Method and apparatus for identifying excimer laser energy models - Patents.com

Abstract

The present invention discloses a method and apparatus for identifying an energy model of an excimer laser. The method includes the steps of: constructing a gated loop network for recognizing an energy model of an excimer laser; setting energy collection conditions in a single laser pulse manner within a plurality of preset times, and collecting a training data set for recognizing an energy model of an excimer laser; and training the constructed gated loop network using the training data set, and terminating the training when a training end condition is reached, to obtain an energy model of an excimer laser. Using the method provided by the present invention, the maximum error between the pulse energy generated by the recognized excimer laser energy model and the actual pulse energy is less than or equal to 1.5%, which can meet the simulation requirements for the energy characteristic control of an excimer laser.

Description

[Technical field]

[0001] The present invention relates to a method for identifying a laser energy model, in particular to a method for identifying an energy model of an excimer laser based on a gated loop network, and at the same time to an apparatus for recognizing a corresponding energy model of an excimer laser. [Background technology]

[0002] With the rapid development of laser technology, laser technology has been widely used in various fields. Among them, excimer lasers are widely used in industrial, medical, scientific research and other fields due to their characteristics of short wavelength, high power and narrow linewidth. In particular, rare gas halide excimer lasers have become the most major laser light source in the current semiconductor lithography industry due to their characteristics such as high peak power of output laser, large single pulse energy, and ultraviolet wavelength. The energy characteristics of excimer lasers are one of the three important indicators (energy, linewidth, wavelength) of excimer lasers for lithography, and directly determine the processing accuracy, yield and main dimensions of semiconductor lithography equipment. Therefore, the construction of an energy model of excimer lasers (i.e., excimer laser output energy model) has become the basis for the research and control of laser energy characteristics.

[0003] In the study of the emission energy of an excimer laser, the closer the energy model of the excimer laser used is to the actual emission energy law, the more advantageous the research is. However, the energy model of an excimer laser is a complex nonlinear model, and it is difficult to obtain an accurate model from theoretical derivation. Summary of the Invention [Problem to be solved by the invention]

[0004] The first technical problem to be solved by the present invention is to provide a method for recognizing an energy model of an excimer laser.

[0005] Another technical problem that the present invention aims to solve is to provide an apparatus for recognizing an energy model of an excimer laser. [Means for solving the problem]

[0006] To achieve the above objectives, the present invention adopts the following technical solutions:

[0007] According to a first aspect of an embodiment of the present invention, a method for identifying an energy model of an excimer laser comprises: Step S1: constructing a gate loop network to recognize the energy model of an excimer laser; Step S2 of setting energy collection conditions in a single laser pulse manner for a plurality of preset times to collect a training data set for recognizing an energy model of the excimer laser; and The method includes a step S3 of training the constructed gated-loop network using the training data set, and terminating the training when a training end condition is reached, thereby obtaining an energy model of the excimer laser.

[0008] Preferably, the gated loop network comprises gated loop units corresponding to a plurality of time series, each of the gated loop units comprises an input layer, a hidden layer and an output layer, the input layer is connected to the hidden layer, the hidden layer is connected to the output layer, and the hidden layers of adjacent gated loop units are connected to each other.

[0009] Preferably, the energy collection condition is a time interval of a single laser pulse and a discharge high voltage value, and the time interval of the single laser pulse is a time interval from a previous laser pulse to a current laser pulse.

[0010] Preferably, the hidden layer of each gate loop unit has a reset gate

[0011]

number

[0012]

number

[0013] The update gate b z(t) represents the amount of information brought to this moment by the state at the previous moment, and is expressed by the following equation.

[0014]

number

[0015] In the above equation, σ denotes the activation function, which is a sigmoid function. b W Z represents the input weight matrix of the update gate, b x(t) represents the input variables of the current gate loop network, b U Z represents the transfer matrix of the hidden layer state of the update gate, b h(t-1) represents the hidden layer state at the previous instant.

[0016] The reset gate b r(t) represents the degree to which the current state ignores the previous instantaneous state and is given by:

[0017]

number

[0018] In the above formula, b W r represents the input weight matrix of the reset gate, b Ur represents the transfer matrix of the hidden layer state of the reset gate.

[0019] The hidden layer state candidates b h(t) is the hidden layer state b It is used to aid in the calculation of h(t) and is given by the following formula:

[0020]

number

[0021] In the above formula, b W represents the input weight matrix of the candidate hidden layer states; b U represents the transfer matrix of the candidate hidden layer state for the hidden layer state at the previous moment,

[0022]

number

[0023] The current hidden layer state b h(t) is expressed as follows.

[0024]

number

[0025] Preferably, the output layer of the gate loop unit obtains the pulse energy E(t) of the excimer laser at a current instant according to the following formula:

[0026]

number

[0027] In the above equation, y(t) represents the energy factor of the pulse energy at the current instant, b W y represents the weight matrix from the hidden layer state to the output layer, and W Erepresents the conversion factor of the output scale.

