JP6465349B2 - Diagnostic method and apparatus for bonding wire current magnetic field distribution inspection of power semiconductor device - Google Patents

Description

  The present invention is a semiconductor device for electric power, especially called an IGBT (Insulated Gate Bipolar Transistor), which is a post-process of a device widely used for power supplies, inverters, etc. of electric equipment and electronic equipment of 1 kW or more, that is, by mounting a chip. The present invention relates to a technique for detecting defective products in a packaging process and preventing failure of the device in the market.

  The IGBT market is broadly divided into three fields: industrial equipment, in-vehicle equipment, and consumer equipment. The major products for industrial equipment are motor control inverter applications for trains, industrial robots, and machine tools. For in-vehicle devices, there are many applications for hybrid motor drive motors and car / air conditioner control inverters. Consumer products are mainly used for camera strobes and air conditioner inverters. In particular, the market is expanding due to IGBT demand in hybrid vehicles and electric vehicles.

  In a power semiconductor module using an IGBT, a plurality of power semiconductor devices are arranged in close contact with each other on the same substrate, and bonding wires for joining each power semiconductor device and terminals are also arranged at high density. When bonding between a power semiconductor device and a terminal with a bonding wire, bonding wire bonding failure may occur due to the state of an oxide film on the bonding portion or the bonding wire surface, a change in bonding wire mounting pressure during operation, or the like. In other words, even though it seems to be completely joined, if the joint is incomplete, the resistance of the joint varies, and the current does not flow evenly to each power semiconductor device, which is large in some power semiconductor devices. It will lead to current flowing and destroying. Further, if the bonding strength is low, there is a possibility that the bonding portion may be disconnected due to mechanical vibration or heat cycle load.

  If such a bonding wire bonding failure can be inspected after the bonding process in the manufacturing stage, failure and trouble after shipment can be prevented. In order to simultaneously measure the currents flowing through a plurality of bonding wires at the time of inspection after the bonding process, Patent Document 1 discloses a magnetic flux detection device that measures the magnetic flux generated by the current flowing through the bonding wires as many as the number of bonding wires. Sensors are disclosed that are stacked at intervals.

  Non-Patent Document 1 points out problems such as a bonding wire bonding failure occurring due to repeated switching current on / off, resulting in an increase in contact resistance of the bonded portion.

  Patent Documents 2 to 4 disclose techniques related to a bonding method. These techniques detect the chip current via the bonding wire and can only measure the total current of the chip. Further, it is a contact type measurement, and current distribution cannot be measured.

  In Patent Document 5, a material testing machine configured to perform a material test by applying a load (external force) to a test piece includes a plurality of detection magnetic detection means for detecting magnetic data of the test piece during the material test. Although a material testing machine that detects magnetic data at different locations in a test piece has been disclosed, currents flowing through a plurality of bonding wires are not simultaneously measured.

  In Patent Document 6, an energizing means for energizing a current of a predetermined frequency between a pair of electrodes of the object to be measured, and a plurality of parts of the object to be measured are positioned opposite to each other, and are generated by currents flowing through the parts. A magnetic field detecting means for detecting a magnetic field to be output and outputting a signal representing the detected magnetic field; and a frequency component extracting means for extracting a signal component having a frequency equal to the predetermined frequency from the signal representing the detected magnetic field output from the magnetic field detecting means. And a current distribution detecting means for detecting the magnitude and direction of current respectively flowing at a frequency equal to the predetermined frequency in a plurality of portions of the object to be measured from the signal component extracted by the frequency component extracting means. In the distribution measuring apparatus, one electrode of the pair of electrodes of the object to be measured has a plurality of electrode terminals respectively joined to a plurality of different positions of the object to be measured. The means supplies currents having different frequencies to the plurality of electrode terminals, and the frequency component extracting means is a signal component having a frequency equal to the different frequencies from a signal representing a detected magnetic field output from the magnetic field detecting means. A current distribution measuring device is disclosed in which the current distribution detecting means detects the magnitude and direction of the current flowing in a plurality of portions of the object to be measured for each of the different frequencies.

  However, this current distribution measuring apparatus has a detection structure in which a plurality of magnetic field sensors are arranged. However, one sensor is configured to detect two orthogonal magnetic field directions, and the power generation cell characteristics of the photovoltaic power generation panel are reduced. The purpose is to detect in the direction of the current magnetic field vector, and it cannot be applied to an application for simultaneously measuring the currents flowing through a plurality of bonding wires.

  Non-Patent Document 2 discloses a current signal distribution measuring device having a non-contact sensor having a small coil on a bonding wire of an IGBT chip in order to prevent an initial failure of a power module. There is no mention of countermeasures against impacts.

JP 2013-76569 A European Patent Publication No. 26775541 U.S. Pat. No. 8,541,892 US Patent No. 8004304 JP 2000-155083 A Japanese Patent No. 5152280

Hamidi et al, "Reliability and lifetime evaluation of different wire bonding technologies for high power IGBT modules" Microelectronics Reliability 39 (1999) 1153-1158. Tsukuda et al, "High-throughput DBC-assembled IGBT screening for power module," CIPS 2014.

  As described above, in the power semiconductor module, the power semiconductor device electrode and the package electrode are joined by a plurality of bonding wires, and the difference in thermal expansion coefficient between the power semiconductor device electrode and the package electrode In addition, a structure capable of transmitting a large current while absorbing vibration and shock when mounted on a vehicle is formed.

  Bonding wires are bonded by a bonding wire mounting device. However, if the bonding wires are defective or the contact resistance value is high due to bonding wire bonding failure, the bonding wires are diverted to other normal bonding wires.

