rate control valve interposed in the processing liquid line, and a flow rate controller for controlling the degree of opening of the flow rate control valve so that a flow rate of the processing liquid detected by the flow rate sensor becomes a target flow rate, the method comprising: a step for sampling flow rate data which is an output of the flow rate sensor; a step for fitting the sampled flow rate data to a mathematical expression representing a transient response curve; and a step for determining at least one of the presence or absence of abnormality in the discharge of the processing liquid from the nozzle, the type of abnormality in the discharge, and the presence or absence of a sign of occurrence of abnormality in the discharge on the basis of at least a parameter value in the mathematical expression to which the fitting has been performed.

Description

METHOD FOR MONITORING DISCHARGE AMOUNT OF PROCESSING LIQUID AND SUBSTRATE PROCESSING APPARATUS

[0001] The present disclosure relates to a method for monitoring a discharge rate of a processing liquid and a substrate processing apparatus.

[0002] The manufacturing process of a semiconductor device includes a liquid processing step in which a processing liquid is supplied to a substrate such as a semiconductor wafer to perform liquid processing on the substrate. In the liquid processing step, the processing liquid is supplied from a nozzle to the surface of the substrate, which is held and rotated by a substrate holding and rotating mechanism such as a spin chuck. Patent Document 1 describes monitoring log data output from a flow rate detection sensor to confirm that the processing liquid is supplied to the substrate at an appropriate flow rate defined in the processing recipe.

[0003] JP 2015-146069 A

[0004] The present disclosure provides a technique for automatically and accurately monitoring flow rates without requiring a highly skilled operator.

[0005] According to one embodiment of the present disclosure, there is provided a monitoring method for monitoring the ejection of processing liquid from a nozzle in a processing liquid supply mechanism comprising a nozzle that ejects processing liquid, a processing liquid line through which processing liquid supplied from a processing liquid supply source flows to the nozzle, a flow sensor and a flow control valve interposed in the processing liquid line, and a flow controller that controls the opening of the flow control valve so that the flow rate of the processing liquid detected by the flow sensor becomes a target flow rate, the monitoring method comprising: a sampling step of sampling flow rate data that is the output of the flow sensor; a fitting step of fitting the sampled flow rate data to a mathematical equation representing a transient response curve; and a determination step of determining at least one of whether or not there is an abnormality in the ejection of processing liquid from the nozzle, the type of the ejection abnormality, and whether or not there are signs of the occurrence of the ejection abnormality, based on at least parameter values ​​in the fitted mathematical equation.

[0006] According to the above embodiment, it is possible to automatically and accurately monitor the flow rate without requiring a highly skilled operator.

[0007] 1 is a schematic cross-sectional view of a substrate processing system according to an embodiment of the substrate processing apparatus; FIG. 2 is a piping system diagram showing a main part of an example of the configuration of a processing liquid supply mechanism used in the substrate processing system; FIG. 3 is a graph showing a transition of a flow rate from the start of discharge of a processing liquid (line 1) and a transient response curve after fitting (line 2); FIG. 4 is a graph showing a transient response curve of a step response in a second-order lag system; FIG. 5 is a graph showing how a damping coefficient ζ changes with each increase in the number of discharges; FIG. 6 is a graph explaining an application example of principal component analysis; FIG. 7 is a graph explaining an application example of principal component analysis; FIG. 8 is a graph explaining an application example of principal component analysis; FIG. 9 is a graph explaining fitting using FFT and a Lorentz function for a first example (DIW discharge); FIG. 10 is a graph explaining fitting using FFT and a Lorentz function for a second example (IPA discharge); FIG. 11 is a graph explaining fitting using FFT and a Lorentz function for the second example; FIG. 12 is a graph explaining an example of an abnormal pattern; FIG. 13 is a graph explaining an example of an abnormal pattern;

[0008] Hereinafter, an embodiment will be described with reference to the accompanying drawings.

[0009] 1 is a diagram showing a schematic configuration of a substrate processing system according to this embodiment. In the following, to clarify the positional relationship, mutually orthogonal X-axis, Y-axis, and Z-axis are defined, and the positive direction of the Z-axis is defined as the vertically upward direction.

[0010] 1, the substrate processing system 1 includes a loading / unloading station 2 and a processing station 3. The loading / unloading station 2 and the processing station 3 are provided adjacent to each other.

[0011] The loading / unloading station 2 includes a carrier placement section 11 and a transport section 12. A plurality of carriers C are placed on the carrier placement section 11, each of which accommodates a plurality of substrates, in this embodiment semiconductor wafers (hereinafter referred to as wafers W), in a horizontal position.

[0012] The transfer section 12 is provided adjacent to the carrier placement section 11 and includes a substrate transfer device 13 and a transfer section 14. The substrate transfer device 13 includes a wafer holding mechanism that holds the wafer W. The substrate transfer device 13 is capable of moving horizontally and vertically and rotating about a vertical axis, and transfers the wafer W between the carrier C and the transfer section 14 using the wafer holding mechanism.