[0028] Preferably, the training data set is an average of actual laser pulse energies at the same position in each corresponding laser burst mode under each discharge high voltage.

[0029] Here, all the actual laser pulse energies under each discharge high voltage and in each corresponding laser burst mode are the actual laser pulse energies collected within a preset time after setting the energy collection condition in the single laser pulse mode.

[0030] Preferably, the loss function of the gated loop network is

[0031]

number

[0032] In the above formula, E t (t) represents the average value of the actual laser pulse energy at the same position in the laser burst mode under the discharge high voltage of the training sample at the current instant, E(t) represents the excimer laser pulse energy sequence output from the gate loop network, and n represents a specific instant.

[0033] Preferably, the constructed gated loop network comprises the steps of training it using said training data set.

[0034] Step S31: Randomly select a training sample from the training data set, and input the energy collection conditions corresponding to the average value of all actual laser pulse energies in this training sample into the gated-loop network one by one, thereby training the gated-loop network to obtain a sequence of pulse energies under the set burst mode.

[0035] Step S32: Calculate a loss function of the gated-loop network corresponding to the currently selected training sample, and update the training parameters of the gated-loop network according to the loss function.

[0036] Step S33: Calculate the error between the pulse energy sequence currently outputted from the gate loop network under burst mode and the training sample.

[0037] Step S34: Steps S31 to S33 are repeated until a training end condition is reached, and then the training is ended and an energy model of the excimer laser is obtained.

[0038] Preferably, the condition for terminating the training of the gated loop network is a predetermined number of training times or a predetermined number of cyclical executions of steps S31 to S33, and the maximum error between each pulse energy in the burst mode pulse energy sequence output by the gated loop network each time and the pulse energy at the same position in the training sample is 0.15 mJ or less.

[0039] According to a second aspect of an embodiment of the present invention there is provided an apparatus for recognising an energy model of an excimer laser comprising a processor and a memory, the processor reading a computer program or instructions in the memory to perform the following operations:

[0040] Construct a gate loop network to recognize the energy model of the excimer laser and determine its input variables.

[0041] Energy collection conditions are set for a number of preset times and in the form of single laser pulses to collect a training data set for recognizing an energy model of the excimer laser.

[0042] The constructed gated-loop network is trained using the training data set, and when a training end condition is reached, the training is terminated and an energy model of the excimer laser is obtained. Effect of the Invention

[0043] The method and apparatus for identifying the energy model of an excimer laser provided by the present invention can identify the energy model of an excimer laser relatively accurately through a gate loop network. This excimer laser energy model can verify the generation of pulse energy with a maximum error between the actual pulse energy and the energy model of an excimer laser of less than 1.5%, so as to meet the simulation requirements of the energy characteristic control of the laser. This excimer laser energy model can facilitate the simulation research and analysis of the energy control method of an excimer laser, and shorten the experiment time, which is of great significance to the improvement of the energy stability control and dose accuracy control of an excimer laser. [Brief description of the drawings]

[0044] [Figure 1] 1 is a flowchart of a method for identifying an energy model of an excimer laser provided by an embodiment of the present invention. [Diagram 2] FIG. 1 is a network structure diagram of a hidden layer of a gate loop network in a method for identifying an energy model of an excimer laser provided by an embodiment of the present invention. [Diagram 3] This figure shows the single pulse energy change regularity of an excimer laser in the constant discharge high voltage operation mode. [Figure 4] FIG. 13 is a diagram showing the maximum error between the pulse energy obtained from the excimer laser energy model and the actual pulse energy under each discharge voltage in the method for identifying an excimer laser energy model provided by an embodiment of the present invention. [Diagram 5] FIG. 11 is a curve diagram showing the change in maximum pulse energy error during the training process of a gate loop network in a method for identifying an energy model of an excimer laser provided by an embodiment of the present invention. [Figure 6a]FIG. 11 is a comparison diagram between the single pulse energy generated by the excimer laser energy model and the actual single pulse energy when the discharge voltage is 1550V in the method for identifying the excimer laser energy model provided by the embodiment of the present invention. [Figure 6b] FIG. 11 is a comparison diagram between the single pulse energy generated by the excimer laser energy model and the actual single pulse energy when the discharge voltage is 1550V in the method for identifying the excimer laser energy model provided by the embodiment of the present invention. [Figure 6c] FIG. 11 is a comparison diagram between the single pulse energy generated by the excimer laser energy model and the actual single pulse energy when the discharge voltage is 1550V in the method for identifying the excimer laser energy model provided by the embodiment of the present invention. [Figure 7a] FIG. 11 is a comparative diagram between the actual pulse energy of a laser with a repetition rate of 1KHz to 4KHz and the pulse energy generated by the model when the constant discharge high voltage is 1500V in the operating condition in the method for identifying the energy model of an excimer laser provided by an embodiment of the present invention. [Figure 7b] FIG. 11 is a comparative diagram between the actual pulse energy of a laser with a repetition rate of 1KHz to 4KHz and the pulse energy generated by the model when the constant discharge high voltage is 1500V in the operating condition in the method for identifying the energy model of an excimer laser provided by an embodiment of the present invention. [Figure 8] FIG. 13 is a diagram showing the maximum error between the actual pulse energy of a laser with a repetition rate of 1 KHz to 4 KHz and the pulse energy generated by the model when the operating condition is a constant discharge high voltage of 1500 V in the method for identifying the energy model of an excimer laser provided by an embodiment of the present invention. [Figure 9] FIG. 2 is a schematic diagram showing the structure of an apparatus for recognizing an energy model of an excimer laser provided by an embodiment of the present invention; DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0045] The technical contents of the present invention will be described in more detail below in combination with the attached drawings and specific embodiments.

