Artificial intelligence algorithm for computing pressure values in a vacuum chamber

Abstract

The present invention relates to the determination of the pressure inside a vacuum chamber evacuated by a pumping system comprising an ion pump which may be useful for controlling the operation of the vacuum pump and / or for measuring the pressure in a vacuum system coupled to the ion pump. A machine-learning algorithm is used for outputting values of the actual pressure based on operating parameters of the ion pump, considering also the contribution of the ion pump leakage current; in particular, an inventive method comprises: measuring an overall current value of the ion pump, estimating a leakage current value of the ion pump through the machine-learning algorithm, subtracting the estimated leakage current value from the measured overall current value thus obtaining an actual current value, and determining the actual pressure by dividing the actual current value by a conversion rate of the ion pump.

Description

[0001] ARTIFICIAL INTELLIGENCE ALGORITHM FOR COMPUTING PRESSURE VALUES IN A VACUUM CHAMBER

[0002] The present invention relates to a computer-implemented monitoring method for monitoring the pressure in a vacuum chamber through an ion pump, a computer-readable storage device with a computer program configured to implement the method and a digital controller configured to implement the method.

[0003] The use of ion pumps in vacuum pumping systems is now widespread due to their advantages in the vacuum applications; in fact, ion pumps can operate in ultra-high vacuum (UHV) systems (as they can reach pressures even lower than 101 1mbar), they have a simple structure, and they do not cause noise and vibrations.

[0004] Ion pumps are often used as gauges to measure the pressure in a vacuum system, thanks to the known dependence of said pressure on the ion current.

[0005] Ion pumps, however, are affected by a phenomenon called “leakage current” which leads to an incorrect reading of the pressure. This problem derives from spurious currents which can arise when a high voltage is applied to the ion pump. Triode pumps are particularly affected by this problem. At low pressures, the leakage current value can be close to the value of the pressure-dependent ionization current, making useless the ion pump pressure readout.

[0006] A sputter ion pump with penning-trap current sensor is described in the U.S. patent 11289315. In this patent, a machine-learning engine is mentioned; however, the engine is used so that the drive controller can be programmed to find optimal voltage settings, but there is no reference to leakage current monitoring. Moreover, the solution of this patent focuses on maintaining a desired pressure in the vacuum chamber regardless of any leakage current contribution.

[0007] There is no need to explain the ion pumps working principles and their characteristic features, aspects already described for example in the aforementioned U.S. patent 11289315 (and in many other) and known to the person skilled in the art.

[0008] As already explained, the above prior art document does not address the problem of estimating the actual and correct pressure in a vacuum chamber evacuated by means of a pumping system comprising an ion pump, by properly considering the value and contribution of the leakage current at any time.

[0009] It is the general object of the present invention to provide a method and a system for obtaining such an actual and correct estimate of such pressure.

[0010] Such object is achieved through the method having the features set out in the annexed main independent claim as well as in the other annexed dependent claims.

[0011] The present invention will become more apparent from the following description that refers to a non-limiting example to be considered with the aid of the attached figures wherein:

[0012] Figure 1 is a block diagram of a system implementing a method according to the present invention through an A. I. model,

[0013] Figure 2 is a block diagram of a test bench configuration used for experimental data collection using three ion pumps,

[0014] Figures 3a, 3b, 3c show results of tests carried out on the three ion pumps about the relation between the current I measured by their electronic hardware and the pressure in the vacuum chamber,

[0015] Figure 4 shows leakage current computations plotted in a C vs P plan,

[0016] Figure 5 shows a comparison between predicted values of the leakage current and the corresponding actual measured values,

[0017] Figure 6 shows three plots (Fig. 6A, Fig. 6B, Fig. 6C) of leakage current percentage values as a function of different parameters, and

[0018] Figure 7 is a block diagram of an electronic hardware controlling an ion pump fluidly coupled to a vacuum system.

[0019] One way to obtain actual and correct estimate of pressures values in a vacuum chamber evacuated by a pumping system comprising an ion pump is by realizing a system configured to carry out a method according to the present invention, in particular by implementing a digital controller that integrates a software configured to carry out a method according to the present invention.

[0020] The method described in this patent application is able to trace and handle the possible presence of leakage current (LC) in ion pumps (diode or triode configuration) operating in different working conditions, using A.I. models integrated for example in a digital controller. An additional advantage of the method according to the present invention is the possibility to detect the leakage current values without any pressure sensors integrated into the vacuum system. This can be accomplished without turning off or modifying the operating parameters of the pump (such as the supply voltage).