[0013] The processing station 3 is provided adjacent to the transport section 12. The processing station 3 includes a transport section 15 and a plurality of liquid processing units 16. The plurality of liquid processing units 16 are provided side by side on both sides of the transport section 15.

[0014] Transfer section 15 includes therein substrate transfer device 17. Substrate transfer device 17 includes a wafer holding mechanism that holds wafer W. Substrate transfer device 17 is capable of moving horizontally and vertically and rotating about a vertical axis, and transfers wafer W between delivery section 14 and liquid processing unit 16 using the wafer holding mechanism.

[0015] The liquid processing unit 16 performs a predetermined substrate processing on the wafer W transferred by the substrate transfer device 17 .

[0016] The substrate processing system 1 also includes a control device 4. The control device 4 is, for example, a computer, and includes a control calculation unit 18 and a storage unit 19. The storage unit 19 stores programs that control various processes executed in the substrate processing system 1. The control calculation unit 18 controls the operation of the substrate processing system 1 by reading and executing the programs stored in the storage unit 19.

[0017] The program may be recorded on a computer-readable storage medium and installed from the storage medium into the storage unit 19 of the control device 4. Examples of computer-readable storage media include a hard disk (HD), a flexible disk (FD), a compact disk (CD), a magnetic optical disk (MO), and a memory card.

[0018] In substrate processing system 1 configured as described above, first, substrate transfer device 13 in load / unload station 2 removes wafer W from carrier C placed on carrier placement unit 11 and places the removed wafer W on transfer unit 14. Wafer W placed on transfer unit 14 is then removed from transfer unit 14 by substrate transfer device 17 in processing station 3 and transferred into liquid processing unit 16.

[0019] The wafer W carried into the liquid processing unit 16 is subjected to liquid processing by the liquid processing unit 16, and then carried out of the liquid processing unit 16 by the substrate transfer device 17 and placed on the delivery section 14. Then, the processed wafer W placed on the delivery section 14 is returned to the carrier C on the carrier placement section 11 by the substrate transfer device 13.

[0020] Next, the configuration of the processing liquid supply mechanism 30 that supplies the processing liquid to the liquid processing unit 16 will be described with reference to FIG.

[0021] The processing liquid supply mechanism 30 has a processing liquid circulation system 31. The processing liquid circulation system 31 has a tank 32 that stores the processing liquid and a circulation line 33 connected to the tank 32. The circulation line 33 is provided with devices such as a pump 34, a temperature regulator such as a heater 35, and a filter 35F. Driving the pump 34 creates a circulation flow of the processing liquid, which is pumped from the tank 32 to the circulation line 33 and returns to the tank through the circulation line 33. A connection area 37 of the circulation line 33 is connected to a plurality of supply lines (processing liquid lines) 36. Each supply line 36 supplies the processing liquid to a corresponding one of the liquid processing units 16. A back-pressure valve 35B is provided in the circulation line 33 downstream of the connection area 37, so that the liquid pressure of the processing liquid in the connection area 37 of the circulation line 33 is maintained at a substantially constant level.

[0022] 2 shows one supply line 36 and one corresponding liquid treatment unit 16. The other supply lines 36 and other liquid treatment units 16 have the same configuration.

[0023] A nozzle 38 is provided at the downstream end of supply line 36. Nozzle 38 is supported by a nozzle arm (not shown). Wafer (substrate) W is held in a horizontal position and rotated about a vertical axis by a substrate holding and rotating mechanism called a spin chuck or the like installed in a chamber (not shown) of liquid processing unit 16. A processing liquid is ejected from nozzle 38 onto the surface of rotating wafer W, thereby performing at least one liquid processing step on wafer W. Liquid processing unit 16 may be one well known in the technical field of semiconductor manufacturing equipment, and detailed configuration illustration and description thereof will be omitted.

[0024] A liquid flow controller 40 is provided in the supply line 36. The liquid flow controller 40 includes a flow sensor 42, a flow control valve 44, and a control circuit (flow controller) 46. The liquid flow controller 40 controls the aperture of the flow control valve 44 based on a flow command signal received from the control unit 4 (see also FIG. 1 ), thereby controlling the discharge flow rate of the processing liquid discharged from the nozzle 38 onto the wafer W. At this time, the control circuit 46 controls the aperture of the flow control valve 44 so as to achieve the commanded flow rate based on the deviation between the commanded flow rate defined by the flow command signal and the measured flow rate detected by the flow sensor 42. The control circuit 46 outputs the detected flow rate received from the flow sensor 42 to the control unit 4 as log data (data linking the detected flow rate with time). The control unit 4 may link the detected flow rate with time. The control unit 4 monitors the discharge of the processing liquid discharged from the nozzle 38 based on the log data, according to a procedure described in detail below.

