JP7837540B2 - Pressure-type flow control device - Google Patents

Description

This invention relates to a pressure-type flow control device used in semiconductor manufacturing equipment, chemical plants, and the like, and more particularly to a pressure-type flow control device equipped with a self-diagnostic function.

In semiconductor manufacturing equipment and chemical plants, various flow meters and flow control devices are used to control the flow rates of material gases, etching gases, and other fluids. Among these, pressure-type flow control devices are widely used because they can precisely control the mass flow rate of various fluids using a relatively simple mechanism combining a control valve and a throttling section (e.g., an orifice plate or critical nozzle). Unlike thermal-type flow control devices, pressure-type flow control devices have superior flow control characteristics, allowing for stable flow control even when the primary supply pressure (pressure upstream of the control valve) fluctuates significantly.

Some pressure-type flow control devices adjust the flow rate by controlling the fluid pressure upstream of the throttling section (hereinafter sometimes referred to as the upstream pressure P1). The upstream pressure P1 is usually controlled by adjusting the opening degree of a control valve installed upstream of the throttling section. When the critical expansion condition (upstream pressure P1 / downstream pressure P2 ≥ approximately 2: in the case of argon gas) is met, the velocity of the gas flowing through the throttling section is fixed at the speed of sound, and the mass flow rate of the gas flowing downstream of the throttling section is known to be proportional to the upstream pressure P1 regardless of the magnitude of the downstream pressure P2 downstream of the throttling section. Therefore, flow rate control is possible by controlling the upstream pressure P1.

International Publication No. 2017/170174 International Publication No. 2021/111979

However, depending on the type of gas used, especially with prolonged use, corrosion or clogging may occur in the throttle section of a pressure-type flow control device. In this case, the opening area of the throttle section changes, causing a discrepancy between the flow rate calculated from the upstream pressure P1 and the actual flow rate, which was initially accurate. Furthermore, since pressure-type flow control devices typically lack flow meters such as thermal sensors, it is impossible to directly measure the actual flow rate of the gas flowing through the channel after the throttle opening has changed. Therefore, pressure-type flow control devices equipped with a self-diagnostic function to determine whether or not there is an abnormality in the throttle section have been proposed.

In the pressure-type flow control device described in Patent Document 1, the drop in upstream pressure after closing the control valve is measured during the period satisfying the critical expansion conditions, and self-diagnosis is performed by comparing this with a reference value. Furthermore, it has been found that the relationship ln(P1(t)/P1i) = -α・t (where P1(t) is a function of the upstream pressure P1 with respect to time t, and P1i is the initial upstream pressure) holds during the pressure drop period. By comparing the slope α obtained from the pressure measurement result with the reference slope α0, it is possible to determine whether or not an abnormality has occurred in the throttling section. The self-diagnosis method described in Patent Document 1 has the advantage that, regardless of the current flow rate, self-diagnosis can be started from any flow rate or any initial upstream pressure, such as at the end of a semiconductor manufacturing process.

However, even if self-diagnosis reveals clogging or enlargement of the throttling mechanism, while a warning can be issued to the user, it is difficult to continue using the device in that state. Therefore, conventionally, when an abnormality in the throttling mechanism was detected, replacement of the throttling mechanism was necessary, but replacing only the throttling mechanism was not easy. Thus, a practical solution was often to replace the entire pressure-type flow control device with a new one.

Replacing pressure-type flow control devices each time significantly increases costs and reduces convenience. This is particularly problematic in environments where the diameter of the throttle section is prone to change, such as when using highly corrosive gases, as the frequency of replacement increases. Therefore, there was a demand for a pressure-type flow control device that could be used stably for longer periods, even in environments where the diameter of the throttle section is prone to change.

The applicant has developed and commercialized a method for controlling the flow rate by comparing initial baseline pressure drop data with pressure drop data during device operation, and taking into account errors in machine gun force. Furthermore, the applicant has focused on a method for calculating the optimal flow rate correction value from pressure drop data during device operation using AI control based on a regressive neural network.

This invention has been made in view of the above points, and its main objective is to provide a pressure-type flow control device that can be used more stably and continuously using AI control.

