Patent Application
20100280646
The invention relates to methods for the production and quality control of bonded connections, in particular wire bonded connections, by means of ultrasound, and to components suitable for carrying out these methods as well as to bonding apparatuses equipped with components of this kind, in particular wire bonders. To this extent, the invention also relates in particular to wedge-wedge bonding. The overall aim of the invention is in particular that of improved productivity and improved reliability of the quality control in comparison with conventional techniques.
Wedge-wedge bonding is a generally recognized and reliable technique for establishing electrical contact between a chip and a substrate. It is used both in power electronics and in microelectronics. To monitor the quality of the bonding process, a specific percentage of the products are taken during the course of production and subjected to a destructive test, and the quality of the overall production batch is assessed by means of statistical methods.
The desire for further quality control, in particular in areas where safety is an issue, has led to the development in recent years of several methods that allow in-line monitoring to be performed in addition to the destructive test specified above. Broadly speaking, two methods should be mentioned, on the one hand the nondestructive mechanical testing of a bonded connection, in which low forces act on the connection (tensile or shearing forces), in order to test its strength, on the other hand a method based on the measurement of signals that can be obtained directly from the ultrasonic generator or the position sensor of the transducer.
The second method uses as measured variables the current flowing in the transducer, the voltage at the output of the ultrasonic generator or the deformation of the wire, or a combination of these variables.
Serving as correcting variables for defining a process window are the bonding force, the ultrasonic power and the bonding time. These methods are described in the patents U.S. Pat. No. 4,606,490, EP 0 368 533 B1 and EP 0 540 189 B1. An essential feature of these methods is that the variables measured vary with a profile that follows a characteristic curve predetermined by the user. A weakness of this method is that even poor bonded connections may exhibit such deformation and current profiles. Consequently, the set procedure specified is in any event necessary, but in no way sufficient for determining the quality of a bond, and is therefore unsuitable for detecting with certainty a bond of inferior quality.
An attempt to rectify this shortcoming can be found in the patents EP 1 023 139 B1 and EP 1 342 201. In the earlier methods, only one measured variable was considered and taken as a basis as a measure of quality. In recent methods, it is attempted to combine the aforementioned variables and derive quality criteria from them. Instead of the current, here the electrical admittance of the transducer is used as a variable that is to the greatest extent independent of the ultrasonic power. It is attempted to use the combination of the admittance profile and the deformation of the wire, and the correlation between them, to achieve a more reliable finding as to the quality of a connection.
These methods do not achieve the aim either, since the measured variables used do not correlate sufficiently with the physical processes taking place in the connecting zone. For example, the electrical resonance (phase angle between current and voltage=zero) does not correspond to the mechanical resonance (phase angle between the speed of the tip of the tool and the voltage=zero), so that systems which only use the electrical variables are not operated with mechanical resonance. Neither the speed of the tip of the tool nor the friction between the parts being connected, that is to say the physical variables that are of decisive importance directly for the formation of the connection, are recorded by the aforementioned methods.
The use of a sensor mounted in the membrane is described in the applicant's EP 1 789 226 A1 and WO 2006/032316 A1.
The described methods that have so far become known are therefore not suitable for acquiring the data essential for assessing the quality of a connection and processing them correspondingly. To this extent, these methods cannot satisfy the requirements for quality monitoring in fully automatic bonders.
According to a first aspect, the invention is based on a method for quality control during ultrasonic bonding, in particular ultrasonic wire bonding, in which a transducer/bonding tool unit and an ultrasonic generator are used and in which, during the bonding, measurement signals for one or more parameters that can vary during the bonding are sensed by means of one or more sensors for assessing the quality of the bond and/or for influencing the bonding, in particular by means of controlling or regulating the bonding process.
Against the background of the difficulties explained at the beginning with respect to the prior art, it is to this extent an object of the invention firstly to provide improved possibilities for quality control, in particular to include further physical variables that are essential for, or informative about, the formation of a connection.
The solution achieving this object according to the invention is based on the recently discovered realization that, in particular, the speed of the tip of the wedge and the friction between the parts to be connected are physical variables or parameters of the bonding process that are essential for the formation of a connection, and to this extent are informative for the assessment of the bonding operation, and that these variables or parameters should therefore be extracted from a sensor signal and made available to an assessment unit. This assessment unit can then be based on a suitable model as a reference, on the basis of which the assessment of the process data is possible.
