DE102019209561A1 - Method and device for optimizing a circuit board material for the production of a circuit board using a Bayesian optimization process - Google Patents

Abstract

The invention relates to a method for determining a material composition of a printed circuit board to be produced, the material composition being determined by a material parameter set with material parameters, with the following steps: - Carrying out (S1-S9) a Bayesian optimization method based on a probabilistic conductivity model, in particular a conductivity Gaussian process model or a Bayesian neural network model, as a quality function, whereby the probabilistic conductivity model simulates a dependence of conductivity differences to a given conductivity model depending on material parameters, whereby during the Bayesian optimization process circuit boards, which are manufactured according to test material parameters, with regard to their thermal conductivity and its processability are measured, with the probabilistic conductivity model and a probabilistic processability model, in particular ere a processability Gaussian process model or a processability Bayesian neural network model with the respectively measured thermal conductivity and the respectively measured processability can be trained or updated, with an acquisition function based on the probabilistic conductivity model being optimized, taking into account the probabilistic processability model, to each to obtain a next test material parameter set for measuring the thermal conductivity and the processability of a printed circuit board manufactured accordingly, and selecting (S10) a set of material parameters for the material composition of the printed circuit board to be manufactured in order to obtain a predetermined or maximized thermal conductivity according to the probabilistic conductivity model.

Description

Technical area

The invention relates to the manufacture of printed circuit boards, and in particular to methods for optimizing the printed circuit board material with regard to its thermal conductivity and processability. In particular, the present invention relates to measures for the selection of material parameters for the manufacturing process of printed circuit boards.

Technical background

Circuit boards for electronics are used to hold and make electrical contact with electronic components. The dissipation of locally generated heat via the circuit board plays a significant role in the service life and performance of circuit board electronics. When the electronics are operated, heat is generated depending on the electrical power converted in the electronic components. Especially for applications in power electronics, in which the components generate a high amount of heat, new materials with high thermal resistance and conductivity are being developed.

The thermal conductivity of the printed circuit board can be varied through the selection, nature and proportions of the materials. In particular, the thermal conductivity can be increased by using fillers, which, however, increase the viscosity and thus have a negative effect on processability.

So far, the thermal conductivity can be roughly estimated depending on the material parameters using a deterministic mixing rule. Only materials and proportions are taken into account here, but not the influence of filler shape and size. A model that can reliably predict the processability of a circuit board depending on its material composition is also not known.

Optimizing the material therefore requires many experiments, which are, however, time-consuming, since the circuit board has to be manufactured and measured accordingly. Thus, only a satisfactory material for a circuit board is determined by the empirical procedure.

Disclosure of the invention

According to the invention, a method and a device for determining material parameters for the material composition of a circuit board according to claim 1 and a device and a manufacturing method for a circuit board according to the independent claims are provided.

Further refinements are given in the dependent claims.

• - Durchführen eines Bayes'schen Optimierungsverfahrens basierend auf einem probabilistischen Leitfähigkeitsmodell, insbesondere einem Leitfähigkeits-Gaußprozessmodell oder einem Bayes'schen neuronalen Netz-Modell, als Qualitätsfunktion, wobei das probabilistische Leitfähigkeitsmodell eine Abhängigkeit von Leitfähigkeitsunterschieden zu einem vorgegebenen Leitfähigkeitsmodell über Materialparametern nachbildet, wobei während des Bayes'schen Optimierungsverfahrens Leiterplatten, die entsprechend Test-Materialparametern gefertigt werden, hinsichtlich ihrer thermischen Leitfähigkeit und ihrer Verarbeitbarkeit vermessen werden, wobei das probabilistische Leitfähigkeitsmodell und ein probabilistisches Verarbeitbarkeitsmodell, insbesondere ein Verarbeitbarkeits-Gaußprozessmodell oder ein Verarbeitbarkeits-Bayes'sches neuronales Netz-Modell, mit der jeweils gemessenen thermischen Leitfähigkeit und der jeweils gemessenen Verarbeitbarkeit trainiert oder aktualisiert werden, wobei eine Akquisitionsfunktion basierend auf dem Leitfähigkeits-Gaußprozessmodell unter Berücksichtigung des probabilistischen Verarbeitbarkeitsmodells optimiert wird, um jeweils einen nächsten Test-Materialparametersatz zur Vermessung der thermischen Leitfähigkeit und der Verarbeitbarkeit einer entsprechend hergestellten Leiterplatte zu erhalten, und
• - Auswählen eines Materialparametersatzes für die Materialzusammensetzung der herzustellenden Leiterplatte, um entsprechend des probabilistischen Leitfähigkeitsmodells eine vorgegebene oder maximierte thermischen Leitfähigkeit zu erhalten.

