CN101726542B - Method and system for monitoring and identifying ultrasonic bonding quality in package and interconnection of chips on line - Google Patents
Method and system for monitoring and identifying ultrasonic bonding quality in package and interconnection of chips on line Download PDFInfo
- Publication number
- CN101726542B CN101726542B CN2009103077466A CN200910307746A CN101726542B CN 101726542 B CN101726542 B CN 101726542B CN 2009103077466 A CN2009103077466 A CN 2009103077466A CN 200910307746 A CN200910307746 A CN 200910307746A CN 101726542 B CN101726542 B CN 101726542B
- Authority
- CN
- China
- Prior art keywords
- bonding
- current signal
- value
- driving current
- fundamental frequency
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Expired - Fee Related
Links
Images
Landscapes
- Wire Bonding (AREA)
Abstract
The invention discloses an online monitoring and identifying method and an online monitoring and identifying system for ultrasonic bonding quality in package and interconnection of chips. The method and the system analyze drive current signals of a transducer in a bonding process and extract the characteristics of the drive current signals, then send the extracted characteristic quantity into a neural network-based identifier to output an output value which indicates whether bonding is successful or not so as to judge whether the present bonding is successful or not in real time and provide a remedial chance or mark for unsuccessful bonding. Therefore, the method and the system improve the production efficiency and reliability of the package and interconnection of the chips and greatly save the cost of the package of the chips.
Description
Technical field
The invention belongs to the ultrasonic bond field in the Chip Packaging interconnection, relate to ultrasonic bonding quality on-line monitoring method of discrimination and system in a kind of Chip Packaging interconnection.
Background technology
Ultrasonic bond is widely used in the microelectronics Packaging interconnection, and it is that the ultrasonic vibration that utilizes transducer to produce acts on bonding wire (ball) and the pad, realizes the method for the metallic bonding between tinsel (ball) and pad.Current microelectronics Packaging interconnection more than 90% all adopts the ultrasonic lead key connection technology to realize, and the off-line sampling Detection is all adopted in the detection of para-linkage quality, and this method poor in timeliness waits and inspects bonding failure by random samples, often produce a large amount of substandard products, caused yield rate to decline to a great extent.Along with the cost that is encapsulated in the chip manufacturing increases (at present near 80%) gradually, bonding speed is accelerated (15 ~ 20 line/second), solder pad space length dwindles (60 ~ 40um), industry expectation development ultrasonic bonding quality on-line monitoring and discrimination technology improve bonding reliability and yield rate.Because ultrasonic bond is the complicated dynamic process of a multivariate coupling, never form practical bonding quality on-line monitoring and method of discrimination at present.
Nearly 10 yearly correlations research mainly contains: 1998, people such as S.W OR installed extra piezoelectric sensor in the piezoelectric pile of ultrasonic lead key connection transducer, and judge bond strength with the output signal change procedure of piezoelectric sensor in the bonding process.Similarly, people such as Michael embedded sensor in the transducer piezoelectric pile in 2002, and according to the bonding of " good " and " bad ", particularly polluted surface of piezoelectric sensor signal distinguishing bonding.Though these methods can detect the bonding failure, the sensor that embeds in piezoelectric pile has had a strong impact on transducer performance, is not applied in the actual production.Also have some then to wait the method for judging bonding quality,, also be difficult to be applied in the actual production owing to need to increase the measuring equipment that overbalances the bonding apparatus several times by the vibration of measuring lead wire distortion and measurement transducer.
Summary of the invention
Technical matters to be solved of the present invention provides ultrasonic bonding quality on-line monitoring method of discrimination and the system in a kind of Chip Packaging interconnection.This method and system can be implemented under any bonding conditions, analysis and feature extraction by transducer drive electric signal in the para-linkage process, whether thereby real-time judge is worked as time bonding successful, and for the failure bonding chance or the mark of remedying is provided, to improve the production efficiency and the reliability of Chip Packaging interconnection, can save the cost of Chip Packaging greatly thus.