To achieve the object, the invention proposes first and foremost the method steps that, during the bonding, at least one speed profile measurement signal that is representative, in particular qualitatively or with respect to the phase profile, of the time/speed profile of the tip of the ultrasonic tool in the direction of oscillation thereof is sensed. To this extent, the time/speed profile of the tip of the tool or wedge represents a parameter (bonding parameter) which, as a measured variable or as a variable parameter of the bonding operation, allows observation thereof. The measurement signals generated from it can be used as input variables for a subsequent evaluation, in particular for assessing the quality of a bond, and/or for controlling and/or regulating the bonding operations at the time and/or subsequently. There is the possibility that one or more further input signals from various sources are additionally included. All variables relevant to the process come into consideration, for example from the kinematics the wire deformation, from the ultrasound system the frequency, the phase difference between current and voltage and the ultrasonic current, and from additional sensors, for example the mechanical ultrasonic oscillation and the phase between wedge speed and ultrasonic current. The parameters or measured variables may be recorded continuously or virtually continuously or at a suitable clock rate. The numerical values thereby determined for a respective measured variable during the production of a bonded connection, i.e. during a bonding operation, may be combined as a column of values to form a vector (feature vector), which is available for further evaluation and use. The values thereby recorded in a time interval, and optionally grouped together to form a vector, are also referred to hereafter in their sequence over time as profiles. It is considered to be suitable if the recording or resolution of the time/speed profile of the tip of the tool in the direction of oscillation thereof is in such fine time increments that the time profile generally made up of sine and/or cosine components is thereby reproduced. To this extent unequivocal recording can be performed in real time for these and, if need be, further measured variables, preferably using FPGA-based electronics, which can be connected to a DDS ultrasonic generator by means of an additional interface.
There is also the possibility of deriving further process-relevant variables from one or more of the aforementioned signals determined with respect to measured variables or bonding parameters. In particular, there is the possibility of using the sensor signals determined from the speed profile of the tip of the tool to obtain derived variables which are associated with or represent the friction between the tip of the tool or wedge and the wire, the friction between the wire and the substrate, the coupling of the wire to the substrate and the deflection or speed of the tip of the tool or wedge. The generation of derived signals from measurement signals may be carried out for example by processing, and in particular transformation, of one or more measurement signals (for example also filtering) and/or by computational combination of a number of measurement signals (for example two, three, etc.). Also coming into consideration as a derived variable, for example, is the mechanical admittance as a quotient of the two measured variables, tool speed and generator voltage. Other examples are the electrical admittance as a quotient of the current and voltage of the generator as well as the electrical impedance as the inverse value of the admittance.
Furthermore, variables based on similarity functions (texture comparisons) represent further possible input variables, in particular also for the calculation of a quality index, in particular according to further aspects of the invention. Each of these input variables represents a vector, the length of which is dependent on the recording duration, the sampling rate and optionally applied pre-processing operations.
According to a preferred development of the method according to the invention, it is preferred that the speed profile measurement signal is recorded by means of a piezo sensor, which is preferably disposed in/on a mounting of the ultrasonic transducer and by means of which it is possible in particular to measure technically a transverse extension perpendicular to a propagating ultrasonic excitation wave. A piezo sensor may preferably be used for this purpose, such as that described in the applicant's WO 2006/032316 A1. The content of this document is hereby incorporated in full in the present application, also for the purpose of including one or more of the features described therein in claims of the present application. Investigations with such a sensor have surprisingly found that the speed of the tip of the tool or wedge, measured with a laser interferometer for purposes of comparison, coincides with the sensor voltage with respect to the phase position. On the basis of this realization, it is proposed according to a development of the method according to the invention that the time/generator voltage profile fed to the ultrasonic transducer from the ultrasonic generator is measured, that the phase difference between the time/generator voltage profile and the speed profile measurement signal is determined, for example by means of a phase comparator, and that the oscillation frequency of the ultrasonic tool is set or changed by means of the ultrasonic generator in such a way that said phase difference is reduced, preferably to substantially or precisely zero. This succeeds in producing a state of mechanical resonance, which offers considerable advantages in the production of bonded connections. This is so on the one hand because, in this state, there is an optimum transmission behavior of the piezo oscillation to the welding tool and on the other hand only particularly small losses occur. According to the invention, there is the possibility of approaching the state of mechanical resonance in a controlled manner. This was not possible in the past, for one reason because there was no possibility of measuring the time/speed profile at the bonding apparatuses, and to this extent a corresponding use was also unknown. Instead it was attempted in the past, though only with limited success, to improve the transmission behavior and losses by the setting of electrical resonance. This involved using a phase comparator to determine the phase difference between the transducer current and the transducer voltage and using a phase regulator for adjustment involving changing the oscillation frequency to zero. However, the state of electrical resonance does not generally coincide with the desired state of mechanical resonance, so that the setting thereof was not possible in a controlled manner. It was then found that, with the tool oscillating, the speed profile is so similar to the measurement signal of the piezo sensor up to high spectral components that the mechanical resonance can be set with great accuracy. A PLL, i.e. phase locked loop or closed loop, known per se, may be used for example for regulating the phase difference to the value zero.