  This change in device current shunt is intended to measure the total current flowing in the power semiconductor device in the semiconductor device pulse switching characteristics test in a very short time in the power semiconductor test process. It is impossible to detect a subtle characteristic change of the current signal flowing through the. Power semiconductor device device products that pass through these tests and have hidden defects are mounted on vehicles and various power control equipment on the market, and operate on harsh environments for a long time. The bias of the load which carries out the remarkably reduces the element lifetime of the semiconductor device for electric power, and has the danger of causing an unexpected malfunction.

  In addition, these power semiconductor device products are mounted in energy control sections such as important transport mechanisms of social infrastructure or natural energy generation while being reduced in size and integrated with high power, and their application range is expanding with great momentum. I am doing.

  For this reason, these functional failures cannot be stopped by simply stopping the function of the apparatus, but greatly affect the basis of social life.

  The present invention is a device for non-contact inspection and diagnosis by detecting a current magnetic field for abnormalities such as bias of current distribution flowing through a plurality of bonding wires and wiring paths, which are wiring materials of power semiconductor devices, and a device caused by wiring defects, etc. The purpose is to inspect, diagnose and detect the cause of hidden defects that may impair the long-term reliability of products and eliminate defective products.

In order to solve the above-described problem, a first configuration of the present invention is a bonding wire current magnetic field distribution inspection diagnostic method for a power semiconductor device in which each of a plurality of power semiconductor devices and a substrate are joined by a plurality of bonding wires. ,
A plurality of current sensors for detecting a magnetic field generated by currents flowing through the plurality of bonding wires and outputting a current corresponding to the magnetic field as a current of each bonding wire are provided,
A feature point comparison between a reference current magnetic field distribution pattern previously collected by the current sensor and a newly collected current magnetic field distribution pattern is performed, and a difference between pre-designated feature points is evaluated. A bonding wire current magnetic field distribution inspection diagnostic method for a power semiconductor device, characterized by diagnosing an abnormality when recognized.

  According to a second configuration of the present invention, in the first configuration, each current sensor including a time axis of a waveform is compared with a current magnetic field distribution pattern which is a previously collected reference and a newly collected current magnetic field distribution pattern. A sample difference integrated signal obtained by integrating the current magnetic field signal waveform difference on the time axis is plotted on the space axis of the current sensor array, and the occurrence of spatial bias is evaluated and detected as an abnormal feature, and abnormality determination is performed. And

  According to a third configuration of the present invention, in the first configuration, the difference between the current magnetic field signals of the respective current sensors is sampled with respect to the current magnetic field standard signal that is a previously collected reference and the newly collected current magnetic field inspection signal. It calculates | requires for every period, integrates | accumulates the said difference to a time axis | shaft, compares the obtained difference integrated value for every current sensor, and the determination criterion set beforehand, and performs abnormality determination, It is characterized by the above-mentioned.

  According to a fourth configuration of the present invention, in any one of the first to third configurations, the amplitude and frequency characteristics of the output signals of the plurality of current sensors are corrected in advance so that all output characteristics substantially coincide with each other. It is characterized by performing.

  According to a fifth configuration of the present invention, in any one of the first to fourth configurations, the reference current magnetic field distribution pattern and the newly collected current magnetic field distribution pattern are current magnetic fields detected by the current sensor, respectively. A signal obtained by removing environmental noise (hereinafter simply referred to as “environmental noise”), which is unnecessary magnetic field radiation noise existing in the inspection environment, which significantly disturbs the current magnetic field signal that should be collected from is used.

  According to a sixth configuration of the present invention, in any one of the first to fourth configurations, the reference current magnetic field distribution pattern and the newly collected current magnetic field distribution pattern are each an environmental noise detected by the current sensor. An original signal including the signal is used.

A seventh configuration of the present invention is a bonding wire current magnetic field distribution inspection diagnostic device for a power semiconductor device in which each of a plurality of power semiconductor devices and a substrate are joined by a plurality of bonding wires.
A plurality of current sensors for detecting a magnetic field generated by currents flowing through the plurality of bonding wires and outputting a current corresponding to the magnetic field as a current of each bonding wire;
A feature point comparison between a reference current magnetic field distribution pattern previously collected by the current sensor and a newly collected current magnetic field distribution pattern is performed, and a difference between pre-designated feature points is evaluated. An abnormality diagnosis means for diagnosing an abnormality when recognized is provided. A bonding wire current magnetic field distribution inspection diagnostic apparatus for a power semiconductor device.

According to an eighth configuration of the present invention, in the seventh configuration, the abnormality diagnosing means sets a time axis of a waveform with respect to a current magnetic field distribution pattern as a reference collected in advance and a newly collected current magnetic field distribution pattern. A difference integration means for integrating the difference of the current magnetic field signal waveform of each included current sensor on the time axis;
An abnormality determination unit that plots the sample difference integration signal integrated by the difference integration unit on the spatial axis of the current sensor array, evaluates and detects the occurrence of a spatial bias as an abnormal feature, and performs abnormality determination It is characterized by that.

According to a ninth configuration of the present invention, in the seventh configuration, the abnormality diagnosing means uses a current magnetic field standard signal as a reference collected in advance and a current magnetic field inspection signal newly collected as a current of each current sensor. A difference accumulation unit for each current sensor that obtains the difference of the magnetic field signal for each sample period and accumulates the difference on the time axis;
Comparing the obtained difference integrated value for each current sensor with a preset criterion, abnormality determination means for performing abnormality determination is included.

  The tenth configuration of the present invention is characterized in that in any of the seventh to ninth configurations, there is provided output characteristic correction means for each current sensor that performs correction in advance so that all output characteristics substantially coincide.

  According to an eleventh configuration of the present invention, in the abnormality diagnosis means according to any one of the seventh to tenth configurations, the reference current magnetic field distribution pattern and the newly collected current magnetic field distribution pattern are respectively obtained by the current sensor. A signal obtained by removing detected environmental noise is used.

  According to a twelfth configuration of the present invention, in the abnormality diagnosis means according to any one of the seventh to tenth configurations, the reference current magnetic field distribution pattern and the newly collected current magnetic field distribution pattern are obtained by the current sensor, respectively. An original signal including detected environmental noise is used.