[0046] In order to solve the problem that there is a large difference between the pulse energy sequence generated by the conventional excimer laser energy model and the actual excimer laser output energy sequence, as shown in Figure 1, the embodiment of the present invention first provides a method for identifying an excimer laser energy model, which mainly includes the following steps:

[0047] Step S1: Build a gate loop network to recognize the energy model of the excimer laser.

[0048] In an embodiment of the present invention, an excimer laser energy model is obtained using gated-loop network recognition. The gated-loop network includes gated-loop units corresponding to multiple time series, and each gated-loop unit has an input layer, a hidden layer and an output layer. The input layer is connected to the hidden layer, the hidden layer is connected to the output layer, and the hidden layers of adjacent gated-loop units are also connected.

[0049] Here, the input layer of the gated loop unit corresponding to the current time series receives the input variables of the current moment and inputs the input variables to the hidden layer of said gated loop unit, which obtains the hidden layer state of the current moment based on the received hidden layer state of the previous moment and the input variables, which is input to the output layer of this gated loop unit on the one hand to obtain the excimer laser pulse energy of the current moment, and on the other hand to the hidden layer of the gated loop unit corresponding to the next time series and used as input, where each moment corresponds to one time series.

[0050] Specifically, through the characteristic analysis of the output energy of the excimer laser, it can be found that the condition variables that directly affect the output energy of the excimer laser are the time interval of a single laser pulse and the discharge high voltage value. Here, the time interval of a single laser pulse is the time interval from the previous laser pulse to the current laser pulse. Therefore, the input variables of the gate loop network select the time interval of a single laser pulse and the discharge high voltage value, as shown in equation (1).

[0051]

number

[0052] In the above formula, b x(t) represents the input variable of the gate loop network at the instant t, HV(t) represents the discharge high voltage value of the laser pulse of the excimer laser at the instant t, and Δt(t) represents the time interval of the laser pulse of the excimer laser at the instant t. b W in represents the input scale transformation matrix.

[0053] As shown in Figure 2, the hidden layer of each gate loop unit has a reset gate b r(t), the update gate b z(t) and candidates for hidden layer states b h(t) is the hidden layer state at the previous moment. b h(t-1) and the input variables at the current moment b The input is x(t), and the hidden layer state at the current moment is b When the hidden layer state at the previous moment is used, h(t) is obtained by processing the hidden layer state at the previous moment.

[0054] Here, the update gate b z(t) represents the amount of information brought to this moment by the state at the previous moment, and is used to capture the long-term dependency relationship over time. Its expression is shown in Equation (2).

[0055]

number

[0056] In the above equation, σ represents the activation function, which is a sigmoid function. b W z represents the input weight matrix of the update gate, b x(t) represents the input variables of the gate loop network at the current instant, b U z represents the transfer matrix of the hidden layer state of the update gate, b h(t-1) represents the hidden layer state at the previous instant.

[0057] Reset Gate b r(t) represents the degree to which the current state ignores the state at the previous moment and is used to capture short-term dependencies in the time series, and is the reset gate. b The activation function of r(t) is selected to use a sigmoid function, denoted as σ, and its expression is as shown in Equation (3).

[0058]

number

[0059] In the above formula, b W r represents the input weight matrix of the reset gate, b x(t) represents the input variables of the gate loop network at the current instant, b U r represents the transfer matrix of the hidden layer state of the reset gate, b h(t-1) represents the hidden layer state at the previous instant.

[0060] Candidate hidden layer states b h(t) is the hidden layer state b Used to assist in the calculation of h(t), the activation function of the candidate hidden layer state is selected and used as the tanh function, and its expression is as shown in Equation (4).

[0061]

number

[0062] In the above formula, b W represents the input weight matrix of the candidate hidden layer states, b x(t) represents the input variables of the gate loop network at the current instant, b U represents the transfer matrix of the candidate hidden layer state for the hidden layer state at the previous moment,

[0063]

number

[0064] The current hidden layer state b The expression for h(t) is shown in equation (5).