Technical conceive of the present invention is: development is judged the method for bonding quality by detecting the transducer electric signal, can not influence the performance of transducer, need not increase the hardware input of equipment simultaneously.The foundation of this method is the understanding to ultrasonic bond mechanism.According to the understanding of present para-linkage mechanism, real online monitoring of bonding is divided into three phases: at first, and lead-in wire and the pad motion that under bonding action, contacts, and produce at first and be out of shape; Ultrasonic energy is delivered to bonding face through luffing bar and chopper then, produces little bonding point at the surface of contact periphery, begins to take shape bond strength; Last ultrasonic energy increases contact area, little bonding point quantity increases, the bonding degree of depth increases, and forms ultrasonic bond intensity thus.In the said process, transducer system impedance, natural frequency change with the variation of bonded interface state, and are reflected as the variation of transducer drive electric current and voltage, and the bonding process difference, the intensity difference, and its change procedure is also different.Therefore, in conjunction with bonding mechanism understanding, the feature of excavating change procedure finds the feature that reflects the intensity generative process, just can realize online detection of bonding quality and differentiation.
For solving the problems of the technologies described above, the technical solution adopted in the present invention is:
Ultrasonic bonding quality on-line monitoring method of discrimination in a kind of Chip Packaging interconnection is characterized in that, may further comprise the steps:
Steps A: the driving current signal of in bonding process, gathering transducer; Obtain following 8 eigenwerts according to the driving current signal of gathering:
1) the fundamental frequency signal effective value mean value avereng of described current signal;
2) fundamental frequency effective value minimum value min in the bonding process of described current signal;
3) maximal value of the fundamental frequency effective value of current signal described in the bonding process and the difference diff of minimum value;
4) described current signal is carried out modulation signal zero passage after the demodulation process Num0 that counts;
5) the variance var1 in the fundamental frequency effective value t1-t2 of described current signal, t1 gets 50-70ms, and t2 gets 90-110ms; [t1, t2 be the value difference under different bonding conditions, and what more than provide is general value.】
6) the mean value mean1 in the fundamental frequency effective value t1-t2 of described current signal, t1 gets 50-70ms, and t2 gets 90-110ms; [t1, t2 be the value difference under different bonding conditions, and what more than provide is general value.】
7) described current signal is carried out the Frequency point of amplitude maximum after the FFT conversion, i.e. fundamental frequency value f;
8) described current signal is carried out the amplitude famp of the Frequency point correspondence of amplitude maximum after the FFT conversion;
Step B): constructing neural network, steps A 8 eigenwerts) are as 8 input quantities of neural network; Neural network is output as one, and output valve is 0 or 1,0 expression bonding failure, 1 expression bonding success;
Step C): according to steps A) gather many groups sample of 8 eigenwerts to step B) in neural network train; After training successfully, use this neural network that the ultrasonic bonding quality in the Chip Packaging interconnection is carried out on-line monitoring and differentiation.
In described steps A) in:
The computing formula of the fundamental frequency signal effective value mean value avereng of described current signal is
The computing formula of fundamental frequency effective value minimum value min is in the bonding process of described current signal
min=min(D
1(51:1000));
The computing formula of the maximal value of the fundamental frequency effective value of current signal described in the bonding process and the difference diff of minimum value is diff=max (D
1(51:1000))-min (D
1(51:1000));
D wherein
1Be the fundamental frequency signal effective value of current signal, D
1Computing formula be
S is described current signal,
D
1(51:1000) D is got in expression
1The the 51st to the 1000th point, time corresponding is 10-195ms;
4) computing formula of described variance var1 is
5) computing formula of described mean value mean1 is
Wherein D is D in the 60-100ms
1Mean value.
Ultrasonic bonding quality on-line monitoring judgement system in the interconnection of a kind of Chip Packaging is characterized in that, comprises the data collector of the driving current signal that is used to gather transducer; Be used for the signal that the data harvester is gathered is carried out the data processing equipment of data processing; Be used for differentiating the neural network identifier of whether bonding success according to input quantity; Described data collector is connected with data processing equipment, and 8 eigenwerts of described data processing equipment output are as the input quantity of described neural network identifier; The output quantity of described neural network identifier is an identification result, and the value of output quantity is 0 or 1,0 expression bonding failure, 1 expression bonding success; Described 8 eigenwerts are as follows:
1) the fundamental frequency signal effective value mean value avereng of described current signal;
2) fundamental frequency effective value minimum value min in the bonding process of described current signal;
3) maximal value of the fundamental frequency effective value of current signal described in the bonding process and the difference diff of minimum value;
4) current signal described in the para-linkage process carries out modulation signal zero passage after the demodulation process Num0 that counts;
5) the variance var1 in the fundamental frequency effective value t1-t2 of described current signal, t1 gets 50-70ms, and t2 gets 90-110ms; [t1, t2 be the value difference under different bonding conditions, and what more than provide is general value.】
6) the mean value mean1 in the fundamental frequency effective value t1-t2 of described current signal, t1 gets 50-70ms, and t2 gets 90-110ms; [t1, t2 be the value difference under different bonding conditions, and what more than provide is general value.】
7) described current signal is carried out the Frequency point of amplitude maximum after the FFT conversion, i.e. fundamental frequency value f;
8) described current signal is carried out the amplitude famp of the Frequency point correspondence of amplitude maximum after the FFT conversion.