[0025] In addition to the components shown in the figure, the supply line 36 may be provided with a drain line equipped with an on-off valve for a suck-back mechanism to prevent liquid from accumulating near the nozzle of the supply line 36. The supply line 36 may also be provided with an additional on-off valve. Furthermore, in order to constantly maintain a high temperature of the liquid in the supply line 36, a return circulation line equipped with an on-off valve may be provided, branching from the upstream side of the liquid flow controller 40 and connected to the circulation line 33. The above-mentioned optional additional configurations are not shown in FIG. 2 for the sake of simplicity.

[0026] The processing liquid supply mechanism 30 does not need to have the processing liquid circulation system 31, and the supply line 36 can be connected to a processing liquid storage tank or a processing liquid supply source provided as a factory utility. The substrate processing system 1 can be configured to include processing liquid supply mechanisms 30 in a number corresponding to the number of types of processing liquid to be supplied to one wafer.

[0027] [Description of Monitoring Method] Next, a method for the control device 4 to monitor the discharge of the processing liquid based on the log data will be described. The calculations involved in this monitoring may be performed by the control device 4 of the substrate processing system 1, or alternatively, by a host computer provided in the semiconductor manufacturing factory where the substrate processing system 1 is installed, or by a host computer provided in a location remote from the semiconductor manufacturing factory. The control device 4, host computer, and host computer described here can be considered as control devices that implement the monitoring method.

[0028] <Detection (monitoring) of abnormalities during a transient period> The monitoring method can be broadly divided into: a sampling step of sampling detected flow rate data (flow rate data), which is the output of the flow sensor 42; a fitting step of fitting the sampled flow rate data to a mathematical formula representing a transient response curve; and a determination step of determining, based on at least parameter values ​​in the mathematical formula that has been fitted, at least one of whether or not there is an abnormality in the ejection of the processing liquid from the nozzle, the type of the ejection abnormality, and whether or not there are signs of the occurrence of the ejection abnormality.

[0029] In the sampling step, as described above, for example, the control circuit 46 (or the control unit 4) samples the output (detected flow rate) of the flow rate sensor 42 at a predetermined sampling frequency and outputs the sampled flow rate as log data in which the detected flow rate is linked to time. Based on the log data, the control unit 4 creates a two-dimensional graph, as shown in FIG. 3, with the horizontal axis representing time and the vertical axis representing flow rate.

[0030] When the control unit 4 creates a graph to provide visual information to the operator, the control unit 4 may display the created graph on a visual user interface such as a display. Displaying the graph on a display is not essential. When not displaying the graph, the control unit 4 may perform calculations equivalent to creating a graph without creating the graph.

[0031] Line 1 in Figure 3 is a line created by plotting a large number of log data corresponding to the sampling frequency on a graph, and shows the change over time in the actual flow rate detected by the flow sensor 42. Line 2 in Figure 3 is a transient response curve created by fitting a large number of log data that are the basis of line 1 to a mathematical formula. There are various known programs for fitting raw data to a mathematical formula, and an appropriate one can be selected and used from among them.

[0032] In this embodiment, the phenomena monitored are the rise in flow rate when a constant target flow rate (set value) is applied to the control system from a zero flow rate state, and the subsequent minute fluctuations in flow rate after the (apparent) stable state is reached. The rise in flow rate when a constant target flow rate is applied to the control system from a zero flow rate state can be represented using a transient response curve in the step response (unit step response).

[0033] Figure 4 shows the transient response curve of the step response in a second-order lag system. Here, ζ is the damping coefficient. ωn is the natural angular frequency, and t is time. Therefore, the horizontal axis can be considered the time axis. y(t), for example, is the flow rate. When ζ is changed, the results are as follows. If 0≦ζ<1, meaning the system has complex poles, it will oscillate, but if ζ=0, it will continue to oscillate forever, and the larger ζ is, the faster the oscillation converges to the target value. If 1≦ζ, meaning the system has real poles, it will not oscillate, and the larger ζ is, the slower the convergence to the target value.

[0034] In the liquid processing unit 16, when the processing liquid is discharged from the nozzle 38 onto the wafer W, the ideal flow rate is one that does not oscillate immediately after startup and quickly converges to the target value (ζ = 1 and ωn is relatively large). To achieve this as much as possible, control parameters such as feedback gain are optimized. In actual operation, some oscillation may be tolerated as long as excessive overshoot and undershoot do not occur. Specifically, the control parameters may be appropriately set to, for example, a damping coefficient ζ = 0.5 to 0.6. This facilitates management of the total flow rate of the processing liquid discharged onto the wafer, reduces the amplitude of oscillation in the flow rate of the processing liquid, and improves the stability of the flow rate (i.e., stabilizes the thickness of the processing liquid film formed on the wafer surface). It is also possible to achieve convergence to the target flow rate without substantially causing any overshoot (corresponding to a case where ζ is 1 or greater).

[0035] The equations expressing the step response in a second-order lag system are, for example, as follows, and these equations are well known.