A pressure-type flow control device according to an embodiment of the present invention is
The aperture section,
A control valve provided on the upstream side of the aforementioned throttling section,
An upstream pressure sensor that detects the upstream pressure, which is the pressure in the flow path between the throttling portion and the control valve,
A pressure-type flow control device comprising a control mechanism having a self-diagnostic function that diagnoses using pressure drop data of the upstream pressure measured by the upstream pressure sensor,
The control mechanism is
The system includes a calculation unit that, using initial setting values for determining the opening degree of a control valve obtained from multiple pressure-type flow control devices, corrects these initial setting values obtained from measured flow errors, and uses data on the change in upstream pressure over time after closing the control valve from a predetermined pressure . It then compares this with pressure drop data constructed based on weight values connecting multiple nodes constituting a regressive neural network, and if it determines that the change is outside the acceptable range, it executes a calculation formula to obtain a corrected value by correcting the initial setting values .
The system is configured to calculate operational correction data for modifying the initial setting value from operational pressure drop data, which is obtained by closing the control valve at any point during operation and observing the change in upstream pressure over time.

The AI prediction formula is constructed from pressure drop data and correction values obtained during pre-shipment inspections. The correction values are then recalculated using pressure drop data from the operational pressure-type flow control device to ensure accurate flow control.

In the above configuration, a downstream pressure sensor is further provided to detect the downstream pressure, which is the pressure in the downstream flow path of the throttling section. The operational pressure drop data is acquired during the period after the control valve is closed, while the critical expansion condition is satisfied, indicating that the pressure ratio of the upstream pressure to the downstream pressure is equal to or greater than a specified value.

In this case, the control mechanism also includes a data recording unit that records the operational pressure drop data and operational correction data that corrects the initial setting values , and the calculation unit can further calculate and update the weight values.

Furthermore, in these cases, the control mechanism may also acquire operational pressure drop data and/or operational correction data for modifying the initial setting values of other pressure control devices, and update the weight values.

Furthermore, in these cases, the control mechanism can also control multiple pressure control devices.

According to the pressure-type flow control device of the present invention, by using pressure drop data and correction values acquired from multiple pressure-type flow control devices to determine weights and derive a calculation formula, it is possible to calculate the optimal correction value from the pressure drop data of the operating pressure-type flow control device, thereby enabling more continuous and accurate flow control.

This is a schematic diagram showing a pressure-type flow control device according to an embodiment of the present invention. This graph shows the changes in upstream pressure P1 and downstream pressure P2 when the valve is closed from a gas supply state. This graph shows the change in the pressure drop of the upstream pressure when there is an abnormality in the opening. Regarding the method for determining the AI-based correction value calculation formula, (a) shows the initial correction value for each flow rate setting, (b) shows the actual flow rate as the initial correction value, and (c) shows the flow rate control correction value calculated from the actual flow rate. This shows a schematic diagram of a neural network. This graph shows one embodiment of flow rate self-diagnosis and correction value determination according to an embodiment of the present invention.

The embodiments of the present invention will be described in detail below with reference to the drawings, but the present invention is not limited to the embodiments described below.

Figure 1 shows a pressure-type flow control device 10 according to this embodiment. The pressure-type flow control device 10 comprises a throttling section 2 provided in the flow path 1, a control valve 6 provided upstream of the throttling section 2, an upstream pressure sensor 3 for detecting the upstream pressure P1 between the throttling section 2 and the control valve 6, a downstream pressure sensor 4 for detecting the downstream pressure P2 downstream of the throttling section 2, and a temperature sensor 5 for detecting the temperature between the throttling section 2 and the control valve 6.

For the upstream pressure sensor 3 and downstream pressure sensor 4, for example, semiconductor piezoresistive diffusion pressure sensors or capacitance manometers are used, and for the temperature sensor 5, for example, resistance thermometers or thermistors are used. For the control valve 6, for example, a piezoelectric element-driven valve (hereinafter sometimes referred to as a piezoelectric valve) is used, which opens and closes a metal diaphragm valve body using a piezoelectric actuator. A piezoelectric valve is a valve (proportional valve) that can be opened to any degree by adjusting the driving voltage applied to the piezoelectric element. Furthermore, for the throttling section 2, for example, an orifice plate or a critical nozzle is used, and the opening diameter of the throttling section 2 is set to, for example, 10 to 2000 μm.