According to a first aspect, a method for determining a material composition of a printed circuit board to be manufactured is provided, the material composition being determined by a material parameter set with material parameters, with the following steps:

• - Carrying out a Bayesian optimization method based on a probabilistic conductivity model, in particular a conductivity Gaussian process model or a Bayesian neural network model, as a quality function, the probabilistic conductivity model simulating a dependency of conductivity differences to a given conductivity model using material parameters, with during the Bayesian optimization method Printed circuit boards that are manufactured according to test material parameters are measured with regard to their thermal conductivity and their processability, the probabilistic conductivity model and a probabilistic processability model, in particular a processability Gaussian process model or a processability Bayesian neural network model, be trained or updated with the respective measured thermal conductivity and the respectively measured processability, whereby an acquisition nsfunction is optimized based on the conductivity Gaussian process model, taking into account the probabilistic processability model, in order to obtain a next test material parameter set for measuring the thermal conductivity and the processability of a correspondingly manufactured circuit board, and
• - Selecting a material parameter set for the material composition of the printed circuit board to be produced in order to obtain a predetermined or maximized thermal conductivity in accordance with the probabilistic conductivity model.

The above method uses a Bayesian optimization method to optimize material parameters for the material composition of printed circuit boards. Since a larger number of material parameters can be optimized with the aid of the Bayesian optimization method than can be done manually with justifiable effort, the quality of the material composition of circuit boards can be significantly improved. In addition, the influences of the manual selection of Material parameters can be reduced to the circuit board properties and especially their variability.

In order to also take known knowledge into account in addition to the Bayesian optimization method, an already known circuit board model based on expert knowledge about the relationship between material proportions and thermal conductivity can be used. Such a combination of a machine learning model and an existing circuit board model makes it possible to optimize a material composition of a circuit board material according to predetermined performance parameters. On the one hand, the machine learning model learns inaccuracies related to the circuit board model and thus enables precise predictions to be made about conductivity. At the same time, the machine learning model enables the processability of the circuit boards to be taken into account.

One idea of the above method is a hybrid machine learning model that combines a circuit board model for estimating a thermal conductivity with a probabilistic conductivity model, e.g. a conductivity Gaussian process model, for modeling the inaccuracy in the determination of the thermal conductivity by the circuit board model and a probabilistic processability model, e.g. a processability Gaussian process model for evaluating the processability of the material. In this way, an optimal solution is achieved with little experimental effort. The material composition of the printed circuit board is optimized using a Bayesian optimization process. Since a larger number of material parameters can be optimized with the aid of the Bayesian optimization method than can be done manually with justifiable effort, the quality of the circuit board material in terms of thermal conductivity and processability can be significantly improved.

Furthermore, the acquisition function can be optimized based on the probabilistic conductivity model with a secondary condition, the secondary condition providing that the processability of a circuit board with respect to the test material parameter set to be determined should have a minimum processability specified in particular by a processing threshold value.

In particular, an estimation uncertainty of the processing Gaussian process model, in particular a standard deviation of the expected processing measure, can be taken into account when evaluating the secondary condition.