Ultrasonic bond is the effect by ultrasonic vibration and chopper pressure, tinsel (ball) is welded on chip bonding pad and the substrate draw-foot, thus the technology that the circuit of chip and substrate is linked together.In bonding process, along with the progressively formation of bonded interface strength of joint, the interface operating mode that tinsel (ball), chopper, chip, substrate are formed constantly changes.Make system mechanics characteristic (resonant frequency, the vibration shape etc.) change on the one hand, show as the change of chopper vibration vibration frequency and driving signal frequency; Make the power of system consumption change on the other hand, show as the change of PZT (piezoelectric ceramics) drive current and voltage magnitude.Make damping and bonded interface microscopic characteristics between system's chopper, tinsel (ball) and the substrate change simultaneously, show as the change of current amplitude and resonance frequency.And the difference of these signals will cause different bonding qualities, and therefore, the variation of extracting these signals can be judged the quality of bonding quality.
According to above-mentioned thinking, content of the present invention comprises: gather the transducer drive current voltage signal, adopt methods such as wavelet analysis, Fast Fourier Transform (FFT) then, extract the variation characteristic of signal in bonding process, and this feature imported trained neural network, thereby obtain the whether successful judgement of bonding.
Ultrasonic bonding quality on-line monitoring system of the present invention is made up of data collecting card, signals collecting program, signal analysis program, neural network procedure etc.
Workflow of the present invention is:
At first adopt conventional method to configure bonding parameter, guarantee on most of bonded energy bonding, the Usage data collection card is gathered 200 groups of above transducer drive current and voltage sample data under this bonding conditions, preserves these data and uses for sample analysis.
Use is drawn and is cut test machine and test these bonding samples, according to industrial standard determines the whether success of these bondings.
Use signal analysis program that the data of above-mentioned preservation are carried out analyzing and processing, extract the feature relevant with bonding quality.
As input, above-mentioned drawing cut test machine test result (whether successful) as output with above-mentioned feature, and the input artificial neural network is trained, and trains successfully this network of back preservation, can drop into actual online use.
During online use, data acquisition, signal analysis and neural network procedure and the real-time communication of ultrasonic bond system, after bonding is finished, whether the ultrasonic bonding quality on-line monitoring system will be judged this time bonding according to the signal characteristic of this time bonding successful, and feed back to bonding system, if the bonding failure can be reminded the user, otherwise carry out next bonding.
After changing bonding parameter, repeat above-mentioned 1 ~ 5 step to get final product.
Ultrasonic bonding quality on-line monitoring and judgement system in the Chip Packaging interconnection comprise:
The data acquisition system (DAS) of transducer drive current, voltage signal comprises signal sensing circuit and acquisition software in energy synchronous acquisition real online monitoring of bonding;
The Signal Analysis System of an above-mentioned signal characteristic of energy extract real-time;
The bonding quality of energy real-time learning and recognition feature is judged system's (being an artificial neural network software in the present invention);
The ultrasonic bonding quality on-line monitoring system, it is characterized in that: the transducer drive electric signal of acquisition system collection is easy to detect, gather the performance that can't influence bonding machine energy converting system of building of required sensing circuit, also do not increase expensive equipment and cost simultaneously; Use digital signal processing methods such as wavelet analysis, Fast Fourier Transform (FFT) above-mentioned signal to be handled in real time the feature relevant that comprises in the picked up signal with bonding quality; Use artificial neural network, judge when the whether success of time bonding according to these eigenwerts through training in advance.
The beneficial effect that the present invention had:
The present invention is by the influence of each parameter para-linkage result in the para-linkage process, extract and closely-related 8 characteristic quantities of bonding quality, discerning according to these characteristic quantity para-linkage success or failure, specifically, the present invention passes through the analysis and the feature extraction of transducer drive electric signal in the para-linkage process cleverly, whether thereby real-time judge is worked as time bonding successful, and for the failure bonding chance or the mark of remedying is provided, to improve the production efficiency and the reliability of Chip Packaging interconnection, can save the cost of Chip Packaging greatly thus.