The pressure-type flow control device 10 also includes a control mechanism (or control circuit) 7 connected to each of the sensors 3, 4, and 5 and the control valve 6. The control mechanism 7 is configured to perform a self-diagnostic function, closing the control valve 6 and measuring the flow pressure when the semiconductor manufacturing process ends (when the gas supply to the process chamber is stopped). The control mechanism 7 in this embodiment incorporates a CPU, memory, A/D converter, etc., on a circuit board, and includes a computer program for performing the operations described later; it is realized through a combination of hardware and software. In this embodiment, the control mechanism 7 is shown as being located inside the pressure-type flow control device 10, but some or all of the components of the control mechanism 7 may be located outside the pressure-type flow control device 10.

The upstream side of the pressure-type flow control device 10 is connected to a gas supply source (not shown), and the downstream side is connected to the process chamber 12 of a semiconductor manufacturing apparatus via a shut-off valve 11 (typically an on/off valve). A vacuum pump 13 is connected to the process chamber 12, allowing for vacuuming of the process chamber 12 when gas is supplied from the upstream side of the pressure-type flow control device 10. In the embodiment shown in Figure 1, the shut-off valve 11 is located outside the pressure-type flow control device 10; however, the shut-off valve 11 may be built into the pressure-type flow control device 10. For example, an AOV (air-driven valve) or a solenoid valve can be used as the shut-off valve 11.

In a semiconductor manufacturing process, when supplying gas to the process chamber 12, the control mechanism 7 uses the output of the upstream pressure sensor 3 (and the downstream pressure sensor 4 and temperature sensor 5) to calculate the flow rate and controls the control valve 6 so that the flow rate passing through the throttling section 2 becomes the set flow rate. The calculated flow rate may be displayed as a flow rate output value on the display unit of the external control device.

More specifically, when the critical expansion condition (P1/P2 ≥ approximately 2: in the case of argon gas) is met, the calculated flow rate is determined from the output of the upstream pressure sensor 3 according to the relationship Q = K1P1 (where K1 is a proportionality constant that depends on the type of fluid and the fluid temperature), and the control valve 6 is feedback controlled so that the calculated flow rate is the same as the set flow rate. Furthermore, under noncritical expansion conditions, the calculated flow rate is determined according to the relationship Q = K2P2 ^m (P1-P2) ⁿ (where K2 is a proportionality constant that depends on the type of fluid and the fluid temperature, and the exponents m and n are values derived from the actual flow rate), and the control valve 6 is feedback controlled so that the calculated flow rate is the same as the set flow rate.

Furthermore, the pressure-type flow control device 10 of this embodiment is configured to perform a self-diagnosis by acquiring pressure drop data at the end of the process when the control valve 6 transitions to the closed state and the gas supply is stopped. It is also possible to perform the self-diagnosis by closing the shut-off valve 11 in addition to the control valve 6 at the end of the process. In this case, the self-diagnosis process can be performed on any gas supply line without affecting semiconductor manufacturing using other gas supply lines.

The solid line graph in Figure 2 shows the changes in upstream pressure P1 and downstream pressure P2 when the control valve 6 is closed. As shown in Figure 2, when the control valve 6 receives a closing command at time t1 and remains closed thereafter, the upstream pressure P1 decreases from the initial upstream pressure P1i during gas flow, and the downstream pressure P2 also decreases from the initial downstream pressure P2i.

In this embodiment, self-diagnosis is performed based on P1(t) measured during the critical expansion period Δt between times t1 and t2, i.e., the period during which P1/P2 ≥ approximately 2 (in the case of argon gas) satisfies the critical expansion conditions. By performing self-diagnosis while confirming whether critical expansion conditions are actually present, it becomes possible to secure the maximum effective pressure drop data acquisition period regardless of the design of the fluid supply control system or the semiconductor manufacturing process. Furthermore, in recent years, the downstream pressure P2 during gas supply (i.e., the initial downstream pressure P2i in Figure 2) is sometimes relatively high. However, since self-diagnosis is performed after determining whether critical expansion conditions are actually present, the diagnostic accuracy can be improved. Additionally, because the period during which critical expansion conditions can be maintained is extended by closing only the control valve 6, self-diagnosis can be performed more easily.