  According to the present invention, abnormalities such as uneven distribution of current flowing in a plurality of bonding wires and wiring paths, which are wiring materials of power semiconductor devices, are inspected and diagnosed in a non-contact manner by detecting a current magnetic field, wiring defects, etc. It is possible to inspect, diagnose and detect the cause of hidden defects that may impair the long-term reliability of the device, and eliminate defective products.

It is an apparatus block diagram which shows the structural example of the electric current magnetic field distribution analyzer which implements the bonding wire electric current magnetic field distribution test | inspection diagnosis of the power semiconductor device which concerns on embodiment of this invention. It is a figure which shows the example of a structure of the module 20 which concerns on this Embodiment, and the influence of various current magnetic fields, (a) showed the arrangement | positioning of a coil with the front view with respect to the bonding wire of the semiconductor device (IGBT) 10 for electric power. (B) is a side view. It is a software functional block diagram of the electric current magnetic field distribution analysis apparatus 100 which concerns on this Embodiment. It is explanatory drawing of the correction | amendment by the single-shot rectangular-wave response waveform data for performing the characteristic correction of each current sensor output characteristic of the sensor array module 20 which concerns on this Embodiment. Currents expressed in three dimensions in the space axis direction, the signal change time axis direction, and the signal amplitude direction, which are the bonding wire arrangement directions when eight bonding normal bonding wires are measured by the sensor array module 20 of 16 channels. 2 shows a distribution pattern of magnetic field waveforms (current magnetic field distribution pattern), (a-1) is a plan view of a bonding wire, (a-2) is a plan view of current magnetic field measurement data of each current sensor output, (B-1) is a side view of the bonding wire, (b-2) is a side view of the current magnetic field measurement data, (c-1) is a perspective view of the bonding wire, and (c-2) is the current magnetic field measurement data. (D) is a front view of the current magnetic field measurement data, (e) is a side view of the bonding wire firmly bonded to the substrate, and (f) is a bonding state of (e). A current magnetic field waveform diagram showing the normal response of the. Of the eight bonding wires, when the current flowing through each bonding wire when the sixth bonding wire B6 is poorly bonded is measured by the 16-channel sensor array module 20, the current magnetic field distribution expressed three-dimensionally as described above. FIG. 4A is a plan view of a bonding wire, FIG. 2A-2 is a plan view of current magnetic field measurement data of each current sensor output, and FIG. 4B-1 is a side view of the bonding wire. (B-2) is a side view of the current magnetic field measurement data, (c-1) is a perspective view of the bonding wire, (c-2) is a perspective view of the current magnetic field measurement data, and (d) is a current magnetic field. Front view of measurement data, (e) is a side view of a bonding wire in which one of the bonding wires is poorly bonded and has resistance, and (f) is a current magnetic field waveform diagram showing the response in the state of (e). is there It is explanatory drawing which shows the cause of the dispersion | variation in which the current sensor (from CS1 to CS16) of the sensor array module 20 is influenced by other than bonding wire bonding failure when measuring the current flowing through the bonding wire. FIG. 6 is a front projection view of a three-dimensional distribution of current magnetic fields, and shows a relationship of a sample differential integration signal (c) with respect to a standard sample average signal waveform (b) when a bonding wire bonding failure occurs in the inspection sample signal (a). It is a thing. Sample differential integration signal FIG. 8C shows patterns of some defective examples as seen from the shape of the floor noise D. FIG. 8A shows a case where one peak stands out, and FIG. 8B shows a peak. (C) is close to multiple (two) peaks, (d) is close to the peak, (e) is low at the peak In this case, (f) shows a case where a low peak occurs as a whole but the floor noise level D is exceeded in some places. Is indicative of the test pulse response of bonding wires, S _T test pulse signal applied to the input gate of the power semiconductor component being tested, S _C is the output signal CS16 from the current sensors CS1 of the sensor array module 20 One representative example is shown. It is explanatory drawing which shows the setting method of the 1st determination criterion 200 and the 2nd determination criterion 201 which are the method of calculating | requiring a difference signal integrated value from a sample difference signal, and two types of determination criteria used for defect determination. 12 is an actual value plot showing an actual measurement example analyzed using the first determination criterion 200 shown in FIG. 11 and determined by the determination threshold value 1 (1σ) and the determination threshold value 2 (2σ). It is a flowchart which shows the processing step of the bonding wire electric current magnetic field distribution inspection diagnostic method of this invention performed by GO / NO GO determination function 133 of FIG.

Hereinafter, embodiments of the present invention will be specifically described with reference to the drawings.
FIG. 1 is a diagram showing a configuration of an inspection / diagnosis apparatus centering on a current / magnetic field distribution analysis apparatus 100 that performs bonding wire current / magnetic field distribution inspection / diagnosis of a power semiconductor device (IGBT) according to an embodiment of the present invention. This current magnetic field distribution analysis apparatus 100 is a current magnetic field generated by a current flowing through a plurality of bonding wires (bonding wires B1 to B8 in FIGS. 2A and 2B) of a power semiconductor device (IGBT) 10 to be inspected. An output signal from the sensor array module 20 that simultaneously measures non-contact is input, a mathematical operation is performed, an increase / decrease variation in current distribution on the bonding wire is inspected, and a manufacturing defect inherent in the power semiconductor device 10 is detected. Is detected. The series of product defect inspection determination functions described above are accommodated in the IGBT test inspection apparatus 40 and operate in parallel with the device pulse switching characteristic test for testing the performance specifications of the power semiconductor device 10.