According to one embodiment, the predefined conductivity model can correspond to a non-stochastic mathematical model that approximates the dependency of the thermal conductivity of a circuit board on one or more of the material parameters with which the circuit board in question is formed.

In particular, the conductivity difference can indicate a difference, in particular a difference, between a conductivity determined by the predefined conductivity model and a thermal conductivity of a printed circuit board that is manufactured by a material composition determined by the respective material parameters.

• - die Art eines Matrixmaterials,
• - die Art eines oder mehrerer Füllstoffe,
• - den Anteil des Matrixmaterials,
• - den Anteil eines oder mehrerer der Füllstoffe,
• - der Größe der Füllstoffpartikel,
• - der Größenverteilung der Füllstoffpartikel,
• - der Form der Füllstoffpartikel.

Furthermore, the material parameters can be selected from one or more of the following parameters:

• - the type of matrix material,
• - the type of one or more fillers,
• - the proportion of the matrix material,
• - the proportion of one or more of the fillers,
• - the size of the filler particles,
• - the size distribution of the filler particles,
• - the shape of the filler particles.

Provision can be made for the material parameter set to be selected depending on a predetermined confidence level, so that only one of material parameter sets is selected in which the sum of the expected value of the probabilistic processability model and the maximum limit of a tolerance band that indicates that the actual value of the Processability with a probability specified by the confidence level within the tolerance band is greater than the specified minimum processability.

• - ein Bayes'schen Optimierungsverfahren basierend auf einem probabilistischen Leitfähigkeitsmodell als Qualitätsfunktion durchzuführen, wobei das probabilistische Leitfähigkeitsmodell eine Abhängigkeit von Leitfähigkeitsunterschieden zu einem vorgegebenen Leitfähigkeitsmodell abhängig von Materialparametern nachbildet, wobei während des Bayes'schen Optimierungsverfahrens Leiterplatten, die entsprechend Test-Materialparametern gefertigt werden, hinsichtlich ihrer thermischen Leitfähigkeit und ihrer Verarbeitbarkeit vermessen werden, wobei das probabilistische Leitfähigkeitsmodell und ein probabilistisches Verarbeitbarkeitsmodell mit der jeweils gemessenen thermischen Leitfähigkeit und der jeweils gemessenen Verarbeitbarkeit trainiert oder aktualisiert werden, wobei eine Akquisitionsfunktion basierend auf dem probabilistischen Leitfähigkeitsmodell unter Berücksichtigung des probabilistischen Verarbeitbarkeitsmodells optimiert wird, um jeweils einen nächsten Test-Materialparametersatz zur Vermessung der thermischen Leitfähigkeit und der Verarbeitbarkeit einer entsprechend hergestellten Leiterplatte zu erhalten,
• - Einen Materialparametersatz für die Materialzusammensetzung der herzustellenden Leiterplatte auszuwählen, um entsprechend des probabilistischen Leitfähigkeitsmodells eine vorgegebene oder maximierte thermischen Leitfähigkeit zu erhalten.

According to a further aspect, a device for determining a material composition of a printed circuit board to be produced is provided, the material composition being determined by a material parameter set with material parameters, the device being designed to:

• - Carry out a Bayesian optimization method based on a probabilistic conductivity model as a quality function, the probabilistic conductivity model simulating a dependency of conductivity differences to a given conductivity model depending on material parameters, with printed circuit boards manufactured according to test material parameters during the Bayesian optimization method, are measured with regard to their thermal conductivity and their workability, the probabilistic conductivity model and a probabilistic workability model being trained or updated with the respective measured thermal conductivity and the respective measured processability, with an acquisition function based on the probabilistic conductivity model being optimized taking into account the probabilistic workability model, in order to obtain a next test material parameter set for measuring the thermal conductivity and the processability of a correspondingly manufactured circuit board,
• - Select a set of material parameters for the material composition of the printed circuit board to be manufactured in order to obtain a predetermined or maximized thermal conductivity according to the probabilistic conductivity model.