[0065]

number

[0066] In the above formula, b z(t) represents the update gate, b h(t-1) represents the hidden layer state at the previous moment, b h(t) is a candidate for the hidden layer state b h(bar)(t)

[0067]

number

[0068] The output of the gate loop unit is the current hidden state. bBased on h(t), the pulse energy E(t) of the excimer laser at the current moment is obtained based on equations (6) and (7). Specifically, the energy factor y(t) of the pulse energy at the current moment is obtained by equation (6), and the activation function of the energy factor y(t) is selected and used as a sigmoid function, which is denoted as σ.

[0069]

number

[0070] In the above formula, b W y represents the weight matrix from the hidden layer state to the output layer, b h(t) represents the hidden layer state at the current instant.

[0071] Based on the energy factor y(t) of the pulse energy at the current instant and equation (7), obtain the pulse energy sequence E(t) of the excimer laser at the current instant.

[0072]

number

[0073] In the above formula, W E represents the conversion factor of the power scale, and y(t) represents the energy factor of the pulse energy at the current instant.

[0074] By combining equations (1)-(7), an embodiment of the present invention constructs a gated loop network that recognizes the energy model of an excimer laser.

[0075] Step S2: Set energy collection conditions in a single laser pulse manner within a plurality of preset times, and collect a training data set for recognizing an energy model of the excimer laser.

[0076] The collection condition of a single pulse of energy means the time interval and discharge high voltage value of a single laser pulse, i.e., the time interval from the previous laser pulse to the current laser pulse and the discharge high voltage corresponding to the current laser pulse.

[0077] The training data set used to recognize the energy model of an excimer laser is the average value of the actual laser pulse energies at the same position corresponding to each laser's burst mode under each discharge high voltage. That is, one discharge high voltage corresponds to the average value of the actual laser pulse energies at the same position of each laser's burst mode. Here, all the actual laser pulse energies corresponding to each laser's burst mode under each discharge high voltage are the actual laser pulse energies collected after setting the energy collection condition in the single laser pulse mode within a preset time.

[0078] Here, the average values ​​of the actual laser pulse energies at the same positions corresponding to the burst modes of each laser under each discharge high voltage in the training data set constitute a series of average values ​​of the actual laser pulse energies in the burst modes, which are used as training samples in the training data set.

[0079] The test data set used to recognize the energy model of the excimer laser is the average value of the actual laser pulse energy at the same position corresponding to the burst mode of each laser under at least one discharge high voltage. The collection method of all the actual laser pulse energies corresponding to the burst mode of each laser under each discharge high voltage is the same as the collection method of the training data set. Here, the average value of the actual laser pulse energy at the same position corresponding to the burst mode of each laser under each discharge high voltage in the test data set constitutes the average value series of the actual laser pulse energy in the burst mode, which is used as the test sample in the test data set.

[0080] Specifically, in semiconductor lithography applications, excimer lasers operate in burst mode, so each moment can be made to correspond to each time series, and each time series can be made to correspond to a burst mode. The pulse energy series of an excimer laser collected in burst mode is shown in Figure 3. From the figure, it can be seen that in each pulse energy series collected in burst mode, the first few pulse energies (e.g., E in Figure 3) are m,1 , E m,2 , E m+1,1 and E m+1,2 ) show overshoots of different magnitudes, and the stable region behind the burst mode also shows fluctuations in energy (for example, E in Fig. 3). m,i and E m+1,j ) When the discharge high voltage is constant, the pulse energy of the excimer laser has a large relationship with its position in the pulse energy sequence collected in the burst mode. Therefore, when analyzing the characteristics of the excimer laser energy, the pulse energies at different positions in the pulse energy sequence collected in the burst mode are analyzed separately.

[0081] In order to remove the noise contained in the pulse energy of the laser pulse energy sequence collected in Burst mode, the average value of the pulse energies at different positions in each laser pulse energy sequence collected in Burst mode is taken using Equation (8).

[0082]

number

[0083] Where:

[0084]

number

[0085] When collecting a training data set for recognizing the energy model of an excimer laser, a KrF excimer laser that generates a wavelength of 248 nm is taken as an example. Since the wavelength of the excimer laser also affects the collected actual pulse energy data, in the course of the experiment, the wavelength is controlled to 248.327 nm using a feedback technique. The excimer laser operates at a repetition rate of 1 KHz, and each laser pulse is set with its time interval and discharge high voltage value, and an energy detector is used to collect the pulse laser energy of the laser under discharge high voltages of 1400 V, 1450 V, 1550 V and 1600 V, respectively, for one minute each. The energy detector collects the actual pulse energy series in 100 burst modes under each discharge high voltage of the excimer laser for one minute, and the actual pulse energy series in each burst mode includes 250 pulse energies. Here, in the actual pulse energy series in each burst mode, the time interval between pulses corresponding to each pulse energy is the inverse of the repetition rate, and the unit may be ms. For the actual pulse energy sequence of 100 burst modes under each discharge high voltage, calculate the average value of the actual laser pulse energy at the same position in the burst mode of the laser under each discharge high voltage according to formula (8), starting from the first pulse energy of the actual pulse energy sequence in each burst mode, respectively, to obtain the average value sequence of the actual pulse energy in one burst mode corresponding to each discharge high voltage. Take the example of calculating the average value of the actual laser pulse energy at the first position in the burst mode of the laser under any discharge high voltage. Just add up the first actual pulse energies of the pulse energy sequence in 100 burst modes and then divide by 100.