  The current magnetic field distribution analyzer 100 receives analog signal outputs of a plurality of current sensor outputs from the sensor array module 20 (in the present embodiment, the configuration of 16 channels CS1 to CS16 in FIG. 2), and the high-speed digitizer 110. After high speed sampling, it is converted to a digital signal. This digital signal is subjected to digital mathematical calculation in the software analysis calculation processing unit 120, and after performing frequency component correction for performing preprocessing for correcting variation in the signal characteristics of each current sensor output of the sensor array module 20, the calculation is performed in advance. Measure the current distribution of the characteristic response waveform that flows through the bonding wire by the test pulse based on the judgment criteria data for which the range setting and threshold setting have been applied, extract the characteristic displacement, and determine whether the power semiconductor device is normal or abnormal Make a decision. Specifically, description will be made based on the software functional block diagram shown in FIG.

  The sensor array module 20 has a structure as shown in FIG. The coils C1 to C16 are arranged in a straight line so as to be parallel to the bonding wires B1 to B8. (B) shows the positional relationship seen from the side surface of bonding wire B1-B8 and coil C1-C16. Currents are generated in the coils C1 to C16 by the current magnetic fields BWC1 to BWC8 generated by the current flowing in the bonding wires, and are input to the amplifier circuits A1 to A16 that perform signal amplification and frequency correction for the coil-specific output characteristics to be amplified and frequency characteristics. The corrected signal is input to the high-speed digitizer 110 of the current magnetic field distribution analyzer 100 of FIG. 1 as the output of the current sensors CS1 to CS16. In the current sensor structure of the 16-channel independent series of coils C1 to C16 and the amplifier circuits A1 to A16 configured as described above, there may be a case where a subtle difference in characteristics between the channels occurs due to manufacturing variations. This variation in output characteristics hinders subsequent detection of the current distribution of the bonding wires B1 to B8, so that it is necessary to correct the variation characteristics in the software analysis calculation processing unit 120 described above.

  In FIG. 2, when the sensor array module 20 is mounted on the IGBT test inspection apparatus 40 and the bonding wire current magnetic field distribution inspection diagnosis of the power semiconductor device 10 is performed, the bonding wire current magnetic fields BWC1 to BWC8 other than the original inspection purpose are used. Fig. 1 shows an example of a situation in which various environmental magnetic field noises are caused by the structure inside the inspection apparatus. The test current terminal 50 contacts the substrate electrode terminal to which the bonding wires B1 to B8 of the power semiconductor device 10 are connected, and supplies the test current 51. The test current terminal magnetic field TTC generated by the test current 51 at this time is a sum of current magnetic field values of the bonding wire current magnetic fields BWC1 to BWC8 generated by the bonding wires (depending on the number of power semiconductor devices packaged in the case of an IGBT module). The sum of the current magnetic field values becomes larger than that) and affects the coils C1 to C16 of the sensor array module 20 as noise.

Unlike the current magnetic field in the vicinity of the bonding wire, these environmental magnetic field noises can be understood as wide-range noise components that spread over a wide range inside the inspection apparatus. At the time of measurement, if the position where the coils C1 to C16 of the sensor array module 20 are located in the vicinity of the bonding wires B1 to B8 is a reference position 52, the position is affected by the bonding wire current magnetic fields BWC1 to BWC8 and the test current terminal magnetic field TTC. Is also affected at the same time. On the other hand, when the positions of the coils C1 to C16 are displaced from the bonding wires B1 to B8 and are located at the displacement position 53, the test current terminal magnetic field TTC is affected to the same extent, but the influence of the bonding wire current magnetic fields BWC1 to BWC8. Will fade.
From the above, by subtracting the output values of the current sensors CS1 to CS16 detected at the deviation position 53 from the output values of the current sensors CS1 to CS16 detected at the reference position 52, the influence of the environmental magnetic field noise, which is a wide-range noise component, is subtracted. It is also possible.

Next, functions of the software processing configuration of the software analysis calculation processing unit 120 will be described with reference to FIG. The high-speed digitizer control function block 122 interposes a human operation interface. The high-speed digitizer 110, which is hardware, performs sampling speed setting, input scaling setting, input channel selection setting, input polarity switching, input impedance switching, input coupling AC / DC switching, the setting control such as setting input trigger level (trigger input to the first rising edge of the test pulse S _T of FIG. 10 and the sampling start.). The data capture function block 124 performs control to capture the digital signal converted by the high-speed digitizer 110 into the acquired data memory 125 that is a storage unit. The automatic range data acquisition function block 123 reads the sampled digital signal stored in the acquisition data memory 125, and the high-speed digitizer so that the sampling value of the A / D converter in the high-speed digitizer 110 is the most efficient scaling. A control instruction is issued to the control function block 122. The data acquisition function block 124 is instructed to acquire a plurality of times of sample waveform data so that it can be acquired at the optimum value of A / D conversion. The acquired digital signal is delivered to the frequency component correction function block 126 which is the next processing block.

  The frequency component correction function block 126 is sensitive to the output characteristics of the current sensors CS1 to CS16 in the sensor array module 20 of FIG. 2 for the waveform digital signal after the scale optimization is confirmed by the automatic range data acquisition function block 123. Frequency characteristic correction and gain adjustment are performed for various variations. A specific calculation method of calculation correction parameters will be described with reference to waveforms (a) to (c) in FIG.

Two calculation correction parameters for correcting the amplification degree and frequency characteristic of the outputs of the current sensors CS1 to CS16 of the sensor array module 20, that is, the amplitude correction value A _j and the attenuation time constant correction value B _j are shown in FIG. 4B created based on the output waveform data Vi of each of the current sensors CS1 to CS16 with respect to the test pulse input (a) used for correction shown in FIG. 4 and waveform data measured in advance for a normal power semiconductor device. It is obtained by performing comparison correction with the ideal reference output waveform V _ref shown. Originally, the amplification degree and frequency characteristic of each of the amplifier circuits A1 to A16 constituting each of the current sensors CS1 to CS16 of the sensor array module 20 are designed to match the reference output waveform _Vref for the test pulse input (a). However, there are subtle differences in output characteristics due to variations in the pattern forming process of the coils C1 to C16 constituting the current sensors CS1 to CS16 and subtle variations in the parts constituting the amplifier circuits A1 to A16. V _i shown in FIG. 4 (b) shows one example of the output waveforms of the respective current sensors CS1~CS16 before correction of the sensor array module 20 to be corrected.
Correction for the output waveform data V _i of the respective current sensors CS1~CS16 is derived by the following equation (1).