On the other hand, the dashed line graph in Figure 2 shows the changes in upstream pressure P1 and downstream pressure P2 when the shut-off valve 11 is closed simultaneously with the control valve 6. The upstream pressure P1 decreases from the initial upstream pressure P1i during gas flow, and the downstream pressure P2 increases from the initial downstream pressure P2i during gas flow. That is, pressure fluctuations occur so that the pressure difference between the upstream and downstream sides of the throttling section 2 is eliminated, and the upstream pressure P1 and downstream pressure P2 converge to substantially the same equilibrium pressure value P'' over time. Although the critical expansion period Δt' between times t1 to t2', which satisfies the critical expansion conditions, is shorter than in the case of the solid line, the self-diagnosis process can be performed without affecting semiconductor manufacturing using other gas supply lines because the shut-off valve 11 is closed.

The pressure ratio P1/P2, which indicates the lower limit of critical expansion conditions, varies depending on the type of gas. For example, it is 2.05 for argon gas, but 1.90 for hydrogen and 1.89 for nitrogen; each gas has its own specific value. Furthermore, critical expansion conditions also change depending on the upstream gas temperature. Therefore, the control mechanism 7 may be configured to determine the critical expansion condition determination formula during self-diagnosis based on at least one of the gas type and gas temperature.

The following describes a specific example of self-diagnosis. After closing the control valve 6, the control mechanism 7 checks whether critical expansion conditions are present based on the measured upstream pressure P1 and downstream pressure P2. It also compares the pressure drop data P1(t) of the upstream pressure P1 measured during the critical expansion period Δt with the pre-stored reference pressure drop data Y(t). The reference pressure drop data Y(t) is generally measured before factory shipment and stored in the control mechanism 7's memory. While the reference pressure drop data Y(t) here is the normal pressure drop data measured before factory shipment, it may also be measurement data from abnormal conditions, previous measurement data, or setting data not based on measurement.

Figure 3 shows the pressure drop curve A1 for the upstream pressure P1 under normal conditions after closing the control valve 6, the pressure drop curve A2 when clogging occurs in the throttling section, and the pressure drop curve A3 when opening enlargement occurs in the throttling section due to corrosion, etc.

As can be seen from Figure 3, when the throttling section 2 is clogged, gas flow becomes more difficult, causing the pressure drop curve A2 to shift upward compared to the normal pressure drop curve A1. In this case, the time required for the upstream pressure P1 to drop to a predetermined value becomes longer. Furthermore, the upstream pressure P1 after a predetermined time has elapsed since closing the control valve 6 will be higher than under normal conditions.

On the other hand, when the opening of the throttling section 2 widens, the gas flows more easily, so the pressure drop curve A3 shifts downward compared to the normal pressure drop curve A1. In this case, the time required for the upstream pressure P1 to drop to a predetermined value becomes shorter. Furthermore, the upstream pressure P1 after a predetermined time has elapsed since closing the control valve 6 is lower than in the normal state.

Next, we will explain how to derive the formula for calculating the correction value using AI with a recurrent neural network in this invention.

Typically, pressure-type flow control devices undergo flow calibration as part of pre-shipment testing. Specifically, this involves correcting the set value for controlling the opening and closing of the control valve 6 in order to control the upstream pressure. When using a pressure-type flow control device, the control mechanism 7 sets the upstream pressure according to the flow rate setting value requested from an external source (e.g., semiconductor manufacturing equipment) (for example, 50% if 50% of the maximum flow rate is required), with the maximum flow rate that the pressure-type flow control device can control set to 100%. For example, when the request from the external source is for a setting of 100%, the calculation unit of the control mechanism 7 (e.g., CPU) sends an opening command for the control valve 6 as 48,000 digits. The value of 48,000 digits is a digital signal value within the control mechanism 7. When the critical expansion condition is met, the flow rate is controlled by controlling the upstream pressure. While this varies depending on the orifice diameter (which is the throttling section 2), for example, when flowing at 100% flow rate, this value is conveniently determined as an internal command value to control the control valve 6 to set the upstream pressure to 300 kPa. More specifically, if the control valve 6 is a piezo valve, 48,000 digits commands, for example, to set the piezo voltage to 80 V. Furthermore, as shown in Figure 4(a), when the flow rate is set to 10%, a command of 4800 digits is sent to control the control valve 6 to control the upstream pressure, and when the flow rate is 1%, a command of 480 digits is sent.