Figure list

• 1 eine schematische Darstellung eines Funktionsblockdiagramms zur Veranschaulichung des Verfahrens zur Optimierung der Materialzusammensetzung für die Herstellung einer Leiterplatte;
• 2 ein Funktionsblockdiagramm zur Veranschaulichung der Funktion des Bestimmens eines Materialparametersatzes für eine Materialzusammensetzung einer Leiterplatte; und
• 3 ein Flussdiagramm zur Veranschaulichung der Verfahrensschritte zum Ermitteln der Materialzusammensetzung für die Herstellung einer Leiterplatte.

Embodiments are explained in more detail below with reference to the accompanying drawings. Show it:

• 1 a schematic representation of a functional block diagram to illustrate the method for optimizing the material composition for the production of a circuit board;
• 2 a functional block diagram to illustrate the function of determining a material parameter set for a material composition of a circuit board; and
• 3 a flow chart to illustrate the method steps for determining the material composition for the production of a printed circuit board.

Description of embodiments

1 shows a schematic representation of an optimization system 1 to determine an optimized material composition for the production of a printed circuit board. The optimization system 1 has a control unit 2 that carries out the method described below and a manufacturing facility 3 drives. The manufacturing facility 3 is designed to provide a mixture of materials when material parameters are specified and to manufacture a printed circuit board from the corresponding material.

With the help of a test facility 4th the thermal conductivity and the processability of the printed circuit board can be determined. The resulting test results are sent as the measurements of thermal conductivity and processability to the control unit 2 transmitted so that it can optimize the material parameters that determine the material composition.

The aim of the method described below is to provide the material composition for the production of a printed circuit board. These circuit boards should have good processability and the highest possible thermal conductivity in order to dissipate heat from hotspots during operation of the applied electronics. Depending on the application of circuit board electronics, such as data processing or power electronics, the requirements for thermal conductivity for circuit boards are different. Since the processability usually decreases with better thermal conductivity, it makes sense to adapt the material composition of the circuit board to the application.

• - die Art eines Matrixmaterials,
• - die Art eines oder mehrerer Füllstoffe,
• - den Anteil des Matrixmaterials,
• - den Anteil eines oder mehrerer der Füllstoffe,
• - der Größe der Füllstoffpartikel,
• - der Größenverteilung der Füllstoffpartikel,
• - der Form der Füllstoffpartikel

In this regard, the method described below provides for a circuit board to be made available for a given minimum processability, which has an optimal conductivity with regard to the type of circuit board electronics to be applied. The material composition of the circuit board is determined from a number of material parameters that can be selected, for example, from one or more of the following parameters:

• - the type of matrix material,
• - the type of one or more fillers,
• - the proportion of the matrix material,
• - the proportion of one or more of the fillers,
• - the size of the filler particles,
• - the size distribution of the filler particles,
• - the shape of the filler particles

The method for optimizing the material parameters is described below using the function diagram of 2 and the flow chart of 3 explained in more detail.

In block 11 or in step S1 a set of material parameters is initially provided for which an evaluation of a thermal conductivity and the processability of a printed circuit board manufactured with it is to be given.

In the block 12 or in step S2 a deterministic expert model for thermal conductivity is applied to the set of material parameters provided and a modeled thermal conductivity is determined.

Simultaneously is in block 13th and in step S3 Using the set of material parameters, a circuit board material and then a circuit board are produced.

In block 14th or in step S4 a thermal conductivity is determined in a first measurement method based on the circuit board.

In block 15th or in step S5 a measure for the processability of the circuit board is determined in a second measuring method. This measure should indicate a variable value to indicate a degree of processability, for example in a range of values between 0 and 1, where 0 can indicate unprocessability and 1 can indicate the best possible processability. The thermal conductivity and the processability are made available to the computing unit.