[0086] Therefore, the training data set for recognizing the energy model of the excimer laser is composed of the average value series of the actual pulse energy in the burst mode corresponding to the discharge high voltages of 1400 V, 1450 V, 1550 V, and 1600 V. That is, there are four average value series of the actual pulse energy in the burst mode, and each discharge high voltage corresponds to a average value series of the actual pulse energy under the burst mode.

[0087] When testing a data set to recognize the energy model of an excimer laser, similarly, take a KrF excimer laser generating a wavelength of 248 nm as an example. The excimer laser operates at a repetition rate of 1 KHz, sets its time interval and discharge high voltage value for each laser pulse, and uses an energy detector to collect the pulse laser energy of the laser under a discharge high voltage of 1550 V for one minute, and calculates the average value of the actual laser pulse energy at the same position of the burst mode of the laser under this discharge high voltage in the same way as training the data set, and obtains an average value series of the pulse energy in one burst mode corresponding to this discharge high voltage. Depending on the actual requirements, the test data set for recognizing the energy model of an excimer laser may include setting its time interval and discharge high voltage value for each laser pulse at a single fixed operating repetition rate or different operating repetition rates, and collecting an average value series of the actual pulse energy in the burst mode corresponding to multiple pre-processed discharge high voltages at preset times. For example, using the same method to obtain the training data set, we can obtain the average values ​​of the actual laser pulse energies at the same position in the burst mode of the laser corresponding to a discharge high voltage of 1500 V when the repetition rates of the excimer laser operation are 1 KHz, 2 KHz, 3 KHz and 4 KHz, respectively.

[0088] Step S3: Using the training data set, train the constructed gated-loop network. When the training end condition is reached, the training is terminated and an energy model of the excimer laser is obtained.

[0089] Since each weight matrix of the gated loop network constructed to recognize the energy model of the excimer laser needs to be acquired by learning, the gated loop network needs to be trained. In practical applications, the training method adopted is the backpropagation through time algorithm (BPTT).

[0090] The gated loop network is applied to recognize the energy model of an excimer laser by taking the burst mode as a time series and defining the loss function of the entire network as shown in equation (9).

[0091]

number

[0092] In the above formula, E t (t) represents the average value of the actual laser pulse energy at the same position of the laser Burst mode under the discharge high voltage of the training sample at the current instant, E(t) represents the excimer laser pulse energy series output from the gate loop network, and n represents a specific instant.

[0093] The input of the activation function is

[0094]

number

[0095]

number

[0096]

number

[0097] When t = n,

[0098]

Number

[0099]

Number

[0100]

Number

[0101] As can be seen from equations (12) to (16), when t < n,

[0102]

Number

[0103]

Number

[0104]

Number

[0105]

Number

[0106] Regarding the input of the activation function as

[0107]

Number

[0108]

number

[0109]

number

[0110] The input of the activation function is

[0111]

number

[0112]

number

[0113]

number

[0114]

number

[0115]

number

[0116] The input of the activation function is

[0117]

number

[0118]

number

[0119]

number

[0120] From equations (9) to (22), we can obtain the method for updating the training parameters of the gate loop network as shown in equations (23) to (29).

[0121]

number

[0122] In the above formula, b W y (i+1) represents the weight matrix from the current hidden layer state to the output layer, b W y (i) represents the weight matrix from the previous hidden layer state to the output layer. b W z (i+1) represents the input weight matrix of the current update gate, b W(i) represents the input weight matrix of the previous update gate. b U z(i+1) represents the transfer matrix of the hidden layer state of the current update gate, b U z (i) represents the transfer matrix of the hidden layer state of the previous update gate. b W(i+1) represents the input weight matrix of the candidate for the current hidden layer state, b W(i) represents the input weight matrix of the candidate for the previous hidden layer state. b U(i+1) represents the transfer matrix of the current hidden layer state candidate to the hidden layer state at the previous moment, b U(i) represents the transfer matrix of the candidate for the previous hidden layer state to the hidden layer state at the previous instant. b W r (i+1) represents the input weight matrix of the current reset gate, b W r (i) represents the input weight matrix of the previous reset gate. b U r (i+1) represents the transfer matrix of the hidden layer state of the current reset gate, b U r (i) represents the transfer matrix of the hidden layer state of the previous reset gate, and λ represents the learning step length.