Although it varies depending on the type of pressure-type flow control device, the pressure-type flow control device used in this embodiment is designed to flow at 600 sccm when set to 100% (hereinafter, the designed flow rate value will be referred to as the reference value). In order to control the upstream pressure, the control valve 6 is controlled according to the digits in the right column shown in Figure 4(a) by receiving commands from the control mechanism 7 according to the setting values of 1, 3, 5, 10, 20, 40, 60, and 100% shown in the table. At this time, when the actual flow rate is measured, an error of about ±0.1 to 0.4 occurs relative to the reference value, as shown in Figure 4(b). This occurs because the orifice diameter, which is the throttling part, differs slightly within the error range of the diameter for each device, and there are differences between devices. Then, the correction value shown in Figure 4(c) is determined by taking into account the errors that occurred for each setting value. Next, to understand the characteristics of the orifice (the throttling section 2), the control valve is closed when the upstream pressure reaches 300 kPa (48,000 digits internally in the control mechanism), and the pressure drop data showing the change in upstream pressure over time is recorded, as shown in Figure 4(d).

Using this pressure drop data and correction value as a set, for example, 1000 data points are used to calculate the weight values that connect multiple nodes constituting a recurrent neural network.

Figure 5 is a schematic diagram of a recurrent neural network. In this embodiment, the input layer 80 contains pressure drop data, and the output layer 82 contains correction values. There are 49 input layers, and from these 49 pressure drop data points, the values for the 9 output layers (the number of correction values shown in Figure 4(c)) are calculated. The number of hidden layers 81 is not particularly limited, but in this embodiment, an example is shown with 64 nodes in layer a, 128 nodes in layer b, 256 nodes in layer c, 128 nodes in layer d, and 64 nodes in layer e. (The value of n in xn is 48, and the value of n in an is 63; the number of n changes according to the number of nodes.)

The number of weights connecting the nodes changes depending on the number of nodes in the hidden layer. Increasing the number of weights (the number of hidden layers and the number of nodes in each layer) improves the expressive power of the neural network, but it also requires a significant amount of computation time. Recent advancements in computer speed have made it possible to significantly increase the number of hidden layers. To input values into the input layer 80 and obtain output values in the output layer 82, it is first necessary to determine the weights. By learning these weight values, accurate prediction becomes possible. In this embodiment, weights connecting each node are calculated from approximately 1000 pressure drop data points and correction values (input layer values (pressure drop data) and output layer values (correction values)) obtained in pre-shipment testing, and the numerical prediction calculation formula y = f(x) in the hidden layer is constructed.

The following describes a specific example of flow rate control in this embodiment. Figure 6 is a flowchart showing the process of deriving a correction value using the numerical prediction calculation formula y = f(x) described above, based on the pressure drop data obtained from the self-diagnosis performed by the pressure-type flow rate control device 10.

As shown in Figure 6, the pressure-type flow control device 10 performs a self-diagnosis to determine the state of the throttling section 2. First, as shown in step S1, when gas is flowing at the set flow rate, for example, 100% of the set flow rate, the control mechanism 7 receives a command to stop the gas supply and issues a closing command (for example, a 0% flow rate command) to the control valve 6.

Next, as shown in step S2, the pressure drop data (operational pressure drop data) of the upstream pressure is measured based on the output of the upstream pressure sensor 3. Preferably, the operational pressure drop data is obtained by sampling the output of the upstream pressure sensor 3 while determining whether the critical expansion condition is met based on the outputs of the upstream pressure sensor 3 and the downstream pressure sensor 4, thereby acquiring the pressure drop data P1(t). This is because if the critical expansion condition is not met, the pressure will not release easily, and the drop characteristics will change, potentially causing problems for the neural network's learning and prediction.

Next, as shown in step S3, the operating pressure drop data is transmitted to the control mechanism 7 and compared with the initial pressure drop data recorded at the time of shipment (or the updated data if the data has been updated) to determine whether the change is within an acceptable range. If the transmitted pressure drop data P1(t) is determined to be within an acceptable range, as shown in step S7, it is determined that no abnormality has occurred in the throttling section and that no correction of the flow control signal is necessary, and the diagnostic process is terminated.