The result of the calculation of the conductivity model is a modeled thermal conductivity. The one in a difference block 16 The difference formed between the modeled thermal conductivity and the measured thermal conductivity is used in a probabilistic conductivity model, in particular in a conductivity Gaussian process model 17th , in one step S6 retrained over the set of material parameters as an input variable or updated with the last measured data point and the measured thermal conductivity.

Furthermore, a workability Gaussian process model is based on the workability measure 18th in step S7 retrained or updated depending on the set of material parameters as inputs.

Basically, the problem is to find material parameters that lead to an improved or optimized circuit board material, i.e. a material composition with the highest possible thermal conductivity and processability that meets a minimum requirement. For this purpose, a quality function (cost function) dependent on the material parameters is evaluated.

In general, Bayesian optimization is used when an unknown function f, a so-called “black box” function, is to be optimized. This unknown function f can only be evaluated and observed (possibly affected by noise) for a value x. The observed value y results as y = f (x) + e, where e denotes the noise. It is also assumed that every evaluation or measurement of the unknown function f is expensive, i.e. Caused costs, in the sense that the implementation of a test procedure for measuring the unknown function causes a lot of effort, as is e.g. is the case when a test procedure is carried out in the test facility. Due to the "expensive" measurement of the unknown function, it is desirable that only a few measurements have to be carried out for the optimization or that the costs for the measurements (especially determined by the time and material expenditure) are as low as possible.

Under certain assumptions, e.g. the continuity of the unknown function, the quality function can be approximated with a Gaussian process regression in a Gaussian process model or a function model. A Gaussian process is a universal function approximator that is used as a surrogate function for the unknown quality function. Usually numerical optimization algorithms are based on many, very cheap evaluations / measurements of the optimization functions. However, if the function evaluations are complex, such as If, for example, one of the above test procedures is carried out, the conventional optimization algorithms can no longer be used. Instead, the Bayesian optimization method is used, which includes modeling of the quality function in the form of a surrogate function. The quality function describes the behavior of the system and gives a quality value depending on the parameters with which the system is operated.

For this purpose, after measuring the quality function at several evaluation points, i.e. Test material parameters and observation of the corresponding functional values, i.e. the respective quality value (thermal conductivity) a model of the quality function can be set up using the Gaussian process. A property of the Gaussian process is that in areas in the vicinity of the measured test material parameters, the model prediction is very good and the quality function is approximated well. This is reflected in a low level of uncertainty in the functional model. Far away from evaluation points, the model predictions about the quality function become poor and the uncertainty increases with increasing distance from the measured test material parameters.

A possible strategy to optimize the quality function is to evaluate the quality function in many different places (e.g. on a regular grid) and to take the lowest observed function value as the result of the optimization. This procedure is inefficient and many measurement processes with the test procedures are necessary with a correspondingly high level of effort to find the optimum.

If several quality functions are relevant for the given optimization task, such as thermal conductivity and processability, several Gaussian process models can be used.

Instead of this approach, a Gaussian process model is used to select new test material parameters.

This is done in block 19th or in step S8 In each recursion, a new test material parameter set for measuring the quality function is selected in such a way that, on the one hand, this improves the Gaussian process model of the quality function, so that the uncertainty of the Gaussian process is reduced. For this purpose, the test material parameters are usually selected in areas in which the quality function has not yet been evaluated (exploration). On the other hand, the new test material parameters for measuring the quality function are selected in such a way that the goal of optimizing the quality function, i.e. to maximize it, in particular to maximize the thermal conductivity, is as quick as possible or with the lowest possible number of measurements with the test Material parameters is achieved. For this purpose, a test material parameter set is preferred which, based on the conductivity Gaussian process model, promises high functional values (exploitation). These two opposing criteria are weighed up using a so-called acquisition function.