Seen as a further suitable development of the method according to the invention is that, during individual bonding operations or after the bonding, friction value/time profiles are determined by means of the speed profile measurement signal as actual profiles or actual vectors of the friction that characterize the variation of the frictional forces changing during the duration of individual bonding operations. The determination of the friction value/time profiles on the basis of the speed/time profile can be performed in a manner with which a person skilled in the art is in fact quite familiar. It may involve determining a sequence of friction values each of which is associated with a specific point in time of a bonding operation and has, for its part, been derived in particular from a greater number of measurement signals of the tip of the tool. The “friction value signal” obtained, or the time profile thereof, may be understood as an aggregate signal during the bonding operation between the bonding tool, the bonding wire and for example the substrate (cf. also the statements made above). The friction signals obtained during a bonding operation, for example, may also be treated for example as a vector for further processing. Correspondingly, vectors for speed values and for friction values and for values derived therefrom, such as for example the admittance, can be formed. According to the invention, there are various possibilities for using these profiles according to requirements for influencing the subsequent bonding and/or for assessing bonding results (quality of a bond). There is the possibility of using the friction value determined in the case of a process state of the bonding operation at a particular time (in the sense explained above as a derived variable) for controlling and/or regulating the further bonding operation and/or subsequent bonding operations, preferably in the course of influencing manipulated variables, such as for example the bonding force, ultrasonic power, bonding time and/or ultrasonic frequency. Alternatively or in combination, there is the possibility of determining a quality value characterizing the quality of a bonding operation or a bonded connection from the friction value/time profile and a setpoint friction value/time profile that is predetermined or a setpoint friction value/time profile that is determined and stored beforehand by means of a computer in a learning phase, preferably according to or by analogy with features explained below. For example, the setpoint profile of the derived friction may also be represented as a vector and compared in its elements step by step with temporally associated elements of the actual vector. For example, a deviation vector may be formed from the respective differences. With respect to the possibility of generating setpoint value profiles of measured variables and/or derived variables in a preceding learning phase, reference is also made in this connection to the description hereinbelow. There is also the possibility of using said quality value for controlling and/or regulating subsequent bonding operations, preferably in the course of influencing manipulated variables such as the bonding force, ultrasonic power, bonding time and/or ultrasonic frequency. A further preferred use may be that the quality value is used for influencing manipulated variables of the bonding process, such as in particular the bonding force, ultrasonic power, bonding time and/or ultrasonic frequency in the course of controlling or regulating and/or for the emission of warning signals when critical values are reached. It has been found that the profiles of the tool speed in the direction of oscillation occurring during individual bonding operations and profiles derived therefrom (in particular the derived friction) can, depending on the bonding conditions, make it possible to provide a more reliable finding about the quality of a bonding operation in comparison with conventional evaluation parameters (in particular the wire deformation). The method according to the invention therefore makes more precise monitoring of the quality of a bond possible in comparison with the prior art, and if need be (for example if the bonding conditions change), improved influencing of manipulated variables to obtain consistently good quality of the bond. In particular, there is the possibility of integrating a setpoint/actual comparison of the derived friction in a regime of controlling and/or regulating bonding processes, with the further possibility that additional parameters could also be taken into consideration.
As far as the previously explained first aspect is concerned, the invention also includes a bonding apparatus, preferably a wire bonder, for the production and quality control of ultrasonic bonded connections, having a transducer/bonding tool unit and an ultrasonic generator as well as at least one sensor for obtaining measurement signals for at least one parameter that can vary during the bonding. The invention proposes that the bonding apparatus has at least one sensor which is suitable for generating a speed profile measurement signal representing the time/speed profile of the tip of the ultrasonic tool in the direction of oscillation thereof. For preferred developments in this respect, reference is also made to the features of claims 9 to 14 and, for possible advantages and effects, reference is made to the description as a whole and the figures. The invention of course also includes in this connection an ultrasonic generator, which can regulate the phase between the speed of the tip of the wedge and the ultrasonic voltage, so that the system operates in phase zero at its mechanical resonance.
On the basis of the relationships and features described above, the invention is consequently also concerned in its first aspect with the generation of process-integrated quality monitoring modules, in particular using friction and mechanical admittance (quotient of the speed of the tip of the wedge and the voltage of the generator), preferably with specifically predetermined reference data (setpoint data). In this case there is the possibility of processing external sensor values. There is also the possibility of operating the transducer-wedge system at its mechanical resonance or optionally at electrical resonance. Derived variables for the friction and the wedge speed can be determined from measured variables.
According to a second aspect, the invention relates to a method for quality control during ultrasonic bonding, in particular ultrasonic wire bonding, in which a transducer/bonding tool unit and an ultrasonic generator are used and in which, during the bonding, measurement signals for one or more parameters that can vary during the bonding are sensed, in particular for assessing the quality of the bond and/or for influencing the bonding.
On the basis of the prior art described, it is the object of this further aspect of the invention to develop methods of the stated type advantageously, so that in particular a more accurate and reliable assessment of the quality of a bond is made possible.