[0036] The above three equations and the graph in Figure 4 are taken from the website https: / / www.flight.tu-tokyo.ac.jp / ~tsuchiya / Control / 14Response.pdf, but similar equations are found on various websites and in literature.

[0037] In this embodiment, the inventors conducted tests based on the above known mathematical formulas and further improved the formulas to find the following formulas (Formulas 1 to 3).

[0038] In Equations 1 to 3 above, -amp corresponds to the value in the braces {}, i.e., the coefficient used to vertically stretch the graph to match the actual flow rate. -ζ is the damping coefficient, which indicates the presence or absence of oscillation (alternating overshoot and undershoot) and the speed of convergence of the oscillation. By monitoring ζ, it is possible to monitor not only the speed of convergence of oscillation but also the presence or absence of oscillation. As shown in Figure 4, when ζ<1, oscillation occurs, but when ζ=1 or ζ>1, oscillation does not occur. When ζ>1, undershoot is maintained and the system slowly converges to the set value. -ωn is the natural angular frequency, and the larger ωn, the faster the rise. Furthermore, the larger ωn, the faster the convergence. By monitoring ωn, it is possible to monitor the rise speed (especially whether the initial rise is steep, i.e., whether it rises in a short time or whether the initial rise is gradual and takes a long time) and the speed of convergence of oscillation. - t0 is a coefficient for shifting the graph along the horizontal axis (time axis). When bubbles occur in the liquid, a delay occurs in the change in discharge flow rate. Although t0 ​​is not included in the general formula for step response described above, the inventors' research has shown that using t0 as a parameter representing the degree of bubble entrapment enables good fitting. t0 of zero indicates no bubble entrapment at all, and it is desirable for t0 to be as small as possible. By monitoring t0, it is possible to determine whether or not an unacceptable level of bubble entrapment has occurred. - y0 is a constant representing the initial value of the flow rate. Setting y0 has the advantage that the same fitting can be used to determine when the flow rate changes from a high value to a low value.

[0039] Fitting was performed for all three equations above (the equation when ζ<1, the equation when ζ=1, and the equation when ζ>1), and the equation with the best fit (coefficient of determination R ２is closest to 1) can be used. In actual operation, it may be known which of the three equations above provides the best fit when the control parameters are determined, so one or two equations to be used for fitting may be determined in advance. This can reduce the computational load.

[0040] As an example, when the raw data of the actual flow rate corresponding to line 1 in FIG. 3 was fitted to the above equation [Equation 2], the parameter values ​​in the equation were as follows: amp = 75.0 ζ = 0.55 ωn = 8.15 t0 = 0.15 y0 = 0.0 The coefficient of determination in this fitting, R ２ is 0.99 (the closer to 1 the better), which shows that the fitting is very good. The time course of the flow rate expressed by Equation 2 after fitting is shown by line 2 in FIG. 3, and it can be seen that it is very similar to line 1, which is the raw data.

[0041] By monitoring the various parameters described above, it is possible to monitor whether appropriate control is being performed during the time period immediately after the start of discharge. The monitoring can be performed, for example, by comparing each parameter with a predetermined reference value. For example, if the actual value of each parameter (the value calculated during fitting) falls below a reference value (for example, a lower threshold value), it can be determined that an abnormality has occurred.

[0042] For example, if the reference value (lower limit threshold) of ζ is 0.50, and the actual value is 0.30, it can be determined that an abnormality has occurred. In this case, a problem occurs in that relatively large overshoots and undershoots are repeated after the start of discharge, and it takes a long time for the flow rate to converge to the target value.

[0043] By continuously monitoring the above-mentioned various parameters during multiple ejections, it is possible to grasp the deterioration or change over time of the device. Specifically, for example, when ζ gradually decreases as shown in the graph of Figure 5, the amplitude of the overshoot and undershoot immediately after the start of ejection gradually increases with each ejection, and the time required for the flow rate to stabilize gradually increases. If the number of ejections (the horizontal axis N of the graph in Figure 5) increases further, the ζ value (the vertical axis of the graph in Figure 5) will fall below the reference value R (the lower threshold indicated by the dashed line in Figure 5), and it is expected that an unacceptable ejection abnormality will occur.

[0044] If a gradual decrease in ζ is observed, or if the difference between the ζ value and the reference value becomes small (for example, when the difference becomes about 0.10), a warning may be issued to the operator using the user interface (display or alarm sound). The operator may then change the control parameters (target flow rate, feedback gain). Alternatively, the control unit 4 may have a function to automatically change the control parameters.