Here, i is a current sensor channel number, A _j is an amplitude correction value, and B _j is an attenuation time constant correction value (= ΔCR).
As shown in FIG. 4 (b), the measured waveform data V _i, or offset voltage is produced V _off for 0V reference (zero voltage), and the amplitude A _ref of the amplitude A _i is the reference waveform amplitude V _i There is a shift, the slope of the attenuation curve is different, and a shift ΔT _i occurs at the time of the falling edge. Therefore, first, as shown in FIG. 4C, after removing V _off , correcting the amplitude A _i , and performing time correction to make the time deviation ΔT _i of the falling edge zero, A _ref = A _i The amplitude correction value A _j is calculated so as to be (in the embodiment, about 0.7 to 1.3).

Next, the decay time constant B _i is corrected. Decay time constant correction value B _j is given by the difference between the constant B _i time constant B _ref and waveform data V _i at the reference output waveform V _ref as shown in FIG. 4 (c). In the case of the example of FIG. 4C, B _i is larger than B _ref , so the correction code is negative. Calculation formula including the sign of B _j is given by B _j = B _ref -B _i. In the calculation correction for the actual waveform data V _i , B _j is selected as a value such that V _ie = 0V as much as possible, as the convergence with respect to V _ref of the falling edge of the waveform. The output correction values (amplitude correction value A _i and attenuation time constant correction value B _j ) of each of the current sensors CS1 to CS16 as a result of these calculations are incorporated as calculation correction parameters of the frequency component correction function block 126 using equation (1). Each current sensor output correction calculation is performed. Note that the calculation correction parameter, that is, the correction value (amplitude correction value A _i and attenuation time constant correction value B _j ) once calculated uses the same correction value unless the sensor array module 20 is exchanged with another individual. . On the other hand, when the sensor array module is replaced with another individual, it is necessary to recalculate the correction value again according to the above procedure and apply it as a correction value unique to that individual.

The data selection function block 128 receives the output digital signal from the frequency component correction function block 126 under the control of the data storage / readout function block 130, selects input / output of the digital signal to / from the acquired data memory 125, and controls data display. The digital signal is selected and transferred to the function block 129 and the GO / NO GO determination function block 133.
The data display control function block 129 receives the digital signal output from the data selection function block 128 under the control of the data storage / readout function block 130 and converts it into various display formats shown below via a human operation interface. The two-dimensional (2D) display function G1 is color-coded for each current sensor, and performs two-dimensional overlapping display in a side view of each current sensor channel of a current magnetic field waveform that combines scaling in the amplitude direction and the time axis direction. The three-dimensional (3D) display function G2 is a display function in which two-dimensional display waveforms for each current sensor are arranged at equal intervals in a space in the physical arrangement direction for each current sensor and viewed from an obliquely upper viewpoint, as shown in FIG. This is a display function for easily grasping a characteristic waveform shape as shown in (c-2) or FIG. 6 (c-2). Further, the viewpoint is switched and the order of the current sensor channels to be displayed is switched through the operation interface. The contour line display function G3 is a two-dimensional imaging display function that colors the amplitude and is captured from above the amplitude axis. As shown in the waveform of (a-2) in FIG. 5 or (a-2) in FIG. This is a function that makes it easy to catch the signal amplitude by color-coding contour lines on a two-dimensional plane. In addition, display color can be assigned via the operation interface.

  The data storage / read function block 130 stores or reads out digital data, scaling values, sampling interval information, and other additional information necessary for calculation, etc., from the acquired data memory 125 via the human operation interface. The digital signal input source switching control is performed for the data selection function block 128, and various display waveform data converted by the data display control function block 129 are converted to the data display control function block 129 by the viewpoint angle and the like. Control is performed to store the acquired data memory 125 together with additional information such as the display current sensor channel order.

The feature point marking function block 131 converts a digital signal of arbitrary waveform data stored in advance in the acquired data memory 125 via its human operation interface into its accompanying data (sampling bit number, sampling scale value, sampling time interval, etc.). At the same time, the data is read out via the data selection function block 128, and the integration calculation interval of the digital signal calculation range used for the determination of the present invention shown in FIGS. 10 and 11 is set.
At the time of feature point marking, the feature point marking display function G4 uses the same waveform display as the contour line display information generated by the contour line display function G3 of the data display control function block 129 to A region that is likely to appear is selected and set by a region of interest marker.
The marked region of interest (integration calculation section) is stored in the comparison data memory 132 together with the digital signal of the waveform information that sets the region.

  The GO / NO GO determination function block 133 includes a digital signal of the waveform to be inspected output from the data selection function block 128, a plurality of digital signals of the basic test waveform stored in the comparison data memory 132, and region of interest information (integration calculation section). ) Is used in accordance with the software algorithm shown in FIG. 13, and “normal”, “bad”, and “suspicious” are displayed on the result display G5. The GO / NO GO determination function block 133 uses a plurality of digital signals of the inspection basic waveform, and the determination threshold value 1 (1σ) and the determination threshold value 2 (2σ) of the first determination criterion 200 shown in FIG. The determination threshold 1 (1σ), the determination threshold 2 (2σ), and the determination threshold 3 (1.5σ) of the second determination criterion 201 are calculated and stored in the comparison data memory 132, and the software algorithm of FIG. Used in each determination step.