[0021] In a preferred solution according to the present invention, the digital controller is configured to perform, through a supervised learning algorithm, a computer-implemented method as described in the following paragraphs.

[0022] Adding the possibility of networking the digital controller, information on the vacuum chamber pressure computation could be shared with the pumps suppliers, in order to further improve the codes behind the controller logic and deepen knowledge on the leakage current phenomenon in ion pumps and, more generally, on vacuum products.

[0023] A single digital controller may be used to control, collect and elaborate data from multiple ion pumps connected thereto. Even though not limited to a specific number of ion pumps, the minimal configuration is 1 : 1 (one controller for one ion pump).

[0024] A further aspect according to the present invention is a computer-implemented method for monitoring and estimating leakage current in ion pumps in a vacuum system which does not include any pressure sensor.

[0025] A training phase allows to determine how to detect and manage the presence of leakage current. This training phase is carried out with the help of a pressure sensor and comprises the following steps: a) mapping the “conversion rate” C which is defined from here on as: P [Torr] C1] where I is the intensity of current measured from the ion pump at a constant pressure value of P b) tracking the presence of the leakage current (LC), through C.

[0026] By the definition according to formula [1], the conversion rate C is such that it is affected by the possible presence of leakage current, since the measured current is given by the ionization current contribution and the leakage current contribution, as follows: _ t _ Ijorl+LC pj

[0027] P P When the ion pump is affected by leakage current the conversion rate C is inflated, especially at lower pressures, where the leakage current has more weight due to the small ionization current contribution Iionin formula [2], Instead, at higher pressures, C is almost not affected by the leakage contribution.

[0028] The current-pressure dependence in an ion pump is known to be linear and independent of the actual pressure so the leakage contribution can be determined at any pressure.

[0029] Once the leakage current is determined, such experimental results are used to train the A.I. model managed by a supervised learning algorithm, in particular a linear regression model which predicts the leakage current value computing the percentage of leakage current (LCperc) on the whole current I measured.

[0030] A typical system based on an A.I. model is shown in figure 1 and may be used both during operation and during training. It considers a time span f= [ to, tf and the collection of measurements of the current from the pump in then it takes the following quantities as the inputs of the model:

[0031] • I current intensity (in particular average value of the current measured in, e.g., nA) in the pump in time span f

[0032] • V supply voltage (in particular average value of the voltage measured in, e.g., V) to the pump in time span f

[0033] • pi lag 1 auto-correlation index of I measured in time span f or, alternatively, higher lag auto-correlation ones such as lag 2, lag 3 but with lower prediction accuracy

[0034] • Oi: standard deviation of I measured in time span f with pi and Gi being optional but their use leading to improved reliability.

[0035] The model of the linear regression for the LCperccan be computed as:

[0036] In order to reach an even more accurate model in terms of prediction of the output it has been verified that is possible to use log( / ) and log(cr / ) instead of I and <7 / due to the high range of variability of these quantities, so a more accurate model is: LCperc= / ?0+ * log(7) + / ?2 * l + / 3 * log / ) + Z?4 * V [4]

[0037] In both formulas and / ?4are different from zero, while the values of each or both of / 32and ?3may be equal to zero according to some embodiments.

[0038] As it can be understood from the above formulas [3] and [4], the dependance of LCperc onI and / or 07 is a monotonic increasing (same or different) function.

[0039] Usually, during training, several time spans are preferably considered with different I and V values.

[0040] The monitoring method, carried out by the controller, comprises the following steps: a) extract the input of the model I, V, pi and Oi from the current signal of the pump; b) obtain a real-time estimate of the leakage current affecting the pump through the A. I. model according to formula [3]; c) provide the real pressure in the chamber through the well-known dependence between pressure and current (cleaned from leakage current).

[0041] The applicability of the method has been validated by experimental tests; in particular, nitrogen pumping tests were conducted on a laboratory test bench, using three different pumps operating simultaneously. A test bench configuration for experimental data collection is schematized in figure 2 where:

[0042] • SIP1, SIP2, SIP3 are the three sputter-ion pumps;

[0043] • G1 is the vacuum gauge;

[0044] • IC1, IC2, IC3 are the ion pump cables;

[0045] • VI is a valve (isolation device);

[0046] In the test bench here described the pressure sensor is a Bayard-Alpert gauge, but other pressure sensors could be easily used such as extractor gauge, cold-cathode, spinning rotor gauge.