[0045] Alternatively, monitoring can be performed via principal component analysis (PCA). Specifically, for example, the treatment liquid is repeatedly discharged from the nozzle at a predetermined target flow rate multiple times. Each time, data on the change in flow rate over time is sampled, at least during the period from the start of discharge until the discharge flow rate stabilizes at the target flow rate (transient period). The sampled raw data is fitted with a transient response curve as described above to calculate the above-mentioned parameters (included in the above-mentioned formula). These parameters can be used as explanatory variables in the principal component analysis. The coefficient of determination obtained during fitting can also be included in the explanatory variables. Furthermore, other parameters related to the discharge flow rate that are not included in the above-mentioned formula can also be included in the explanatory variables. Examples of such parameters include the discharge flow rate set value, the deviation between the discharge flow rate set value and the actual discharge flow rate t seconds after the start of discharge (e.g., t = 1.5 seconds, 2.0 seconds, and 3.0 seconds), the overshoot time (the time during which the actual discharge flow rate exceeds the discharge flow rate set value), and the overshoot rate ((maximum actual discharge flow rate during the overshoot time - discharge flow rate set value) / discharge flow rate set value). For each discharge, a set of explanatory variables composed of the various parameters described above is obtained. Principal component analysis is performed on the set of explanatory variables obtained from multiple discharges, and the first and second principal components, which are composite variables corresponding to each discharge, are calculated and plotted on a two-dimensional graph (two-dimensional map).

[0046] FIG. 6 is a two-dimensional graph plotting the first principal component (PCA Feature 1 on the horizontal axis) and the second principal component (PCA Feature 2 on the vertical axis) calculated based on data obtained when DIW (pure water) was ejected multiple times as a processing liquid from a nozzle at a target flow rate of 2000 ml / min.

[0047] 6, the plots are concentrated in a specific area A surrounded by a chain line, and one plot Pb is located at a distance D away from area A. It is estimated that the ejection behavior of the treatment liquid during the transition period in the ejection cycle corresponding to plot Pb is significantly different from that of the other ejection cycles (the ejection cycles corresponding to the plots in area A), for example, that the overshoot / undershoot behavior is significantly different.

[0048] The actual flow rate data corresponding to plot Pb in the graph of Figure 6 corresponds to curve Cb in the graph of Figure 7, which shows the change in flow rate over time. Curve Cb exhibits an extremely large overshoot, and it takes a long time to converge to the target flow rate. On the other hand, the actual flow rate data corresponding to the plot in area A of the graph of Figure 6 corresponds to the numerous curves Ca in the graph of Figure 7 (which are overlapping and therefore difficult to distinguish). Curve Ca shows a desirable flow rate transition. In other words, curve Ca does not exhibit a problematic large overshoot, and converges to the target flow rate in a short time.

[0049] By discharging DIW many times and accumulating data, it becomes possible to determine the presence or absence of an abnormality based on the plot position on a two-dimensional graph such as that shown in Figure 6, and it also becomes possible to determine the type of abnormality. As more data is accumulated, the accuracy of the determination can be improved.

[0050] FIG. 8 is a two-dimensional graph plotting the first and second principal components calculated based on data obtained when IPA (isopropyl alcohol) was ejected from a nozzle many times as a processing liquid at a target flow rate of 75 ml / min.

[0051] 8, the plots can be roughly divided into plot groups P1, P2, P3, and P4. The plot groups P1, P2, P3, and P4 differ from one another in the total number of ejections from a single nozzle for ejecting IPA. The total number of ejections is greatest in the order of P1, P2, P3, and P4.

[0052] It can be seen that the plots on the graph tend to move from left to right as the total number of ejections increases, which suggests that the ejection behavior of the treatment liquid in the transitional period, such as the overshoot / undershoot behavior, gradually changes as the total number of ejections increases.

[0053] The graph in Figure 9 shows the change over time in the flow rate corresponding to each of the plot groups P1, P2, P3, and P4. Although it is difficult to see because the lines overlap, the amount of overshoot gradually increases in the order of P1, P2, P3, and P4. This phenomenon is thought to be caused by, for example, deterioration of the flow control valve 44 due to wear or other factors, resulting in a decrease in control accuracy.

[0054] By discharging IPA (processing liquid) many times and accumulating data, it is possible to determine whether or not a device (e.g., a flow control valve) has deteriorated over time based on the movement of the plot on a two-dimensional graph. By understanding the tendency of the movement of the plot as the number of discharges increases, it is possible to predict deterioration of the device and change the flow rate setting value, control parameters, etc., or the maintenance schedule of the device.

[0055] <Detection (Monitoring) of Abnormalities During Stable Period> Next, detection of abnormalities such as hunting after the discharge flow rate of the treatment liquid has generally stabilized at the target flow rate (after the repeated overshoot / undershoot typical of the transition period has ended) will be described.

[0056] At the timing when the initial repetition of overshoot / undershoot is expected to end, specifically 10 / ωn seconds after the start of discharge (here, 10 / 8.15 = approximately 1.2 seconds), an FFT (Fast Fourier Transform) is initiated on the raw data of the actual flow rate (data sampled at a predetermined frequency). The FFT converts the raw data of the flow rate into frequency spectrum data (two-dimensional graph) that represents the relationship between frequency [Hz] and amplitude (flow rate amplitude). Then, a Lorentz function is fitted to the peaks that appear in the frequency spectrum data (at least one or all of them, if there are multiple peaks).