Adopt after the method for the present invention, interconnected lead-in wire crash rate can be reduced to below 10/1000000ths from 1,000,000 present/100-50.
Compare the transducer drive electric signal that the present invention is easy to detect by analysis, the quality of judging bonding quality of energy real-time online with other bonding quality determination methods.Sensing circuit does not influence the performance of transducer system, does not need to introduce expensive equipment, has the advantages such as reliability that cost is low, response is fast, efficient is high and improve Chip Packaging greatly.
Description of drawings
Fig. 1 is a workflow diagram of the present invention;
Fig. 2 is a signals collecting sensing circuit synoptic diagram; 1-luffing bar wherein; 2-ultrasonic vibration direction; The 3-chopper; The 4-worktable; 5-PZT.
Fig. 3 is each frequency band signal in the current signal;
Current harmonics when Fig. 4 fails for bonding;
Fig. 5 is ((a) bonding success of signal contrast after the demodulation; (b) bonding failure);
Current signal when Fig. 6 goes between bonding for not having;
Fig. 7 is the fashionable current signal of dangling bonds;
Current signal when Fig. 8 is the contaminated surface bonding [(a) bonding failure, (b) bonding failure (c) bonding success];
Current effective value curve when Fig. 9 fails for the part bonding [(a) current amplitude becomes big suddenly, the bonding failure, (b) the current amplitude fluctuation is violent, the bonding failure];
Figure 10 is trained values and expectation value contrast;
The result of Figure 11 for all bondings are discerned.
Figure 12 changes the bonding quality identification [(a) the neural metwork training result (b) under the change bonding conditions changes the Application of Neural Network result under the bonding conditions] under the bonding conditions.
Embodiment
The invention will be further described below in conjunction with the drawings and specific embodiments.
The present invention realizes on WS3000 type crude aluminum silk ultrasonic lead key connection machine platform.Because the ultrasonic system and the bonding mechanism thereof of ultrasonic bond are all similar, so the present invention also can be applied in hot ultrasonic lead key connection, the hot ultrasonic back bonding system fully.
Embodiment 1:
With WS3000 type crude aluminum silk ultrasonic lead key connection machine is platform, and specific implementation process of the present invention is:
At first make up a multi-channel high-speed data acquisition system that is used for the ultrasonic bonding quality on-line monitoring, realized the synchronous acquisition of transducer drive voltage, current signal in the bonding process.This system is made up of PCI high-speed data acquisition card, acquisition software and signal sensing circuit.Signal sensing circuit as shown in Figure 2.Acquired signal is the electric current of process transducer piezoelectric ceramics and the voltage at its two ends, extraction voltage is U2, electric current converts voltage U 1 to by 9 ohm resistance sampling, because the input impedance of acquisition system is very high, the resistance of current sampling resistor is very little, so the access of signal sensing circuit is to the not influence of former energy converting system.Therefore in the large power supersonic bonding, the PZT driving voltage is higher than the ceiling voltage that data collecting card can bear, and adopts 10 times the BNC probe of decaying to insert capture card, decay back output voltage U 2, and voltage is decayed in the present invention's experiment.
The relation of each road measured signal and actual signal is as follows:
Electric current I=U1/9 * 1000 (mA)
Voltage U=U2 * 10 (V)
Acquisition software adopts LabView to write, and except main acquisition function, software comprises that also each parameter setting of capture card, image data show, the spectrum analysis result shows, the user operates subsidiary functions such as response, data preservation, graunch processing.
After obtaining above-mentioned signal, the present invention has write the Matlab program, use the Wavelet time-frequency analytical approach to extract the feature of current signal, and according to the intensity of bonding, obtained the incidence relation of current signal feature and bond strength under the specific bonding conditions, these associations can be used for the online differentiation of bonding quality.
Wavelet analysis and feature extraction:
1, be 3.5 lattice (about 12.2N), bonding time to be that 3.5 lattice (about 200ms), bonding power are that most bondings can both have good bond strength under the bonding conditions of 4.7 lattice (about 0.8W) at bonding pressure.The transducer driving current signal is carried out WAVELET PACKET DECOMPOSITION, and the effective value curve that obtains its fundamental frequency, frequency multiplication, frequency tripling, quadruple signal as shown in Figure 3.Under this bonding conditions, if current attenuation is slow, the current attenuation amplitude shows then that less than 3mA bonding quality does not form all the time in the bonding process in the 10-195ms; (100-195ms) current average of second half section during bonding is very big, surpasses 12mA, and impedance is very little, shows that then last bonding quality does not form; Finally show as the bonding failure, as shown in Figure 4.