[0047] The tests, carried out on a test bench as previously described, show the relation between the ionization current I measured by the electronic hardware and the pressure P measured in the vacuum chamber given by equation [1], All three pumps have been tested for three different supply voltage values (3000V, 4000V, 5000V) and the results (shown in figures 3a, 3b, 3c) confirmed the linear relation between the measured I and the measured P. Moreover, the nitrogen pumping tests confirm the existence of distinct "conversion rate" values C, which depend on the operating conditions of the pumps. The conversion rate C and the leakage current contribution in a Sputter-Ion Pump (SIP) at different pressures during nitrogen pumping is shown in figure 4.

[0048] The predicted values of leakage current percentage given by the model according to formula [3] (LCperc) are compared with the actual values of leakage current measured (LCperc_true) the results, shown in figure 5, confirm that the accuracy of the model prediction can be improved by collecting supplementary data (as well-known by a person skilled in the art). In figure 6 there are shown three different plots of the LCpercvalues obtained with three sputter-ion pumps (with a nominal pumping speed of 5 1 / s) working at a supply voltage of 5kV as a function of different values of measured current (graph in Fig. 6A), different auto-correlation values (graph in Fig. 6B) and different current standard deviation (graph in Fig. 6C). Each dot represents a measurement relative to a single pump.

[0049] A further aspect of the present invention is a computer-readable storage device comprising a computer program that performs the computer-implemented method for the estimation of pressure in ion pumps. As represented in figure 7, the vacuum system is connected to an electronic hardware with user interface and an external interface to display the information and outputs of the monitoring method.

Claims

CLAIMS1. A computer-implemented method for monitoring the pressure in a vacuum chamber, said chamber being evacuated by a pumping system comprising an ion pump, the method comprising a machine-learning algorithm, having operating parameters of said ion pump as its inputs; wherein said method provides, as an output from the machine-learning algorithm, values of actual pressure inside the vacuum chamber; wherein the values of actual pressure inside the vacuum chamber are obtained by considering the contribution of the ion pump leakage current.

2. Method according to claim 1, comprising:- measuring an overall current value of the ion pump,- estimating a leakage current value of the ion pump through the machine-learning algorithm,- subtracting the estimated leakage current value from the measured overall current value thus obtaining an actual current value, and- determining the actual pressure by dividing the actual current value by a conversion rate of the ion pump.

3. Method according to claim 1 or 2, wherein the machine-learning algorithm comprises a linear regression model in order to estimate the ion pump leakage current.

4. Method according to claim 3, wherein the model of the linear regression provides as an output a percentage of leakage current on a whole current of the ion pump based on the ion pump supply voltage, the ion pump current and optionally an auto-correlation index of the ion pump current within a time span, preferably a lag 1 auto-correlation index, and / or a statistical distribution index of the ion pump current within a time span, preferably a standard deviation.

5. Method according to claim 4, wherein the linear regression is computed as: o + / ?! * I + / ?2* p + / ?3* (7 / + / ?4* V wherein P4are coefficients different from zero,I is the current through the ion pump,V is the voltage across the ion pump, p4is a lag 1 auto-correlation index of the current through the ion pump,(Ji is the standard deviation of the current through the ion pump.

6. Method according to claim 4, wherein the linear regression is computed as: / ?0 + / ?1 * log( / ) + / ?2 * l + / 3 * log / ) + / wherein P4are coefficients different from zero,I is the current through the ion pump,V is the voltage across the ion pump, p4is a lag 1 auto-correlation index of the current through the ion pump,(Ji is the standard deviation of the current through the ion pump.

7. Method according to any of claims from 3 to 6, wherein the machine-learning algorithm provides for preliminary supervised learning which is based on measurements of the pressure inside the vacuum chamber.

8. Method according to any of the preceding claims, wherein the input operating parameters of the ion pump comprise: the supply voltage, the current and preferably one or more of a statistical distribution index of the current, in particular a standard deviation, and an auto-correlation index of the current, in particular a lag 1 auto-correlation index, wherein said input operating parameters are used during execution of the machinelearning algorithm for the vacuum chamber pressure estimation and / or during training of the machine-learning algorithm.

9. A computer-readable storage device comprising: a computer program which is configured to perform the monitoring method according to any of the preceding claims when executed by one or more processors in order to estimate the pressure inside a vacuum chamber evacuated by a pumping system comprising an ion pump.

10. A digital controller configured to perform the method of any of claims 1 to 8.

Images