[0057] An example of frequency spectrum data obtained in this manner is shown in Figure 10. In Figure 10, in addition to the frequency spectrum curve (line 1), line 2 shows the results of fitting several peaks that appear on the frequency spectrum curve by combining multiple Lorentz functions.

[0058] By fitting with a Lorentz function, the peak frequency ([Hz]), peak half-width ([Hz]), and peak height ([arb.units]) can be determined for each peak, and these values ​​can be stored as the feature quantities of that peak. The appearance of a sharp peak (a peak with a high peak height and a small half-width) can be used to determine that an abnormality such as hunting in the flow rate (oscillation of the flow rate around the target value) has occurred. The presence of a sharp peak may be due to vibration in one of the devices (valves, pumps, etc.) of the processing liquid supply mechanism, and such vibration may indicate the possibility of an impending failure of that device.

[0059] When the flow rate is stable without oscillation, as shown in the raw data of the change in flow rate over time in Figure 11, no sharp peak appears in the Lorentzian function, as shown in Figure 12, which shows the corresponding frequency spectrum curve and Lorentzian function. In Figure 11, line 1 represents the flow rate set point, line 2 represents the measured flow rate (raw data), and line 3 represents the valve opening. In Figure 12, line 1 represents the frequency spectrum curve, and line 2 represents the result of fitting one of several peaks that appear in the frequency spectrum curve with a Lorentzian function.

[0060] It is possible to understand what abnormality each peak's feature (peak frequency ([Hz]), peak half-width ([Hz]), peak height ([arb.units])) corresponds to by comparing the feature with the flow data (raw data).

[0061] It is possible to understand what equipment malfunctions may be caused by the transition of the feature values ​​of each peak as the number of discharges increases, based on a large amount of flow rate data obtained by performing a large number of discharges and the corresponding feature values. For example, suppose that the peak height of the same peak frequency gradually increases with each discharge, ultimately leading to a failure of the flow control valve. In this case, the transition of the peak height of that peak frequency may be set as a key monitoring item, and an alarm may be issued to the operator if the peak height exceeds a predetermined threshold.

[0062] Furthermore, it is conceivable that there may be peaks or characteristic quantities that are not related to malfunctions of the device, and such peaks or characteristic quantities may be excluded from the monitoring targets.

[0063] An example of the flow of a learning procedure using fitting to a mathematical expression representing a transient response curve and FFT and Lorentz function fitting will be briefly described below.

[0064] (Step 1) A target flow rate is set for a certain processing liquid (e.g., DIW or IPA), and when the processing liquid is actually discharged, the maximum flow rate, the minimum flow rate, and the difference between the target flow rate and the actual flow rate at several points in time are recorded, as well as the opening of the flow control valve and its change over time. Note that these parameters are also recorded in conventional monitoring methods.

[0065] (Step 2) As described above, the log data of the detected flow rate (data on changes in flow rate over time) is fitted to a mathematical formula representing a transient response curve, and the above-mentioned parameters (amp, ζ, ωn, t0, y0, etc.) are calculated. The calculated parameters are associated with the type of treatment liquid and the set flow rate, and stored in the memory unit 19 of the control device 4.

[0066] (Step 3) As explained above, an FFT is performed on the flow rate data during a stable flow rate period (e.g., a period during which time t satisfies t > 10 / ωn) from the time-varying flow rate data to obtain a frequency spectrum. Then, Lorentzian function fitting is performed on the peaks that appear in the frequency spectrum to obtain the peak frequency, peak half-width, and peak height as feature quantities for each peak. The calculated feature quantities for each peak are associated with the type of processing liquid and the set flow rate and stored in the memory unit 19.

[0067] (Step 4) Based on the results of steps 1, 2, and 3, reference values ​​are set for one or more of the parameters (amp, ζ, ωn, t0, y0, etc.) obtained by fitting to the transient response curve in order to determine whether an abnormality has occurred. Reference values ​​are also set for the feature quantities obtained by Lorentz function fitting. Reference values ​​are also set for the maximum and minimum flow rate values, and for the difference between the target flow rate and the actual flow rate T seconds later (T is an appropriate time elapsed from the start of discharge (not limited to one time)). Furthermore, a reference value for the opening or opening fluctuation of the flow control valve may be set. Reference values ​​may also be set based on the principal component analysis described above. The reference values ​​may be set for each combination of the type of treatment liquid and the set flow rate.

[0068] In actual operation, until sufficient data is accumulated, the setting of the above-mentioned reference values ​​and the construction of a judgment model using the reference values ​​can be performed by a highly skilled engineer or operator. Furthermore, until sufficient data is accumulated, the presence or absence of an abnormality can be determined based on the raw data by a highly skilled engineer or operator.