The object is achieved according to the invention first and foremost in conjunction with the features
The determination of individual quality indices Qi, which is also referred to within the scope of the invention as feature extraction (cf. also the figures), makes it possible to monitor the quality of a bond in a particularly clear and informative way and even, as also explained below, furthermore control and/or regulate the process of producing bonded connections (i.e. the so-called bonding). Within the scope of the invention, multi-variable appraisal or multi-variable monitoring is preferred.
The actual profiles, setpoint profiles and deviation profiles may in principle or in general, i.e. also in connection with the other aspects of the invention, again be, for example, one-dimensional vectors or sequences of values, with values preferably sorted in the time sequence of the underlying measurement signals. To this extent it is also possible to speak in terms of vectors (feature vectors) instead of profiles, that is to say to speak of actual vectors, setpoint vectors, deviation vectors, etc. In a simplified form, they may be understood as columns of values, the number of their numerical values depending, inter alia, on the clock rates used in the measurements and the duration of the bonding operations or measurements. It is in this case preferred that actual vectors and setpoint vectors have the same dimension or length, so that pairs of values of two vectors that are associated with one another in the sequence of steps or at the measuring time (that is temporally), can be respectively evaluated in a particularly clear way for generating values of the deviation vector.
As described above, the invention proposes the generation of feature profiles or feature vectors from measured and/or derived variables. In the calculation of an individual quality index Qi, the underlying feature vector or the deviation vector thereof is preferably converted into a scalar variable. If the number of feature vectors or profiles on which the quality calculation is based is denoted by n, this is a mapping of n feature vectors that may have the same or different dimensions onto a feature vector (result vector) of the dimension n. Each element of the result vector may in this case preferably correspond as a scalar variable to the value of the respectively underlying feature vector.
It is preferably provided that, in the calculation of the quality index Q, at least some of the deviation profiles are weighted individually, and in particular independently of one another, in particular according to information previously stored in a memory. The accuracy of the quality assessment can also be increased in particular by one or more deviation profiles being variably weighted temporally or in their variation (i.e. with respect to the various elements of the deviation vector) during the calculation. The method according to the invention may preferably be performed in such a way that an individual quality index Qi is respectively determined by means of suitable computational means from individual deviation profiles, in particular with individual weighting temporally or in terms of variation, according to stored information and that the quality index Q is calculated from a number of individual quality indices Qi according to a stored algorithm.
These method steps are based on the realization now discovered that different bonding parameters (or variables derived from them) or the measured time profiles thereof may be attributed various levels of significance in the assessment of the quality of a bond with respect to one another and also individual bonding parameters (or variables derived therefrom) within different time intervals of a bonding operation. This means that a model which only takes into consideration individual bonding parameters that are assumed to be significant, but does so over the entire bonding duration, or a model which in principle constantly takes into consideration all technically measured parameters may not be sufficient for an exact quality assessment, depending on conditions of the bonding process and the disturbing influences acting on it. The method according to the invention thus makes it possible to integrate into an automated method for quality control, or into a bonding apparatus suitable for carrying out said method, findings as to which bonding parameters may have a greater or lesser significance in which time periods within the bonding duration. Corresponding findings and relationships may be determined in tests and then archived, for example in a database, an expert system or the like. If bonded connections are then to be produced later under corresponding bonding conditions, i.e. for the same reference system, the archived information can be loaded into a main memory, so that a quality assessment that is tailored to this extent is possible. For example, it may be advantageous under certain preconditions to give the time profile of the wire deformation greater significance toward the end of a bonding operation. Similarly, it would be conceivable, for example, to give a profile representing the varying friction a greater weighting in an early time interval of the bonding phase in comparison with a later time interval. Of course, there is also the possibility if need be of providing certain bonding parameters with the same weighting over the bonding duration. What is decisive is the possibility of being able to weight various bonding parameters individually, and to this extent independently of one another. By analogy, there is also the possibility of using individual weighting of bonding parameters within a regime of controlling or regulating the bonding process in real time for influencing manipulated variables. For example, it would be conceivable for the ultrasonic power to be influenced with greater weighting of a first bonding parameter at the beginning of a bonding operation and with greater weighting of a setpoint/actual comparison for another bonding parameter toward the end of the bonding operation.
According to the above, the method according to the invention also includes the possibility that deviation profiles of individual bonding parameters are only considered in terms of their value at certain time intervals within bonding operations as a result of their individual weighting. This can be achieved, for example, by these parameters being assigned the weighting of zero in other time intervals of the bonding operation. While it is also the case here that the bonding parameters that are always taken into consideration are constantly recorded during the bonding duration, the weighting of individual bonding parameters in the quality assessment may be different during different time intervals. Apart from predetermining the different weightings themselves, there may also be the possibility of individually predetermining the starting points and ending points for the various parameters. As an alternative to or in combination with sudden changes in the weighting of parameters, it is also possible to predetermine weighting functions in which the weighting factor changes in small steps or virtually continuously. This would be possible, for example, by predetermining weighting vectors.