Effective value curve calculation formula is among the figure:
(1)
Wherein, S is a current component signal, and D is the effective value signal.
According to above analysis, after mass data handled as can be known, under this bonding conditions, if in bonding process in the current signal 10-195ms fundamental frequency effective value mean value greater than 13mA, the effective value minimum value is greater than 11mA, and the difference of effective value maximal value and minimum value is less than 3mA, and then bonding is failed.Its signal characteristic extracts as follows:
Fundamental frequency signal effective value mean value avereng, its computing formula is
Fundamental frequency effective value minimum value min in the bonding process, its computing formula is
min=min(D
1(51:1000)) (3)
The difference diff of fundamental frequency effective value maximal value and minimum value in the bonding process, its computing formula is
diff=max(D
1(51:1000))-min(D
1(51:1000)) (4)
D wherein
1Be fundamental frequency effective value, D
1(51:1000) D is got in expression
1The the 51st to the 1000th point, time corresponding is 10-195ms.
2, the transducer driving current signal is carried out demodulation process, can obtain modulation signal wherein.The demodulation of the current signal of failing for bonding success and bonding, the result who obtains has notable difference.As shown in Figure 5, signal is random after bonding when failure demodulation, and signal is tending towards rule after the current signal demodulation during the bonding success.According to this analysis, the extraction eigenwert is
4) the modulation signal zero passage Num0 that counts after the demodulation;
Be less than 4 if modulation signal zero passage in 10-195ms is counted, then will cause the bonding failure.
3, in bonding process, also belong to the bonding failure as no lead-in wire situation occurring.Each frequency band signal of electric current as shown in Figure 6 at this moment.Compare during with normal bonding (Fig. 3), this moment amplitude change mild, though some decay is arranged, not obvious, peak value more than 16mA, and during normal bonding current peak all below 15mA.The feature that show this moment is with the 1st point analysis.
If the situation of unsettled bonding will appear in chopper solid failure during 4 bondings and upwards sliding, promptly chopper does not touch substrate at all during bonding.Almost not decay of current signal in the case changes not obviously, and amplitude is obviously bigger, and peak value is near 23mA.As shown in Figure 7.The feature that show this moment is also with the 1st point analysis.
5, in bonding process,, in bonding process, will have a strong impact on the formation of bonding quality if there are oxidation and pollutant in the chip bonding pad surface.When using this pad to carry out bonding, the transducer drive current signal changes complexity with the pollution condition of bonding, and most of bondings all can be failed.As shown in Figure 8, big when current peak was than normal bonding when contaminated surface carried out bonding, but fashionable littler than dangling bonds, mild if amplitude changes, then bonding failure.Even its bond strength of bonding success is also than just often low in the case.The feature that show this moment is also with the 1st point analysis.
6, in bonding process, catastrophe point appears in the current signal of part bonding failure, as shown in Figure 9, compares with Fig. 3, shows as in the t1-t2 to alter a great deal, and t1, t2 be the value difference under different bonding conditions, and general t1 gets 50-70ms, and t2 gets 90-110ms; Under this bonding conditions, t1-t2 is 60-100ms, and its standard deviation is greater than 8mA, and mean value is less relatively, less than 11mA.Therefore can extract and be characterized as
5) fundamental frequency effective value 60-100ms internal variance var1, its computing formula is
6) mean value mean1 in the fundamental frequency effective value 60-100ms, its computing formula is
Wherein D is a mean value in the 60-100ms.
7, the lead-in wire bonding finally rely on energy converting system resonance so that required ultrasonic energy to be provided, the resonance frequency of system finally is all multifactor relevant with bonded interface state, impedance magnitude etc.After current signal carried out Fast Fourier Transform (FFT), the frequency of its amplitude maximum has reflected the frequency of operation of this time bonding energy converting system, and its size changes near the transducer natural frequency, if this frequency values is away from the transducer natural frequency, then ultrasonic phase-locked failure will cause the bonding failure.This frequency has reflected information such as bonded interface state, impedance, so this frequency also will be relevant with bonding quality, and its amplitude size has been reacted the concentration of energy situation simultaneously, and the big more then energy of amplitude is concentrated more, extracts in view of the above to be characterized as
7) Frequency point of amplitude maximum, i.e. fundamental frequency value f after the FFT conversion
8) the amplitude famp of maximum frequency correspondence after the FFT conversion
If fundamental frequency value f is greater than 60kHz or less than 56.28kHz, the then phase-locked failure of system, bonding also will fail, as if the amplitude famp of maximum frequency correspondence after the FFT conversion less than 50mA
2, then energy is not concentrated in the bonding process, the bonding failure.