[0069] (Step 5) As the number of discharges increases, flow rate data and the corresponding judgment results are accumulated in the memory unit 19 of the control device 4. Based on the accumulated data, normal / abnormal patterns are learned (machine learning), and the judgment model is refined. This enables the control device 4 to automatically judge abnormalities with high accuracy.

[0070] When a new discharge of processing liquid is performed, the above steps 1 to 3 are performed on the flow rate data obtained by the discharge. The new data obtained by the steps 1 to 3 (flow rate, flow control valve opening, parameter values ​​and feature values ​​related to fitting, etc.) is applied to the latest judgment model that has been refined by learning up to that point, thereby determining whether or not there is an abnormality in the current discharge of processing liquid. If an abnormality is determined, an alarm may be generated to notify an operator. If an abnormality is determined and it is determined that the abnormality can be addressed by modifying, for example, a control parameter (e.g., feedback gain), the control parameter may be automatically modified. If a flow rate abnormality that does not fit into any previously experienced pattern is detected, an alarm may be generated to request a skilled operator's judgment.

[0071] Some examples of abnormal patterns are shown below with reference to the graphs of Figures 13 to 16. In the graphs of Figures 13 to 16, line 1 is the flow rate, line 2 is the opening of the flow control valve, and line 3 (shown only in Figure 16) is the set flow rate.

[0072] <Unacceptable Overshoot> The graph in Figure 13 shows an example of an overshoot occurring at the beginning of discharge, reaching 1500 ml / min from a set flow rate of 1000 ml / min. The allowable overshoot value is, for example, 1050 ml / min, and an overshoot such as that shown in this graph is unacceptable. The initial overshoot can be associated with the parameter ζ. Therefore, it can be determined that an unacceptable overshoot has occurred when, for example, both the following two criteria (which can be stored in a memory unit): "ζ < 1" and "the maximum flow rate of the initial overshoot exceeds the allowable value (e.g., 1050 ml / min)" are satisfied.

[0073] <Unstable Flow Rate (Hunting)> The graph in Figure 14 shows an example where, when the set flow rate is 100 ml / min, the flow rate oscillates between 100 and approximately 70 ml / min (hunting occurs), resulting in an unstable flow rate. In this case, the peak half-width ([Hz]), which is a feature obtained by fitting using the Lorentz function described above, is small, and the peak height ([arb.units]) is large. By storing the reference values ​​of these feature values ​​in a memory unit and monitoring them, it is possible to determine whether an "unstable flow rate (hunting)" phenomenon is occurring.

[0074] <No flow rate> The graph in Figure 15 shows a case where the set flow rate is 1000 ml / min, and the actual flow rate remains at approximately 0 ml / min, even though the opening of the flow control valve gradually increases. In other words, the maximum flow rate is zero, and the opening of the flow control valve fluctuates greatly (increases monotonically in stages). By comparing the maximum flow rate and the opening of the flow control valve with reference values, it is possible to determine whether or not the "no flow rate" phenomenon is occurring.

[0075] <Large jumps in flow rate> The graph in Figure 16 shows that a large overshoot occurs at the beginning of discharge, exceeding the allowable value of 1600 ml / min, relative to the set flow rate of 1000 ml / min, followed immediately by an undershoot to 850 ml / min. The overshoot and undershoot are associated with large fluctuations (vibrations) in the opening of the flow control valve. In this case, too, it is possible to determine whether a "large jump in flow rate" has occurred by comparing the maximum flow rate and the opening of the flow control valve with reference values.

[0076] 15 and 16 are anomaly determinations made by conventional methods that are unrelated to fitting to a mathematical formula representing the transient response curve and unrelated to Lorentz function fitting. Such anomaly determinations made by conventional methods may also be incorporated into the anomaly determination procedure according to this embodiment. In the example of FIG. 16, it is conceivable that anomalies may also have occurred in the ζ value and the ωn value, and therefore it is conceivable that a flow rate anomaly can also be detected by comparing the parameter values ​​determined by fitting to a mathematical formula representing the transient response curve with the reference values.

[0077] According to the above embodiment, by fitting flow data to a mathematical formula representing a transient response curve and determining an equipment anomaly based on the resulting parameter values, even an unskilled operator can accurately determine an equipment anomaly. The mathematical formula representing the transient response curve allows transient changes to be expressed by a representative value consisting of one or a small number of parameter values ​​(e.g., overshoot behavior can be expressed by a ζ value), eliminating the need for a skilled operator to make the determination. While conventional methods require a skilled operator to make an empirical determination based on a few points or the overall shape of a flow rate transition graph, this embodiment eliminates this requirement. Furthermore, because continuous monitoring is possible without relying on a skilled operator, the possibility of overlooking an anomaly due to human error is significantly reduced.

[0078] The embodiments disclosed herein should be considered to be illustrative in all respects and not restrictive, and the above-described embodiments may be omitted, substituted, or modified in various ways without departing from the scope and spirit of the appended claims.