[0123] Training the constructed gated-loop network using the training dataset involves the following steps:

[0124] Step S31: Randomly select a training sample from the training data set, and input the energy collection conditions corresponding to the average value of all actual laser pulse energies in the training sample into the gated-loop network one by one, thereby training the gated-loop network to obtain a sequence of pulse energies under burst mode.

[0125] In order to fully utilize the data, the gated loop network randomly selects training samples from the training data set for each training. For example, a training data set (four actual pulse energy series in Burst mode) for recognizing the energy model of an excimer laser is constructed from the average value series of actual pulse energies in Burst mode corresponding to discharge high voltages of 1400V, 1450V, 1550V, and 1600V. The average value series of actual pulse energies in Burst mode with a discharge high voltage of 1600V is randomly selected, and the energy collection conditions (time interval of a single laser pulse and discharge high voltage value) corresponding to the average value of each actual pulse energy in this average value series of actual pulse energies in Burst mode are input one by one to the gated loop network, and after the gated loop network is trained, it outputs one pulse energy series in a burst mode.

[0126] Step S32: Calculate a loss function of the gated-loop network corresponding to the currently selected training sample, and update the training parameters of the gated-loop network according to the loss function.

[0127] The average value series of actual pulse energy in burst mode to obtain a discharge high voltage of 1600V and the pulse energy series in burst mode corresponding to the output discharge high voltage of 1600V after training the gate loop network are substituted into equation (9) to calculate the loss function of the gate loop network. Based on this loss function, the training parameters of the gate loop network can be updated by combining equations (23) to (29).

[0128] Step S33: Calculate the error between the pulse energy sequence under burst mode currently output from the gate loop network and the current training sample.

[0129] Each pulse energy in the pulse energy series in burst mode with a discharge high voltage of 1600 V currently output by the gate loop network is subtracted from the pulse energy at the same position in the training sample (the average value series of actual pulse energies in burst mode with a discharge high voltage of 1600 V) to obtain the error between each pulse energy in the pulse energy series in burst mode currently output by the gate loop network and the pulse energy at the same position in the training sample.

[0130] Step S34: Steps S31 to S33 are repeated until a training end condition is reached, and then the training is ended and an energy model of the excimer laser is obtained.

[0131] The training end condition of the gated loop network may be a predetermined number of training times (e.g., 100,000 times) or a predetermined number of cycles of steps S31 to S33. The maximum error between each pulse energy in the burst mode pulse energy sequence under a certain discharge high voltage output by the gated loop network and the pulse energy at the same position in the training sample is 0.15 mJ or less each time (as shown in FIG. 4, the ... is 0.15 mJ or less each time). Burst The pulse energy error of each mode is less than 0.15 mJ, and since the center energy value is 10 mJ, the relative error is less than 1.5%.

[0132] Training parameters for ultimate excimer laser energy model. b W y , b W z , b U z , b W, b U, b W r , b U r can be obtained by training a gated loop network and generate an energy model of the excimer laser based on equations (1)-(7).

[0133] In the process of training the gate loop network, the output discharges at a certain high voltage. Burst The change of the maximum error of the pulse energy of the mode is shown in Figure 5. The step length on the abscissa is the number of times the gate loop network is trained, and the ordinate is the absolute value of the pulse energy error output from the gate loop network. As can be seen from Figure 5, the maximum error of the pulse energy output from the gate loop network gradually decreases until it is below 0.15 mJ. This proves that the gate loop network constructed according to the present invention for recognizing the energy model of the excimer laser converges in the training process.

[0134] Step S4: Using the test data set, verify the accuracy of the excimer laser energy model.

[0135] The accuracy of the excimer laser energy model is verified using mean series of actual pulse energies in burst mode corresponding to multiple preconditioned discharge high voltages collected at a single fixed operating frequency or at different operating frequencies.

[0136] For example, the excimer laser energy model is verified by selecting the average value series of actual pulse energies in the burst mode with a discharge high voltage of 1550 V. In this average value series of actual pulse energies in the burst mode, the energy collection conditions (time interval of a single laser pulse and discharge high voltage value) corresponding to the average value of each actual pulse energy are input one by one into the gate loop network, and the excimer laser energy model outputs one burst mode pulse energy series.

[0137] By subtracting each pulse energy in the pulse energy series in the burst mode of the excimer laser energy model with a discharge high voltage of 1550V from the pulse energy at the same position in the average value series of the actual pulse energy in the burst mode of the discharge high voltage of 1550V, the error between each pulse energy in the pulse energy series in the burst mode of the excimer laser energy model and the actual pulse energy is obtained. As shown in Figures 6a to 6c, here, Figure 6b shows the change in the actual pulse energy in the burst mode of the discharge high voltage of 1550V, Figure 6c shows the change in the pulse energy in the burst mode of the discharge high voltage of 1550V output from the excimer laser energy model, and Figure 6a shows the overlap between the pulse energy change in the burst mode of the discharge high voltage of 1550V output from the excimer laser energy model and the change in the actual pulse energy. From the figure, it can be seen that the pulse energy change obtained by the excimer laser energy model overlaps well with the actual pulse energy change.