As already mentioned, there is also the possibility of forming derived variables from measurement signals of various parameters, in particular by means of suitable computational means, and in turn determining from the respective actual profiles thereof deviation profiles, by computational comparison with previously stored setpoint profiles of these derived variables, and of these deviation profiles being used for the determination of a quality index that is weighted individually in terms of time. As above, the term comparison should be broadly understood here in the sense of various possibilities of data processing and computational operations. In a simple example, the comparison could comprise a simple subtraction or formation of the difference between pairs of setpoint values and actual values, but other algorithms would also be conceivable. As far as the setpoint profiles predetermined for example by vectors are concerned, there is the possibility of these being specifically predetermined for example in a memory (for example originating from an expert system) or previously determined in a learning phase, preferably according to or by analogy with features subsequently described. For the automation of the method described above, it is also preferred that at least one or more of its steps is/are carried out in a computer-aided manner by means of software. A preferred development is also seen in the possibility of using the determined individual quality indices or the overall quality index for refining the regime of controlling and/or regulating the further bonding operation. Manipulated variables may here again preferably be the bonding force, the ultrasonic power and the bonding time.
As far as its second aspect is concerned, the invention also includes, furthermore, a bonding apparatus, preferably a wire bonder, for the production and quality control of ultrasonically bonded connections, which is formed according to claim 24 as being suitable for carrying out the method described above. For possible preferred developments of the bonding apparatus, reference is also made to the features of claims 25 to 31 and, for possible effects and advantages, reference is made to the description as a whole. The measured variables or parameters, such as current, wire deformation, resonant frequency, wedge speed, phase differences between ultrasonic voltage and current and wedge speed and voltage, as well as the variables derived from the sensor signals, for example admittance or impedance and friction, are compared with a predetermined or learned time profile (setpoint profile). A weighted input variable for the subsequent calculation of the quality index is determined from the deviation of the individual measured variables from the associated setpoint curve or setpoint profile. The weighting of the individual values and the time period in which these values are considered can be set.
According to a third aspect, the present invention relates to a method for quality control during ultrasonic bonding, in which a transducer/bonding tool unit and an ultrasonic generator are used and in which, during the bonding, measurement signals for one or more parameters that can vary during the bonding are sensed for assessing the quality of the bond and/or for influencing the bonding.
Measurements with conventional systems have shown, in particular when using relatively thick wires, that the recorded physical variables vary very greatly in dependence on the bonding surfaces, the substrate materials, the stiffness of the construction, the eigenmodes of the system as a whole, the wedges and wires used, etc. The strongly application-dependent fluctuations of the process variables do not allow the use of characteristic curves as reference data that can be used across all applications in the case of thick wire bonders.
Against this background, it is a further object of the invention to provide a method and an apparatus which can generate suitable reference data (setpoint data) for the assessment of the input data under consideration, or in an analogous sense can learn such data, for a wide variety of applications—in spite of different starting preconditions.
This object is achieved according to the invention first and foremost in conjunction with the features
The sensing of measurement signals by means of the sensors may be carried out continuously, virtually continuously or optionally also at a desired lower clock rate. The chosen term, learning phase, is intended to illustrate that setpoint profiles (setpoint vectors) serving as reference profiles are generated in an automated manner during this phase by means of the bonding method or the bonding apparatus. To this extent, instead of speaking of a learning phase, one could also speak of a generating phase for reference profiles or setpoint profiles.
On the assumption that stable processes that have process parameters ensuring the production of bonded connections of sufficiently high quality exist for the various starting preconditions, the related statistics for each directly measured variable and for each derived variable can be learned. This learning phase serves for generating reference data which serve during later automatic operation as a basis for the calculation of the quality indices. There is therefore no specifically programmed-in characteristic curve for the various processes, but instead the characteristic curves are generated by the system itself under the aforementioned premise.
The same variables as previously described can be used as input variables. On the basis of the statistics learned, deviations can be quantitatively assessed and used for a quality calculation.