  FIG. 5 shows three current sensors CS1 to CS16 arranged three-dimensionally in the space axis direction, signal change time axis direction, and signal amplitude direction when eight bonding normal bonding wires are measured by the sensor array module 20 of 16 channels. 2 shows a current magnetic field distribution pattern which is a distribution pattern of current magnetic field distribution waveforms represented by planes, (a-1) is a plan view of bonding wires B1 to B8, and (a-2) is a view of each current sensor channel. (B-1) is a side view of the bonding wire, (b-2) is a side view of the measurement data, (c-1) is a perspective view of the bonding wire, (c-2) Is a perspective view of the measurement data, (d) is a front view of the measurement data, (e) is a side view of the bonding wire firmly bonded to the substrate, and (f) is a bonding state of (e). Time A current magnetic field waveform diagram showing the normal response. In this normal bonding state, the three-dimensional distribution of the current magnetic field distribution waveform measurement data of all the current sensor channels shows small undulations of the bonding wires B1 to B8 as shown in the front view of (d). Thus, the signal waveform measured using the sensor array module 20 includes the time axis direction (test pulse signal change direction), the spatial axis direction (the arrangement direction of the plurality of current sensors CS1 to CS16 of the sensor array module 20), and each current sensor. The characteristics can be grasped by complex wavefront changes such as ocean waves indicated by the output signal amplitude direction (signal intensity).

  On the other hand, FIG. 6 shows a case where the current flowing through each of the bonding wires B1 to B8 when the sixth bonding wire B6 is defective in bonding among the eight bonding wires is measured by the 16-channel sensor array module 20. FIG. 3A shows a three-dimensional current magnetic field distribution waveform diagram, (a-1) is a plan view of a bonding wire, (a-2) is a plan view of measurement data of each current sensor channel, (b) -1) is a side view of the bonding wire, (b-2) is a side view of the measurement data, (c-1) is a perspective view of the bonding wire, (c-2) is a perspective view of the measurement data, (d ) Is a front view of the measurement data, (e) is a side view of one of the bonding wires having a poor resistance and a minute resistance, and (f) is a current indicating a response in the state of (e). Magnetic field waveform diagram . In this way, if there is a bonding failure in a part of the bonding wire, it will increase the resistance value, the current will decrease, which will result in a decrease in the current magnetic field waveform amplitude, and a subtle response delay of the test pulse response. It appears. By comparing this current magnetic field distribution waveform value with a preset current magnetic field distribution waveform, a defect of the bonding wire B6 can be inspected. However, in practice, as shown in FIG. 7 below, there are many mechanical alignment variations, manufacturing variations, test power supply electrodes inside the IGBT test inspection apparatus 40, inductors for test loads, and the like. Since the generated strong environmental magnetic field noise is generated as a disturbance, it is difficult to perform a defect inspection of the power semiconductor device 10 by a simple differential inspection.

Variations in the current flowing through the bonding wires B1 to B8 are caused by variations in the shapes and inclinations of the bonding wires B1 to B8, as shown in FIG. That is, not all of the bonding wires B1 to B8 are neatly arranged in a straight line, but in order to improve the “surface current” flowing in the vertical direction in the element of the power semiconductor device 10, (a− in FIG. 1) As shown in (b-1) (c-1), (a-1), (b-1), (c-1) in FIG. 6 and (a) in FIG. There are many. In addition, when the bonding wires B1 to B8 are pressure bonded to the device electrode surface or the substrate electrode terminal surface, the bonding wires B1 to B8 are not completely vertical but have a slight inclination as shown in FIG. Further, the sensor array module 20 is fixed to a device test jig inside the IGBT test inspection apparatus 40 at a position close to the bonding wires B1 to B8 of the coils C1 to C16 of the current sensors CS1 to CS16 of the sensor array module 20. There are some relative positional variations (± Δx, ± Δy, ± Δz). When such various effects affect each other in a complicated manner, as shown in FIG. 7D, even if the coils C1 to C16 of the sensor array module 20 are arranged uniformly, each of the current sensors CS1 to CS16 is arranged. The current magnetic field waveform detected by the channel varies as shown in FIG. These variations are recognized as deviation values (FIG. 7 (c-1)) of signal input when detecting the current magnetic field distribution of each current sensor. In general, these variations in relative position often have a trending signal effect on the standard signal level obtained by the output signal values of the current sensors CS1 to CS16. This is applied as floor noise determination using determination threshold 1, determination threshold 2, and determination threshold 3 of the second determination criterion 201 in FIG.
Note that the variation in the current magnetic field distribution waveform due to the mechanical relative position variation is different from the variation in the current magnetic field distribution waveform due to the bonding failure shown in FIG. Based on this, the bonding failure of the bonding wires B1 to B8 is caused by the influence of the mechanical alignment relative variation and the signals of the amplifier circuits A1 to A16 constituting the current sensors CS1 to CS16 of the sensor array module 20 of FIG. Various disturbance noises surrounding the measurement environment such as the influence of thermal noise of LSI circuits that deteriorate the noise ratio (hereinafter referred to as S / N ratio) or the influence of the test current terminal magnetic field TTC generated from the test current terminal 50 described in FIG. In order to discriminate with high accuracy from the output signals of the current sensors CS1 to CS16 accompanying the above, in the present invention, diagnosis is performed by the following method.

  In the present invention, as shown in the front view of the three-dimensional distribution of the current magnetic field shown in FIG. 8, the standard is the current magnetic field distribution of the bonding wires B1 to B8 in a normal power semiconductor device that has been obtained in advance and has no poor bonding. When the difference between the sample average signal (b) and the inspection sample signal (a) is taken, as shown by the sample difference integration signal (c), the integration of the difference integrated on the time axis (in the depth direction in the figure) is obtained. In the sample difference integrated signal (c) obtained by integrating the signal difference values in the time axis direction, it can be determined that the portion having a peak higher than the floor noise D is a bonding wire bonding defective portion. This floor noise D is mainly due to the fact that when a bonding wire bonding failure occurs, the reduced current is caused by the property of diverting to other normal bonding wires.