Based on above analysis, adopt wavelet analysis and Fast Fourier Transform (FFT) (FFT) analytical approach, at bonding pressure is 3.5 lattice (12.2N), the bonding time is 3.5 lattice (200ms), bonding power is under the bonding conditions of 4.7 lattice (0.8W), extracts 8 signal characteristics and is respectively: current signal fundamental frequency effective value mean value, fundamental frequency effective value minimum value in the bonding process, fundamental frequency effective value maximal value and minimum value poor, the number of modulation signal zero crossing after the demodulation, fundamental frequency effective value 60-100ms period internal variance and mean value, the amplitude of Frequency point of amplitude maximum (fundamental frequency) and maximum frequency correspondence after the FFT conversion.When bonding conditions changes, then can be according to analytical approach before, again the electric signal in the para-linkage is analyzed, identical when the eigenwert span is no longer with above analysis when obtaining the bonding success, but can obtain one group of eigenwert of judging that bonding success and bonding are failed certainly according to this method.
The BP Network Design:
It is as follows to choose BP network initial parameter:
Neural network input layer number is 8, the number of the feature of extracting in promptly above the analysis, and hidden layer is 1 layer, and hidden layer node is 15, and output node is 1, is output as linear function output.
Neural network needs training before use.The training function of BP neural network has a lot, adopts the Levenberg-Marquardt training method in the present invention, and this method is under the situation of quadratic sum form good speed of convergence to be arranged at the performance function.
The parameter of training is provided with as follows:
Net.trainParam.time=inf; The maximum training time of %
Net.trainParam.lr=0.05; The % learning rate
Net.trainParam.epochs=1000; The maximum frequency of training of %
Net.trainParam.goal=1e-5; % training requirement precision is promptly expected error
Net.trainParam.min_grad=1e-10; The requirement of % minimal gradient
Net.trainParam.max_fail=5; The maximum frequency of failure of %
In the training process, as long as satisfy any one condition in following 5 conditions, training will stop:
Surpass maximum iteration time epochs.
The performance functional value is less than error criterion goal.
Grad is less than accuracy requirement mingrad.
Train used time overtime restriction time.
The maximum frequency of failure surpasses number of times restriction max_fail.
Training process is: test 300-500 bonding and observe the success or failure of bonding, choose 150 subnormal bondings, do not have the lead-in wire bondings for 60 times, 60 unsettled bondings, 45 contaminated surface bondings train neural network as sample.Desired value and trained values correlation curve are as shown in figure 10 in the training process.
After training finishes, use this neural network 315 bondings of having trained and 315 bondings that carry out are in real time discerned, its result is (wherein 1 expression bonding success, 0 expression bonding failure) as shown in figure 11, as seen from the figure, this neural network can effective recognition go out the situation of all bonding failures.
At last the network application that has trained in the ultrasonic bonding quality on-line monitoring system.Behind the ultrasonic bond, acquisition system is gathered the driving electric signal of piezoelectric ceramics in real time and is sent in the analytic system each time.Analytic system obtains its 8 eigenwerts according to the feature extracting method described in the above explanation.Whether neural network goes out current bonding according to these 8 eigenwert real-time judge successful.1) if judge the bonding success, then carries out the bonding of next bar lead-in wire; 2) if judge the bonding failure, then bonding suspends and the current bonding failure of alarm, and the notifying operation workman gets involved processing, such as putting mark on the current chip so that in follow-up encapsulation process with its rejecting; Perhaps according to predetermined rule, the lead-in wire of bonding (promptly mending line) again on the next door of bonding point.