It is considered to be a suitable procedure that individual setpoint profiles are respectively generated in separate learning phases for different bonding reference systems, i.e. for systems which differ with respect to the bonding conditions. In the case of a first reference system, this could be, for example, a system with a ceramic substrate, in the case of a second, different reference system, it could be, for example, a connector, in the case of a third reference system, it could be, for example, a chip, and so on. The differentiation between the differing bonding conditions, which can lead to different setpoint profiles, is particularly significant in the case of so-called thick wire bonding, since greater deviations can occur here than in the case of thin wire bonding. To this extent it is also regarded as advantageous if, for a specific ceramic substrate, for example, in combination with different wire thicknesses, once again different reference systems are formed or related individual setpoint vectors are created. In practice, the procedure followed for generating setpoint profiles for a specific reference system may be that, during a first bonding operation, all measured variables of interest for a quality assessment during later automatic operation are recorded in parallel and stored in separate vectors, it also being possible to form vectors for variables derived from the measured variables. This step may then be repeated for a second bonding operation and further bonding operations, a bonding collective possibly comprising, for example, one hundred bonding operations (or some other number). If the measured variable profiles are recorded, for example, at a sampling rate of 50 kHz, and if, by way of example, the rate of a bonding operation is 10 m/sec, this results in five hundred measured values for each measured variable or measured variable profile. Settings of the process parameters of the bonding apparatus (for example the bonding force, ultrasonic power and bonding time) with which the bonding process is expected to function so well for the specific reference system that a majority of the bonded connections produced, in particular a statistically significant proportion, have good quality of the bond, and at most a smaller proportion, in particular a statistically less significant proportion, of the bonded connections are of unsuitable quality should be chosen during the learning phase. Such settings may either be preselected by trained personnel on the basis of empirical values or, for example, be taken from a database, an expert system or the like. During the learning phase, the values for all measured variables entering the setpoint curves or setpoint profiles are recorded time-parallel with one another. In this case there is the possibility in various ways of preparing the measurement signals by means of signal processing components (for example computing units, analogous transmission elements, etc.).
The learned setpoint profiles can be used later in a method for producing bonded connections (production operation or automatic operation) for the quality control of the ultrasonic connections. For this purpose, there is the possibility that, for the production of bonded connections for a specific bonding reference system, the setpoint profiles created during an earlier learning phase are provided, for example from a database, for example they are read into a main memory, that, during the time period of bonding operations, measurement signals for one or more parameters are recorded in each case by means of suitable sensors and respectively kept in a memory as temporal actual profiles and that, for at least one parameter, deviation profiles are determined as so-called error vectors from the actual profile and the setpoint profile learned in the learning phase. This may take place in the way already described, for example by computational comparison of values that are temporally associated with one another from the setpoint and actual vectors. It is considered to be a suitable development that a confidence interval of a specific magnitude about the distribution maximum is predetermined for the statistical model, that in the respective distributions the values at the lower interval boundary and/or upper interval boundary are determined, that a lower characteristic boundary curve is formed from the values of the lower interval boundary and/or an upper characteristic boundary curve is formed from the values of the upper interval boundary and that actual profiles determined during the production of bonded connections are computationally compared with the lower and/or upper characteristic boundary curve. In particular, there is the possibility that, if the upper characteristic boundary curve is overshot and/or the lower characteristic boundary curve is undershot, an error signal is generated and/or is stored in a database and/or is associated with the bonded product by means of an identification for the later segregation or repair of said product and/or a user input is requested by a process controlling or regulating system. Alternatives of this kind may also be provided if, for example, the quality index Q calculated for a respective bonding operation (or one or more individual quality indices Qi) over- or undershoot(s) specific limit values. It goes without saying that said deviation profiles can, if need be, be determined for a number of parameters. There is also the possibility that setpoint profiles for derived parameters are determined in the learning phase from the measurement signals for parameters or from expectation profiles, that, during the subsequent production of bonded connections, actual profiles for said derived parameters are also determined from the measurement signals or from actual profiles of measurement signals and that deviation profiles associated with the derived parameters are also determined from setpoint profiles and actual profiles associated with one another for derived parameters. Said deviation profiles of measured variables and/or derived variables may for their part be used for determining individual quality indices and/or an overall quality index, preferably according to one or more features described with reference to the figures. Also in the case of the method described above, there is the possibility of at least one or more method steps being automatically carried out in a computer-aided manner by means of software.
As far as its third aspect is concerned, the invention also further includes a bonding apparatus, preferably an ultrasonic wire bonding apparatus, for the production and/or quality control of ultrasonic bonded connections, according to claim 42, which is formed suitably for carrying out the method according to the invention according to one or more of the features described in this respect above, or is adapted for this purpose. Features for the preferred development of such a bonding apparatus are also specified in particular in claims 43 to 45. For possible effects and advantages, reference is also made to the description as a whole and the figures.
With respect to the statements made above, the invention is concerned with the provision of a self-learning system, the apparatus and the method being suitable for producing statistics that are dependent on the product and the ambience as a basis for quality calculation. In theory, the quality calculation may include any desired number of input variables that can be provided as measured variables and/or calculated in real time as derived variables of the measured variables, for example by transformation of the measured variables, wavelet transformation, estimation of the variance, etc. There is the possibility of generating process-integrated quality monitoring modules using the input variables described (actual profiles or actual vectors) with reference data (setpoint profiles or setpoint vectors), which have been learned in the corresponding learning phases or automatically generated for the different processes. Quality monitoring modules of this kind may be configured as hardware modules and/or as software modules on a bonding apparatus according to the invention (as also in the case of the other aspects of the invention), it being possible within the scope of various aspects of the invention for modules of this kind also to be the subject of independent patent claims. There is the possibility of taking into consideration a constant number of input vectors. There is also the possibility of generating feature vectors by correlation of different measured variables of identical or different time intervals or by correlation of identical measured variables from different time intervals. There is also the possibility of taking into consideration a variable number of input vectors by feedback from a monitoring unit. The invention finally also reflects the possibility of generating process-integrated quality monitoring modules by using a set of input data extended by a monitoring unit as well as an adaptive quality calculation. There is the possibility of the transformation of an error vector to a scalar quality variable.