  FIG. 9 shows some defective patterns of the sample differential integration signal (c) shown in FIG. 8. FIG. 9 (a) shows one peak, and a large bonding wire joint is formed near the current sensor CS12. When a defect is detected, (b) has a plurality of peaks (two in the vicinity of current sensor CS6 and current sensor CS12), and (c) has a plurality of peaks (of current sensor CS10 and current sensor CS12). (D) is close to the edge of the current sensor CS1, (e) is the edge of the current sensor CS1 and the peak is low, (f) Although a very low peak occurs, the case where the floor noise level D is exceeded in the vicinity of the current sensor CS1 or CS2 or in the vicinity of the current sensors CS8 and CS10 is shown. In particular, it is difficult to determine a defect when there is a low signal level of the bonding wire bonding defective portion of the sample differential integration signal (FIG. 8C) under the conditions of (e) and (f) or a random signal as a whole. Become.

  Incidentally, in the embodiment of the present invention, the standard sample average signal [reference waveform] (b) and the test sample signal [current measurement waveform] (a) from which the signal difference is obtained are already obtained from the sensor array module 20 of FIG. As described with reference to the ambient magnetic field noise, the signal includes the ambient magnetic field noise peculiar to the device mechanism structure of the IGBT test inspection apparatus 40 that has performed signal measurement. This is a case where the disturbance of bonding wire current magnetic field distribution of each power semiconductor device is determined based on the signal difference as described above, which may cause a great disturbance noise and cause trouble. For this reason, performing the process of removing the influence of the environmental magnetic field noise in advance when acquiring the N standard waveform signal samples from which the standard sample average signal (b) [reference waveform] is calculated is the standard sample average signal ( b) [Reference waveform] There is an effect of improving the portability of data between inspection apparatuses.

  As one example, as already described with reference to FIG. 2, the reference position 52 of the current magnetic field measurement positions of the bonding wires B1 to B8 of the power semiconductor device 10 of the sensor array module 20 and the deviation away from the bonding wires B1 to B8. The difference of the same test pulse signal response is taken at the position 53.

Specifically, the method of measuring the signal waveform at the two displacement positions moves the measurement position of the sensor array module 20 to the two positions of the reference position 52 and the displacement position 53, respectively, and generates the same test pulse signal _ST . It is also possible to use a method of measuring the current and the coil position of the structure of the current sensors CS <b> 1 to CS <b> 16 of the sensor array module 20 at a reference position 52 and a displacement position 53.
Note that a method for obtaining a signal difference by providing a plurality of coil positions can be performed by analog calculation or digital calculation after the output of the current sensor, and the electromotive force with respect to the magnetic flux of each coil at the reference position 52 and the displacement position 53 can be obtained. It can also be realized by making connections so as to cancel out coil signal output polarities in different winding directions and different winding directions.
The plurality of coils C1 to C8 may be replaced with other magnetic detection elements such as a magnetoresistive element.
By taking the difference between the signal waveforms at the reference position 52 and the displacement position 53, the magnetic shield has a complex shape around the sensitive current sensor and may possibly have a detrimental effect on the current sensor detection signal. The environmental magnetic field noise signal can be canceled without adding a special structure such as a shield.

In the embodiment of the present invention, the IGBT test inspection device 40 that measures the standard sample average signal (b) [reference waveform] and the IGBT test inspection device 40 that measures the test sample signal (a) [current measurement waveform] are provided. The explanation is based on the premise that they are the same device, and the influence of environmental magnetic field noise (such as the test current terminal magnetic field TTC as an example shown in FIG. 2) on the coils C1 to C16 of the sensor array module 20 is considered to be equivalent. The calculation results of the current sensor-specific difference integrated value D _{(i, j} ) in equation (3) described later and the current sensor total difference integrated value D _{A (i)} in equation (4) described later include the environmental magnetic field. The noise is deducted and offset.
Therefore, in the embodiment of the present invention, the step of canceling noise by differentiating the heights of the current sensors CS1 to CS16 and taking the difference between them is omitted, and the reference waveform including the environmental noise is included. The measurement waveform was used. This will be described below.

The numerical value of each square shown in FIG. 11 (a) is signed at each sample time timing of the standard sample average signal [reference waveform] and the inspection sample signal [current measurement waveform] (current sensor output signal (b) in FIG. 10). This is a difference value. In the explanatory diagram, for simplification of explanation, the number is set to 1 or 4 or the like, but actually, the output values of the current sensors CS1 to CS16 (FIG. 11) for detecting the magnetic field current of the bonding wire shown in FIG. (Refer to (b-2)) (It is a difference value of about ± 1000.0 mV), and the actual displacement is about ± 300.0 mV.
The sample difference integration signal (examples of (c) in FIG. 8 and (a) to (f) in FIG. 9) is an integrated value of the integral calculation section Ti of each current sensor (CS1 to CS16) shown in FIG. 11 (c). It is the information which shows the series of each current sensor difference integrated value which is.

The advantage of integrating the difference value of each square in (a) in the integration calculation section Ti is that the instantaneous value of the timing of one sample time, that is, the value of the square in (a) is steep such as spike noise. When the change is captured, it can be leveled as a whole by time axis integration, and when a bonding wire bonding failure occurs, the characteristic change of the signal is likely to appear, that is, the difference value is likely to increase (Example) in the time integral of the first and falling Specifies the longitudinal (Ti)) near the signal S _C of FIG. 10, in being able to further improve the S / N ratio of the detection signal.

  In this embodiment, two reference values, a first determination criterion 200 and a second determination criterion 201, are provided as evaluation criteria. A method for calculating and setting each criterion will be described below.