When bonding conditions changes, adopt said method can extract its characteristic of correspondence and structure monitoring system, be that the quantity of eigenwert under the different bonding conditions and the computing formula and the threshold value thereof of eigenwert are understood different.Following analyzing examples.At bonding pressure is 3 lattice (about 11N), bonding time to be that 3 lattice (about 170ms), bonding power are that most bondings can both have good bond strength under the bonding conditions of 3 lattice (about 0.6W).It is as follows to adopt above-mentioned steps to extract its eigenwert:
1) fundamental frequency signal effective value mean value avereng, its computing formula is
2) fundamental frequency effective value minimum value min in the bonding process, its computing formula is
min=min(D
1(41:880)) (8)
3) fundamental frequency signal second half section attenuation amplitude diff in the bonding process, its computing formula is
diff=max(D
1(470:480))-min(D
1(870:880)) (9)
4) the modulation signal zero passage Num0 that counts after the demodulation;
5) Frequency point of amplitude maximum, i.e. fundamental frequency value f after the FFT conversion;
6) the amplitude famp of maximum frequency correspondence after the FFT conversion;
Owing under this bonding conditions, there not be as shown in Figure 9 sudden change feature of appearance in the signal process, thus aforementioned feature fundamental frequency effective value 60 100ms internal variance var1 and mean value mean1 can not want at this, with simplified network model.The pass of each eigenwert and bond strength is under this bonding conditions: if fundamental frequency signal mean value avereng is greater than 9mA, fundamental frequency effective value minimum value min is greater than 7mA in the bonding process, attenuation amplitude diff is less than 1mA the fundamental frequency signal second half section in the bonding process, the modulation signal zero passage Num0 that counts is less than 5 after the demodulation, and the amplitude famp of maximum frequency correspondence is less than 30mA after the FFT conversion
2, fundamental frequency value f is greater than 60kHz or less than 56.3660kHz, and then bonding is failed.
Adopt above-mentioned 6 eigenwerts to make up neural network, 100 bonding sample training backs are tested other 100 bonding signals, as shown in figure 12, these eigenwerts can well identify the bonding failure.
Adopt after the method for the present invention, interconnected lead-in wire crash rate can be reduced to below 10/1000000ths from 1,000,000 present/100-50.
In this example, sample frequency is 1024kHz.
Claims (1)
1. the ultrasonic bonding quality on-line monitoring method of discrimination during a Chip Packaging interconnects is characterized in that, may further comprise the steps:
Steps A): the driving current signal of in bonding process, gathering transducer; Obtain following 8 eigenwerts according to the driving current signal of gathering:
1) the fundamental frequency signal effective value mean value avereng of described driving current signal;
2) fundamental frequency effective value minimum value min in the bonding process of described driving current signal;
3) maximal value of the fundamental frequency effective value of driving current signal described in the bonding process and the difference diff of minimum value;
4) described driving current signal is carried out modulation signal zero passage after the demodulation process Num0 that counts;
5) the variance var1 in the fundamental frequency effective value t1-t2 of described driving current signal, t1 gets 50-70ms, and t2 gets 90-110ms;
6) the mean value mean1 in the fundamental frequency effective value t1-t2 of described driving current signal, t1 gets 50-70ms, and t2 gets 90-110ms;
7) described driving current signal is carried out the Frequency point of amplitude maximum after the FFT conversion, i.e. fundamental frequency value f;
8) described driving current signal is carried out the amplitude famp of the Frequency point correspondence of amplitude maximum after the FFT conversion;
Step B): constructing neural network, steps A 8 eigenwerts) are as 8 input quantities of neural network; Neural network is output as one, and output valve is 0 or 1,0 expression bonding failure, 1 expression bonding success;
Step C): according to steps A) gather many groups sample of 8 eigenwerts to step B) in neural network train; After training successfully, use this neural network that the ultrasonic bonding quality in the Chip Packaging interconnection is carried out on-line monitoring and differentiation;
In described steps A) in:
The computing formula of the fundamental frequency signal effective value mean value avereng of described driving current signal is
The computing formula of fundamental frequency effective value minimum value min is in the bonding process of described driving current signal
min=min(D
1(51∶1000));
The computing formula of the maximal value of the fundamental frequency effective value of driving current signal described in the bonding process and the difference diff of minimum value is diff=max (D
1(51: 1000))-min (D
1(51: 1000));
D wherein