In the Bayesian optimization method, the measurements for determining the quality function are optimized with the aid of the acquisition function in such a way that they do not necessarily have the lowest uncertainty overall, but rather the highest possible information about the position of the optimum, i.e. H. have the material parameter set at which the highest possible quality value can be achieved.

The acquisition function generally uses parameters of the quality function, which is described by a Gaussian process model, e.g. the Gaussian process mean µ (x) and the Gaussian process standard deviation σ (x). An example is the so-called Lower Confidence Bound (LCB) acquisition function or the Upper Confidence Bound (UCB) acquisition function, which are described as follows: LCB (x) = µ (x) - kσ (x) or UCB ( x) = µ (x) + kσ (x). The factor k is often constant in practice e.g. set to a certain value, such as k = 2. This new criterion can be efficiently minimized or maximized with common gradient-based methods and the minimum of LCB (x) or the maximum of UCB (x) then forms the new test material parameters for the quality function. It should be noted here that an optimization domain must be defined for the optimization of the acquisition function, in which the next test material parameters are searched for. This domain is typically chosen based on experience and / or expert knowledge.

Besides the above acquisition functions, other acquisition functions are known, such as e.g. Expected Improvement (EI), Probability of Improvement (PI) or so-called entropy search methods that are based on information-theoretical considerations.

In the present case, the acquisition function first takes into account the sum of the modeled thermal conductivity and the thermal conductivity determined by the conductivity Gaussian process model as follows: UCB (x) = + ƒ _phys (x) + µ (x) + kσ (x) and maximized this to determine the next test material parameter set. The optimization takes place under a secondary condition that takes into account the processability Gaussian process model. In particular, the above Upper Confidence Bound UCB can be maximized under the constraint that the expected processability according to the processability Gaussian process model exceeds a specified processability threshold of, for example, 0.9 (the processability being better the higher the amount of processing). In particular, the processability forecast error can be taken into account by reducing the expected processability by the corresponding standard deviation (estimation error from the processability Gaussian process model), so that the secondary condition criterion compares the value reduced by the standard deviation of the expected processability with the specified processability threshold.

After optimizing the acquisition function, the next test material parameter set is available, with which a new measurement can be carried out.

A new set of material parameters is selected by applying an acquisition function to the sum of the modeled thermal conductivity and the conductivity difference determined by the conductivity Gaussian process model under the constraint that the expected processability minus a term dependent on its standard deviation is greater than a specified processability threshold.

In step S9 a termination criterion is checked. The termination criterion can relate to the length of time that is to be expended for the optimization of the quality function, or to the number of iterations or to a suitable convergence criterion. If the termination criterion is met (alternative: yes), the optimization should be terminated and the procedure is started with step S10 continued, otherwise (alternative: no) the procedure is continued with step S1 continued.

In step S10 a material parameter set for the material composition of the printed circuit board to be manufactured is selected based on a predetermined minimum permissible thermal conductivity. In particular, the material parameter set can be selected by maximizing the thermal conductivity.

The material parameter set can be selected from the set of material parameter sets for which the sum of the expected value of the conductivity Gaussian process model and the functional value of the predefined conductivity model is greater than the predefined minimum thermal conductivity.

According to a further embodiment, the material parameter set can be selected as a function of a predetermined confidence level, so that only one material parameter set is selected in which the sum of the expected value of the conductivity Gaussian process model and the functional value of the predetermined conductivity model and the maximum limit of a tolerance band around the expected value of the conductivity -Gauss process model, which indicates that the actual value of the thermal conductivity is within the tolerance band with a probability predetermined by the confidence level, is greater than the predetermined minimum thermal conductivity.

Then in step S11 a printed circuit board is produced with a material composition which is indicated by the selected material parameters.