On the other hand, if in step S4 it is determined that the change is outside the acceptable range, as shown in step S5, 49 P1(t) values (where t is, for example, the upstream pressure value acquired every 2 seconds, with the measurement start time set to 0) are input to x0 to x48 of the input layer 80 of the regressive neural network. A calculation unit executes a numerical prediction calculation formula constructed based on the weight values connecting the multiple nodes constituting the input layer 80, the hidden layer 81, and the output layer 82, and outputs nine correction values y0 to y8 from the output layer 82. The output correction values (operational correction data) update the correction values set at the time of shipment (for example, the correction values shown in Figure 4(c)), and as shown in step S6, they are stored in the memory unit of the control mechanism 7 as new correction values.

In this way, a correction value is obtained from the pressure drop data P1(t), which is the result of the self-diagnosis process. Then, the correction formula for generating the internal flow control signal to be used in subsequent flow control operations is determined.

Furthermore, the control mechanism 7 includes a data recording unit that records operational pressure drop data and operational correction data. The calculation unit further calculates and updates the weight values, thereby improving the accuracy of the AI prediction using the neural network. Moreover, acquiring operational pressure drop data and/or correction data from other pressure control devices connected via the network and updating the weight values further improves the accuracy of the AI prediction. Other pressure control devices connected via the network can utilize data not only from pressure control devices installed within the same factory, but also from pressure control devices installed in other factories.

Furthermore, the control mechanism 7 can also utilize a computer system equipped with a high-speed CPU, large-capacity memory, and large-capacity storage device to control multiple pressure control devices. In this case, it is also possible to centrally manage pressure control devices located not only within the same factory but also in other networked factories.

The pressure-type flow control device according to the embodiment of the present invention is suitably used for accurately controlling the flow rate of gas supplied in semiconductor manufacturing equipment and the like over long periods of time.

1 Flow path 2 Constriction section 3 Upstream pressure sensor 4 Downstream pressure sensor 5 Temperature sensor 6 Control valve 7 Control mechanism 10 Pressure-type flow control device 11 Shut-off valve 12 Process chamber 13 Vacuum pump

Claims (5)

The aperture section,
A control valve provided on the upstream side of the throttling section,
An upstream pressure sensor that detects the upstream pressure, which is the pressure in the flow path between the throttling portion and the control valve,
A pressure-type flow control device comprising a control mechanism having a self-diagnostic function that diagnoses using pressure drop data of the upstream pressure measured by the upstream pressure sensor,
The control mechanism is
The system includes a calculation unit that, using initial setting values for determining the opening degree of a control valve obtained from multiple pressure-type flow control devices, corrects these initial setting values obtained from measured flow errors, and uses data on the change in upstream pressure over time after closing the control valve from a predetermined pressure . It then compares this with pressure drop data constructed based on weight values connecting multiple nodes constituting a regressive neural network, and if it determines that the change is outside the acceptable range, it executes a calculation formula to obtain a corrected value by correcting the initial setting values .
A pressure-type flow control device that calculates operational correction data for modifying the initial setting value using the calculation unit, based on operational pressure drop data of the change in upstream pressure over time, acquired after closing the control valve at any time during operation. The system further includes a downstream pressure sensor that detects the downstream pressure, which is the pressure in the downstream flow path of the constricted portion.
The pressure-type flow control device according to claim 1, wherein the in-operation pressure drop data is acquired during the period after closing the control valve, while satisfying the critical expansion condition that indicates the pressure ratio of the upstream pressure to the downstream pressure is equal to or greater than a specified value. The control mechanism is
The pressure-type flow control device according to claim 1 or 2, further comprising a data recording unit that records the operating pressure drop data and the operating correction data for correcting the initial setting value , wherein the calculation unit further calculates and updates the weight value. The pressure-type flow control device according to claim 3, wherein the control mechanism acquires operational pressure drop data and/or operational correction data for correcting the initial setting value of another pressure control device, and updates the weight value. The pressure-type flow control device according to claim 3 or 4, wherein the control mechanism controls a plurality of pressure control devices.