1Be the fundamental frequency signal effective value of driving current signal, D
1Computing formula be
I=0,1 ..., length (S)/200; S is described driving current signal, D
1D is got in (51: 1000) expression
1The the 51st to the 1000th point, time corresponding is 10-195ms;
The computing formula of described variance var1 is
The computing formula of described mean value mean1 is
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2009103077466A CN101726542B (en) | 2009-09-25 | 2009-09-25 | Method and system for monitoring and identifying ultrasonic bonding quality in package and interconnection of chips on line |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2009103077466A CN101726542B (en) | 2009-09-25 | 2009-09-25 | Method and system for monitoring and identifying ultrasonic bonding quality in package and interconnection of chips on line |
Publications (2)
Publication Number | Publication Date |
---|---|
CN101726542A CN101726542A (en) | 2010-06-09 |
CN101726542B true CN101726542B (en) | 2011-06-01 |
Family
ID=42447741
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2009103077466A Expired - Fee Related CN101726542B (en) | 2009-09-25 | 2009-09-25 | Method and system for monitoring and identifying ultrasonic bonding quality in package and interconnection of chips on line |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN101726542B (en) |
Families Citing this family (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103901266A (en) * | 2014-02-17 | 2014-07-02 | 浙江海洋学院 | Identification method of ultrasonic bonding power |
CN104459430B (en) * | 2014-09-01 | 2018-05-08 | 哈尔滨工业大学深圳研究生院 | Ultrasonic lead key connection loses line detector and method |
CN107958850A (en) * | 2017-11-28 | 2018-04-24 | 宁波尚进自动化科技有限公司 | A kind of solder joint welding quality monitoring system and its monitoring method |
CN109975687B (en) * | 2019-03-14 | 2022-07-08 | 深圳市大族封测科技股份有限公司 | Quality detection device and method based on IC bonding lead |
-
2009
- 2009-09-25 CN CN2009103077466A patent/CN101726542B/en not_active Expired - Fee Related
Also Published As
Publication number | Publication date |
---|---|
CN101726542A (en) | 2010-06-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN103575523B (en) | The rotary machinery fault diagnosis method of kurtosis-envelope spectrum analysis is composed based on FastICA- | |
CN106271036B (en) | Ultrasonic wave metal welding quality appraisal procedure, device and ultrasonic metal bonding machine | |
CN102288883B (en) | Oscillation wave partial discharge identifying and positioning method for asynchronous double-end power cable | |
CN101726542B (en) | Method and system for monitoring and identifying ultrasonic bonding quality in package and interconnection of chips on line | |
CN107014668A (en) | A kind of fatigue crack integrated monitoring based on piezoelectricity and smart coat sensor | |
CN108343599B (en) | A kind of water pump assembly intelligent failure diagnosis method based on more granularities cascade forest | |
CN110514957A (en) | Automatic inspection method and platform for transformer substation | |
CN104545887A (en) | Method and device for identifying artifact electrocardiograph waveforms | |
CN104665849A (en) | Multi-physiological signal multi-model interaction-based high-speed railway dispatcher stress detecting method | |
CN103675544A (en) | Power system fault signal detection and waveform identification method based on optimization algorithm | |
CN106441547A (en) | Transformer vibration monitoring method and apparatus | |
CN109649432A (en) | Cloud platform rail integrity monitoring systems and method based on guided wave technology | |
CN106441896A (en) | Characteristic vector extraction method for rolling bearing fault mode identification and state monitoring | |
CN105223466A (en) | A kind of extra high voltage direct current transmission line method of single end distance measurement utilizing modulus maximum ratio | |
CN103267652B (en) | Intelligent online diagnosis method for early failures of equipment | |
CN107350900A (en) | A kind of tool condition monitoring method based on the extraction of chip breaking time | |
CN104459430B (en) | Ultrasonic lead key connection loses line detector and method | |
CN106338237A (en) | Transformer winding deformation detection method based on frequency response impedance method | |
CN110208377A (en) | A kind of more characteristic parameters damage degree assessment method based on Lamb wave | |
CN100593719C (en) | Method for investigating failure connection of engineering structure by signal focus | |
CN103558513B (en) | A kind of aircraft cable network Fault Locating Method based on Graphic Pattern Matching algorithm | |
CN106338662A (en) | Transformer winding deformation diagnosis method based on mathematical morphology | |
CN113869339A (en) | Deep learning classification model for fault diagnosis and fault diagnosis method | |
CN107064298A (en) | The laser detecting method of blower fan blade cracks in a kind of operation | |
CN101241872B (en) | A distributed online detection system for ultrasonic lead key connection quality |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
C17 | Cessation of patent right | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20110601 Termination date: 20120925 |