Claims (11)

A method for determining a material composition of a printed circuit board to be produced, the material composition being determined by a material parameter set with material parameters, with the following steps: - Carrying out (S1-S9) a Bayesian optimization method based on a probabilistic conductivity model, in particular a conductivity Gaussian process model or a Bayesian neural network model, as a quality function, the probabilistic conductivity model being dependent on conductivity differences from a given conductivity model Simulates material parameters, During the Bayesian optimization process, printed circuit boards that are manufactured according to test material parameters are measured with regard to their thermal conductivity and their processability, wherein the probabilistic conductivity model and a probabilistic processability model, in particular a processability Gaussian process model or a processability Bayesian neural network model with the respectively measured thermal conductivity and the respectively measured processability are trained or updated, wherein an acquisition function based on the probabilistic conductivity model and the probabilistic processability model is optimized in order to obtain a respective next test material parameter set for measuring the thermal conductivity and the processability of a correspondingly manufactured circuit board, and - Selecting (S10) a material parameter set for the material composition of the printed circuit board to be manufactured in order to obtain a predetermined or maximized thermal conductivity in accordance with the probabilistic conductivity model. Procedure according to Claim 1 , wherein an acquisition function based on the probabilistic conductivity model is optimized taking into account the probabilistic processability model. Procedure according to Claim 2 , wherein the acquisition function is optimized based on the probabilistic conductivity model with a secondary condition, the secondary condition providing that the processability of a circuit board with respect to the test material parameter set to be determined should have a minimum processability specified in particular by a processing threshold value. Procedure according to Claim 3 , whereby an estimation uncertainty of the probabilistic processability model, in particular a standard deviation of an expected processing measure, is taken into account when evaluating the secondary condition. Method according to one of the Claims 1 to 4th , wherein the specified conductivity model corresponds to a non-stochastic mathematical model that approximates a dependency of the thermal conductivity of a circuit board on one or more of the material parameters with which the circuit board in question is formed. Procedure according to Claim 5 , wherein the conductivity difference is a difference, in particular a difference, between a conductivity determined by the predetermined conductivity model and an actual thermal conductivity Indicates printed circuit board that is manufactured by a material composition determined by the respective material parameters. Method according to one of the Claims 1 to 6th , the material parameters being selected from one or more of the following parameters: - the type of a matrix material, - the type of one or more fillers, - the proportion of the matrix material, - the proportion of one or more of the fillers, - the size of the filler particles, - the size distribution of the filler particles, and - the shape of the filler particles. Method according to one of the Claims 1 to 8th , wherein the material parameter set is selected as a function of a predetermined confidence level, so that only one of material parameter sets is selected in which the sum of the expected value of the probabilistic conductivity model and the functional value of the predetermined conductivity model and the maximum limit of a tolerance band that indicates that the actual The value of the thermal conductivity with a probability predetermined by the confidence level is within the tolerance band, is greater than the predetermined minimum thermal conductivity. Device, in particular a control unit (2), for determining a material composition of a printed circuit board to be produced, the material composition being determined by a material parameter set with material parameters, the device being designed to: - to carry out a Bayesian optimization method based on a probabilistic conductivity model as a quality function, the probabilistic conductivity model simulating a dependency of conductivity differences to a given conductivity model depending on material parameters, During the Bayesian optimization process, printed circuit boards that are manufactured according to test material parameters are measured with regard to their thermal conductivity and their processability, where the probabilistic conductivity model and a probabilistic processability model are trained or updated with the respective measured thermal conductivity and the respectively measured processability, wherein an acquisition function based on the probabilistic conductivity model is optimized, taking into account the probabilistic processability model, in order to obtain a next test material parameter set for measuring the thermal conductivity and the processability of a correspondingly manufactured circuit board, to select a set of material parameters for the material composition of the printed circuit board to be manufactured in order to obtain a predetermined or maximized thermal conductivity in accordance with the probabilistic conductivity model. Computer program with program code means which is set up to implement a method according to one of the Claims 1 to 8th execute when the computer program is executed on a computing unit. Machine-readable storage medium with a computer program stored thereon Claim 10 .

Images