[0079] 38 Nozzle 36 Processing liquid line 42 Flow rate sensor 44 Flow rate control valve 46 Flow rate controller

Claims

1. A monitoring method for monitoring the ejection of a processing liquid from a nozzle in a processing liquid supply mechanism comprising a nozzle for ejecting a processing liquid, a processing liquid line through which the processing liquid supplied from a processing liquid supply source flows to the nozzle, a flow sensor and a flow control valve interposed in the processing liquid line, and a flow controller for controlling the opening of the flow control valve so that the flow rate of the processing liquid detected by the flow sensor becomes a target flow rate, the monitoring method comprising: a sampling step for sampling flow rate data which is the output of the flow sensor; a fitting step for fitting the sampled flow rate data to a mathematical equation representing a transient response curve; and a determination step for determining at least one of the presence or absence of an abnormality in the ejection of the processing liquid from the nozzle, the type of the ejection abnormality, and the presence or absence of signs of the occurrence of the ejection abnormality, based on at least parameter values ​​in the mathematical equation that has been fitted.

2. The monitoring method according to claim 1, wherein the judgment step judges at least one of the presence or absence of an ejection abnormality and the type of the ejection abnormality based on a comparison between one of the parameter values ​​in the fitted formula and a reference value.

3. The monitoring method according to claim 1, wherein the judgment step judges whether there is a sign of an ejection abnormality based on the change in the comparison result between one of the parameter values ​​in the formula obtained by fitting to the flow rate data of each of multiple ejections of the processing liquid and a reference value as the number of ejections increases.

4. The monitoring method of claim 1, wherein the determination step includes a step of calculating a first principal component and a second principal component for each ejection by performing principal component analysis based on multiple sets of parameter values ​​in the formula obtained by multiple ejections, and plotting the first and second principal components for each ejection on a two-dimensional graph, and determining at least one of the presence or absence of an abnormality in the ejection of the processing liquid from the nozzle, the type of the ejection abnormality, and the presence or absence of signs of the occurrence of the ejection abnormality based on the positional relationship between the plots on the two-dimensional graph.

5. The monitoring method of claim 4, wherein the judgment step judges that there is an abnormality in the ejection of processing liquid corresponding to a plot to be judged if the distance on the two-dimensional graph between the plot to be judged and an area where plots obtained when normal ejection is performed are concentrated is greater than a predetermined threshold value.

6. The monitoring method according to claim 4, wherein said determining step determines that there is a sign of an occurrence of an ejection abnormality when the plot on said two-dimensional graph tends to move in a specific direction as the number of ejections increases.

7. The monitoring method according to claim 1, wherein the judgment step further determines whether at least one of the maximum value, minimum value, and difference from a target flow rate at a certain point in time of the detected flow rate of the processing liquid is within a reference value, and if not within the reference value, it is determined that there is an ejection abnormality.

8. The following formulas 1 to 3 are used to represent the transient response curve. The monitoring method according to claim 1, wherein any one of the following is used, in which "y(t)" is the flow rate of the treatment liquid measured by the flow sensor, "amp" is a coefficient for making the value in curly brackets ({}) correspond to the magnitude of the output (flow rate), "ζ" is a damping coefficient, "ωn" is a natural angular frequency, "t0" is a coefficient representing delay due to bubble entrapment, and "y0" is a constant representing the initial value of the flow rate.

9. The monitoring method of claim 1, wherein the judgment step judges, based on parameter values ​​in the fitted formula, at least one of the following as the discharge abnormality: an unacceptably slow rise rate of the discharge flow rate, an unacceptable overshoot or undershoot of the discharge flow rate at the beginning of control, an unacceptable delay in the convergence of the discharge flow rate to the target value, and an unacceptable discharge delay due to bubble entrapment.

10. The monitoring method according to claim 1, further comprising the steps of: applying a fast Fourier transform to sampled flow rate data during a period after the flow rate data output from the flow sensor has become substantially stable, thereby converting the sampled flow rate data into frequency spectrum data; fitting at least one of the peaks included in the frequency spectrum data to a Lorentz function, and determining the peak frequency, peak half-width, and peak height, which are characteristic quantities of the peak; and an additional determination step of determining, based on the characteristic quantities of the peak, at least one of the presence or absence of an abnormality in the ejection of the processing liquid from the nozzle, the type of the ejection abnormality, and the presence or absence of signs of the occurrence of the ejection abnormality.

11. A substrate processing apparatus comprising: a processing liquid supply mechanism including: a substrate holding part for holding a substrate; a nozzle for supplying a processing liquid to a substrate held by the substrate holding part; a processing liquid line through which processing liquid supplied from a processing liquid supply source flows to the nozzle; a flow sensor and a flow control valve interposed in the processing liquid line; and a flow controller for controlling the opening of the flow control valve so that the flow rate of the processing liquid detected by the flow sensor becomes a target flow rate; and a control device, wherein the control device is configured to carry out the monitoring method described in any one of claims 1 to 10.

Images