According to a fourth aspect, the invention also relates to a method for quality control during ultrasonic bonding, preferably ultrasonic wire bonding, in which a transducer/bonding tool unit and an ultrasonic generator are used and in which, during the bonding, measurement signals for a number of parameters that can vary during the bonding are sensed, preferably for controlling the quality of the bond and/or for influencing the bonding.
On this basis, it is an object of the invention to develop a method of this kind advantageously, so that, in particular, improved feedback of operating states is made possible.
The object is achieved according to the invention first and foremost by the features that, during the time period of bonding operations, measurement signals for a number of parameters, such as preferably the current intensity and/or voltage at the ultrasonic generator or transducer and/or wire deformation and/or ultrasonic frequency or resonant frequency and/or tool speed, are recorded by means of sensors and are respectively provided as temporal actual profiles, preferably stored in a memory, with the possibility that one or more actual profiles for variables derived from the parameter(s) is/are formed from measurement signals for one or more parameters of the bonding process, that a number of actual profiles are respectively subjected to a comparative computing operation with a setpoint profile which is stored in a memory and is associated with the respective actual profile with respect to the parameter or the derived variable, a deviation profile being determined in each case for the actual profiles, preferably by comparison of individual values that are temporally associated with one another, that an individual quality index (scalar variable) is respectively calculated by means of suitable computational means from deviation profiles and that a plurality of individual quality indices, at least of one bonded connection, are compared as a bond index group with a number of different memory index groups stored in a memory, preferably in an expert system, which differ from one another by the values of the individual quality indices associated with specific parameters or derived variables, using at least one predetermined similarity criterion, and, if at least one similarity criterion is satisfied, are associated with a memory index group.
The determination of said individual quality indices is also referred to within the scope of the invention as feature extraction. In a simple application example, an individual quality index can be determined from the values or scalars of the associated deviation vector by forming the value. It is preferred that, when a similarity criterion for a memory index group is satisfied, a preferably electrical classification signal is automatically generated, causing information to be output and/or being used for controlling or regulating a bonding apparatus. It is also suitable that different data concerning the operating state of a bonding apparatus, preferably concerning different errors states, are respectively associated with different memory index groups in a memory and that the generated classification signal or error classification signal causes the output of information concerning this operating state or error state. According to a further aspect of the invention, which may also be of independent importance, there is the possibility of the individual quality indices formed during the feature extraction (which already no longer have any time reference) and/or the quality index and/or classification signals of bonded connections produced during bonding operation being analyzed for aberrant bonds by means of a monitoring device performing computing operations, with consideration for known relationships and statistics. To develop the method, the group of individual quality indices belonging to a aberrant bond may be stored in a memory as a new memory index group and be combined with data concerning the operating state or error state. There is the possibility that, from the aberrant bond, measurement signals, preferably actual profiles thereof, are converted statistically into derived variables by means of computational means and the actual profiles thereby formed are investigated for significant features, profiles, etc. According to a further aspect of the invention, which may likewise be of independent importance, there is also the possibility that derived variables for which significant features have been established are used during subsequent bonding processes in the calculation of the quality index. In this way, the number of input vectors for the feature extraction and the future number of values contained in an index group are increased by 1. Such a monitoring system advantageously ensures that the quality monitoring is constantly extended or checked over and best-possible model forming is achieved. According to a further aspect of the invention, which may likewise be of independent importance, there is the possibility that, with predetermined deviations between bond index groups and memory index groups, electrical signals in particular are automatically generated and used, preferably in real time, for controlling or regulating the bonding operations in progress at the time and/or carried out subsequently. To this extent, this is negative feedback in dependence on known memory index groups or error classes. It is also preferred that the method is carried out according to one or more of the features described above in an automated manner after setting and/or maintenance work on a bonding apparatus. It is preferred as being suitable that the method is carried out according to one or more of the features described above in an automated manner, preferably in a computer-aided manner using software.
As far as its fourth aspect is concerned, the invention also includes, furthermore, a bonding apparatus, preferably an ultrasonic wire bonding apparatus, for the production and quality control of ultrasonically bonded connections, which according to claim 57 is suitable for carrying out the method according to the invention, with one or more of the features described above, or is adapted for carrying it out. Preferred possible developments are also specified in claims 58 to 62, reference being made to the description as a whole and the figures for possible effects and advantages.