[0138] As another example, the average value series of actual pulse energy in Burst mode from 1KHz to 4KHz under a discharge high voltage of 1500V is selected to verify the excimer laser energy model, as shown in Figure 7a and Figure 7b. Here, Figure 7a shows the actual pulse energy change in Burst mode from 1KHz to 4KHz under a discharge high voltage of 1500V, and Figure 7b shows the pulse energy change in Burst mode from 1KHz to 4KHz under a discharge high voltage of 1500V output from the excimer laser energy model. Comparing Figure 7a and Figure 7b, it can be seen that the pulse energy obtained by the excimer laser energy model at different repetition rates is consistent with the trend of the laser energy change in Burst mode obtained by actual measurement. The maximum error is shown in Figure 8, as can be seen from the figure, the maximum error at different repetition rates is less than 0.13mJ, i.e., less than 1.5%.

[0139] Based on the relationship between dose accuracy and energy stability, when the dose accuracy of 0.5% for lithography is met, the maximum error of energy stability is 2.74%, and the error of pulse energy generated by the excimer laser energy model provided by the present invention is smaller than the error of the control accuracy of energy stability, so it can be concluded that this model meets the simulation requirements for laser energy characteristic control.

[0140] In addition, as shown in FIG. 9, the embodiment of the present invention further provides an apparatus for recognizing an energy model of an excimer laser, which includes a processor 32 and a memory 31, and may further include a communication component, a sensor component, a power component, a multimedia component, and an input / output interface according to actual needs. Here, the communication component, the sensor component, the power component, the multimedia component, and the input / output interface are all connected to the processor 32. As mentioned above, the memory 31 may be a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, etc. The processor 32 may be a central processing unit (CPU), a graphics processor (GPU), a field programmable logic gate array (FPGA), an application specific integrated circuit (ASIC), a digital signal processing (DSP) chip, etc. The other communication components, sensor components, power components, multimedia components, etc. can all be implemented using general parts installed in conventional smartphones, and will not be specifically described in this specification.

[0141] Also, an embodiment of the present invention provides an apparatus for recognizing an energy model of an excimer laser, comprising a processor 32 and a memory 31. The processor 32 reads computer programs or instructions in the memory 31 to perform the following operations:

[0142] Construct a gated loop network to recognize the energy model of the excimer laser.

[0143] Energy collection conditions are set for a number of preset times and in the form of single laser pulses to collect a training data set for recognizing an energy model of the excimer laser.

[0144] The constructed gated-loop network is trained using the training data set, and when the training end condition is reached, the training is terminated and an energy model of the excimer laser is obtained.

[0145] In addition, an embodiment of the present invention further provides a computer-readable storage medium having stored thereon instructions that, when executed on a computer, cause the computer to execute a method for recognizing an energy model of an excimer laser as described in FIG. 1, the specific implementation of which is not repeated in this specification.

[0146] In addition, an embodiment of the present invention further provides a computer program product including instructions that, when executed on a computer, cause the computer to execute a method for recognizing an energy model of an excimer laser as described in FIG. 1, but the specific implementation thereof will not be repeated in this specification.

[0147] The method and apparatus for identifying the energy model of an excimer laser provided by the embodiment of the present invention can identify the energy model of an excimer laser relatively accurately through a gate loop network. This excimer laser energy model can verify the generation of pulse energy with a maximum error between the actual pulse energy and the energy model of the laser of less than 1.5%, so as to meet the simulation requirements of the energy characteristic control of the laser. This excimer laser energy model can facilitate the simulation research and analysis of the energy control method of the excimer laser, and shorten the experiment time, which is of great significance to the improvement of the energy stability control and dose accuracy control of the excimer laser.

[0148] Above, the method and apparatus for identifying the energy model of an excimer laser provided by the present invention have been described in detail. For those skilled in the art, any obvious modifications made to the present invention without departing from the essential content of the present invention will fall within the scope of protection of the patent right of the present invention.

Claims

1. 1. A method for identifying an energy model of an excimer laser, comprising: Step S1: constructing a gate loop network for recognizing an energy model of an excimer laser; Step S2 of setting energy collection conditions in a single laser pulse manner for a number of preset times to collect a training data set for recognizing an energy model of the excimer laser; and A step S3 of training the constructed gated loop network using the training data set, and terminating the training when a training end condition is reached, and obtaining an energy model of the excimer laser; The loss function of the gate loop network is [0010] and In the above equation, E t (t) represents the average value of the actual laser pulse energy at the same position in the laser burst mode under the discharge high voltage of the training sample at the current moment, E(t) represents the pulse energy sequence of the excimer laser output from the gate loop network, and n represents a specific moment. A method for identifying an energy model of an excimer laser, characterized in that:

2. The gated loop network includes gated loop units corresponding to a plurality of time series; Each of the gate loop units comprises an input layer, a hidden layer and an output layer; 2. The method for identifying an energy model of an excimer laser as described in claim 1, wherein the input layer is connected to the hidden layer, the hidden layer is connected to the output layer, and the hidden layers of adjacent gate loop units are connected.

3. 2. The method for identifying an energy model of an excimer laser according to claim 1, wherein the energy collection conditions are a time interval of a single laser pulse and a discharge high voltage value, and the time interval of the single laser pulse is a time interval from a previous laser pulse to a current laser pulse.

4. The hidden layer of each gate loop unit is a reset gate ｂ r(t), update gate ｂ z(t) and the candidates for hidden layer state [0025] ( ｂ h(t) (The alphabet following the subscript b is in bold. The same applies below.) The update gate ｂ z(t) represents the amount of information brought to this moment by the state at the previous moment, and is expressed by the following formula: [0030] In the above equation, σ represents the activation function, which is a sigmoid function. ｂ W ｚ represents the input weight matrix of the update gate bz(t), ｂ x(t) represents the input variables of the current gate loop network; ｂ U ｚ represents the transfer matrix of the hidden layer state of the update gate bz(t), ｂ h(t-1) represents the hidden layer state at the previous instant, The reset gate ｂ r(t) represents the degree to which the current state ignores the state at the previous instant, and is given by the following formula: [0045] In the above formula, ｂ W ｒ represents the input weight matrix of the reset gate b r(t), ｂ U ｒ represents the transfer matrix of the hidden layer state of the reset gate b r(t), The hidden layer state candidates ｂ h(t) is the hidden layer state ｂ It is used to aid in the calculation of h(t) and is given by the following equation: [0050] In the above formula, ｂ W represents the input weight matrix of the candidate hidden layer states; ｂ U represents the transfer matrix of the candidate hidden layer state with respect to the hidden layer state at the previous moment, [006] represents the Hadamard product, The current hidden layer state ｂ 2. The method of claim 1, wherein h(t) is expressed as follows: [0070]

5. The output layer of the gate loop unit obtains the current instantaneous excimer laser pulse energy E(t) according to the following formula: [0080] where y(t) represents the energy factor of the pulse energy at the current instant, ｂ W ｙ represents the weight matrix from the hidden layer state to the output layer, and W Ｅ 5. The method of claim 4, wherein: represents a conversion factor of power scale.

6. the training data set being an average value of actual laser pulse energies at the same location in each corresponding laser burst mode under each discharge high voltage; The method for identifying the energy model of an excimer laser according to claim 1, characterized in that under each discharge high voltage, all actual laser pulse energies in each corresponding laser burst mode are actual laser pulse energies collected within a preset time after setting the energy collection condition in the single laser pulse mode.

7. training the constructed gated-loop network using the training data set, Step S31: training the gated-loop network to obtain a sequence of pulse energies under a set burst mode by randomly selecting a training sample from the training data set and inputting the energy collection conditions corresponding to the average value of all actual laser pulse energies in the training sample into the gated-loop network one by one; A step S32 of calculating a loss function of the gated-loop network corresponding to the currently selected training sample and updating the training parameters of the gated-loop network based on the loss function; A step S33 of calculating an error between the pulse energy sequence under the burst mode currently outputted from the gate loop network and the current training sample; and The method for identifying an excimer laser energy model according to claim 1, further comprising a step S34 of cyclically executing steps S31 to S33 until a training end condition is reached, and then terminating training and obtaining an energy model of the excimer laser.

8. The method for identifying an excimer laser energy model according to claim 7, characterized in that the condition for terminating the training of the gated loop network is a predetermined number of training times or a predetermined number of times of cyclically executing steps S31 to S33, and the maximum error between each pulse energy in the burst mode pulse energy sequence output by the gated loop network each time and the pulse energy at the same position in the training sample is 0.15 mJ or less.

9. An apparatus for recognizing an energy model of an excimer laser, comprising: A processor and a memory, the processor reading a computer program or instructions in the memory and performing the following operations: An operation of constructing a gate loop network for recognizing an energy model of an excimer laser and determining input variables of the gate loop network; setting energy collection conditions in the form of single laser pulses for a plurality of preset times to collect a training data set for recognizing an energy model of the excimer laser; training the constructed gated loop network using the training data set, and when a training end condition is reached, terminating the training and obtaining an energy model of the excimer laser; The loss function of the gate loop network is [0090] and In the above formula, E t (t) represents the average value of the actual laser pulse energy of the training sample at the same position in the laser burst mode under the discharge high voltage at the current moment, E(t) represents the pulse energy sequence of the excimer laser output from the gate loop network, and n represents a specific moment; 2. A device for recognizing an energy model of an excimer laser.

Images