According to the explanations given above, the invention makes it possible to classify errors that are possible during bonding processes, automatically allocate them to error classes and name bonding errors in an automated manner. To this extent, the invention also includes the possibility of generating correction variables in real time in dependence on input textures in the feature extraction and providing feedback to the process controlling system. There is also the possibility of generating correction variables in the quality calculation in the form of a trend analysis and providing feedback to the process controlling system. A further application possibility is that of monitoring user interventions and checking over maintenance work. The invention also makes it possible to transform an error vector to a scalar quality variable.
As far as the various aspects described are concerned, the invention also independently includes in addition to the bonding apparatuses respectively described for them those components or modules that are suitable or satisfactory for carrying out the individual methods according to the invention. To this extent, corresponding components or modules may also be the subject of independent patent claims.
The aspects of the present invention are described in more detail below with reference to the accompanying figures, in which preferred exemplary embodiments of the method according to the invention and apparatuses according to the invention are represented and in which:
A model for the forming of a connection suitable for assessing the quality of the connection can be generated from the measurement data which, according to one aspect of the invention, are recorded during the learning phase.
If all sampled variables are recorded and the relative frequency of their amplitudes is considered, a two-dimensional probability density function over the amount of the individual vector components is obtained in this way. A probability distribution of this kind is likewise represented only by way of example in
However, it is also possible to use input vectors of which the components are associated with specific values of other measured variables, so that not every vector contains components which can be assigned to equidistant sampling times. The lengths of the feature vectors may likewise differ, since their effect on the system may be of importance for ranges of different sizes of the reference variables.
The statistical models may be chosen in dependence on the input variable. The minimum amount of reference bonds for representative statistics depends on the model chosen. In principle, the aim should be to obtain a random sample that is as large and representative as possible. The minimum size is specifically predetermined by the system as a constant. The size of the range of variance or of the confidence interval (
The quality calculation is preferably divided into two stages, namely the feature extraction or determination of the individual quality indices Qi (
In a next step, the n feature vectors at the input of the feature extraction are mapped onto n scalars at the output. These may once again be regarded as a vector with n components and serve as an input variable for the quality calculation. In the quality calculation, this vector is mapped onto a scalar by a procedure established according to the significance of the components. This scalar is the quality value Q that is being sought. A threshold value which, depending on the configuration, may necessitate intervention by an operator, is a parameter that can be set.
In continuance or development of the aforementioned possibilities, feature vectors (Q1 . . . Qn) or error vectors may be categorized with regard to their associated error in a further step.
The raw data from the signal preprocessing and the ultrasonic generator are not only used for feature extraction (
The monitoring system processes the aforementioned values taking account of the statistics known at the time. It is activated, for example, when the substrate is changed or after a previously established number of bonded connections have been carried out and first investigates the entered data for aberrants (
The dimension of the input vector for the quality calculation/error classification is consequently likewise increased by one. The new feature vector is transferred furthermore from Block 2.3 to Block 2.4 (adaptation of the quality calculation and classification). In this area, the model is checked for consistency on the basis of the new data, optionally adapted and returned to the quality calculation and the error classifier (
According to a further aspect of the invention, with knowledge of the calculated quality and the underlying model, a deviation from the model can be noticed in real time already during the calculation of the vector components which belong to the individual measured variables, and are generally a function of time, and the process can be influenced by suitable parameter adaptation at the time of running the process (see also
According to a further aspect of the invention, which may also be of independent significance, the method steps and features of a bonding apparatus described above can also be used for checking user interventions. On the basis of exact knowledge of a transducer-wedge system, it is also possible, for example, to check whether, for example, the transducer has been correctly fitted after maintenance work and whether the wedge has been properly installed and fixed with the correct pre-tensioning of the wedge screw. The state of the clamping and the bond holding can also be checked for correct functioning on the basis of the learned textures or profiles.
In Block 1 of
In
All features disclosed are (in themselves) pertinent to the invention. Individual aspects of the invention that are described, in particular including individual features thereof, may also be combined with one or more of the other aspects of the invention that are described, in particular including with individual features thereof. The disclosure content of the associated/accompanying priority documents (copy of the prior patent application) is also hereby incorporated in full in the disclosure of the application, including for the purpose of incorporating features of these documents in claims of the present application.
| Number | Date | Country | Kind |
|---|---|---|---|
| 10 2007 054 626.4 | Nov 2007 | DE | national |
The present application is a continuation of pending International patent application PCT/EP2008/065101 filed on Nov. 7, 2008, which designates the United States and claims priority from German patent application number 10 2007 054 626.4 filed on Nov. 12, 2007, the content of which is incorporated herein by reference.
| Number | Date | Country | |
|---|---|---|---|
| Parent | PCT/EP2008/065101 | Nov 2008 | US |
| Child | 12778883 | US |