The first determination criterion 200 in FIG. 11 (f) is calculated based on the integrated values of all the current sensors CS1 to CS16 and the integral calculation section T _i of the absolute value of the sampling difference value of each square. The purpose of the first determination criterion 200 is to evaluate the differential signals of all current sensors at once and to increase the inspection determination speed on the premise that there are not many defective bonding wire current magnetic field distribution variations.
First, an average value obtained by averaging the amplitude of an arbitrary current sensor output value at an arbitrary sampling time point with a sample number N of pre-measured power semiconductor devices that do not include a defect is obtained by Equation (2).

(X _{(i, j, k)} : Signed signal amplitude value at one sampling time point of one current sensor channel for one individual measurement of a power semiconductor device, i: 1 to N powers A value giving the number of individual semiconductor device samples j: A value giving the number of arbitrary current sensors (CS1 to CS16) from 1 to C, where the maximum is C. Integration calculation interval T from k: 1 to T _(It is a value that gives an arbitrary sampling time position of _i , and the maximum is T.)

Next, the difference between the normal N sample signals and the standard sample average signal X _{A (jk)} at an arbitrary sampling point in the integration calculation interval for each current sensor is calculated by the equation (3) and Find N variances. As shown in S _C of FIG. 10, the signal is an amplitude with polarity around the 0V reference value, and the absolute value of each signal is taken and integrated over time in the integration calculation section T _i . The integration result means the difference area between the two signals.

Next, D _{(i, j),} which is the difference integrated value for each current sensor in equation (4), is the value obtained by integrating all current sensors (CS1 to CS16), and the current sensor total difference integrated value is obtained.

Next, the deviation value of the first criterion 200 (see (f) in FIG. 11) obtained by normalizing the standard deviation values of the N normal current sensor total difference products D _{A (i)} is obtained by Expression (5). .
Based on the calculation result, the standard deviation value and the existence probability, the optimum judgment threshold value 1 which is a normal product judgment standard (in this embodiment, set as the allowable margin m = 1σ in FIG. 10 [existence probability of about 68.2% width]) is selected. To do. On the other hand, when used for determination threshold value 2 (determined as 2σ [existence probability of about 95.4% width] in the embodiment) or the like for inferior product determination, It is necessary to fully evaluate the yield in advance and set an optimal value.

Equation (6) represents the difference between the absolute value of the test sample signal (a) [currently measured waveform] and the standard sample average signal (b) [reference waveform] shown in FIG. This is a value (current sensor total difference area) integrated in the integral calculation section T _i , and is an expression for evaluating dispersion by taking the absolute value of the evaluation target. Here, if the absolute values of the waveform values of all of the current sensors CS1 to CS16 are not taken, they are added as they are, and the decreased current amount of the bonding wire bonding defective portion and the increased current portion diverted to other normal bonding wires are offset and become zero. Since there is a high possibility, in this determination, the absolute value is taken and evaluated by the total amount.

(X _{S (j, k)} : Inspection sample signal (FIG. 10A) [Signal amplitude value at any sampling time position of any sensor of [currently measured waveform])
Above, (6) test sample signal given by the equation (FIG. 10 (a)) is normal or abnormal by evaluating in the Current measured waveform] D _AS the determination threshold value 1 (1 [sigma) and the determination threshold value 2 (2 [sigma]) Make a decision.
Specifically, it is “normal” if it is within 1σ, and “abnormal” if it is 2σ or more. Defective product determination content that can be mainly determined by this determination is a pattern in which bonding wire bonding failure portions are clear as shown in FIGS. 9A to 9D (bonding wire compared to floor noise D). This is the case where the peak of the defective joint is conspicuous. In the case of between 1σ and 2σ, various methods are conceivable, but in this embodiment, processing to be taken over by the second determination criterion 201 described later is performed.

The advantage of calculating the total difference integrated value (e) of each of the current sensors CS1 to CS16 in the first determination criterion 200 is that the amplifier circuits A1 to A16 constituting the current sensors CS1 to CS16 of the sensor array module 20 are correlated with each other. The noise in the circuit such as no thermal noise can be reduced by integration. As a result, an improvement in the S / N ratio of the current sensor total difference integrated value (e) can be expected, and determination with higher determination accuracy can be realized.
Next, the second determination criterion 201 set for each of the current sensors CS1 to CS16 will be described.

  The second determination criterion 201 in FIG. 11D is an evaluation value set for each current sensor, and a more detailed evaluation is possible as compared with the first determination criterion 200, but the determination takes time accordingly. .

  Deviation values (see (d) of FIG. 11) of the individual second determination criteria 201 for each current sensor CS <b> 1 to CS <b> 16 are obtained by Equation (7) in the same manner as the procedure of the first determination criteria 200.

In the case of the present embodiment, the current sensor-specific deviation value T _Sj represented by the equation (7) has 16 channels of CS1 to CS16, so that there are 16 independent determination threshold values 1 respectively. Then, there are σ values of different standard deviation values for each current sensor. Similarly, the determination threshold value 2 has an optimum determination threshold value 2 for each current sensor (in this embodiment, 2σ, which is a different standard deviation value for each current sensor).
When used for the threshold value 2 (for example, 2σ) of the second determination criterion 201 for actual defective product determination, there may be a tendency value peculiar to the manufacturing process, which is optimum in consideration of the yield of the actual manufacturing line. It is necessary to set a correct value.

(8) the inspection sample signal (a) [Current measured waveform] a standard sample average signal (b) [reference waveform] and the absolute value of the integration calculation interval T _i at time by the current sensors of the difference shown in FIG. 10 This is the integrated variance value of the evaluation target.
As a result of integration, when the amplitude of the test sample signal (a) [currently measured waveform] is larger than that of the standard sample average signal (b) [reference waveform], the polarity of the integration result of equation (8) is + (plus )become. Conversely, when it is small, it becomes-(minus).

By evaluating the integrated value D _Sj for each current sensor output of the formula (8) with the determination threshold 1 and the determination threshold 2 of the second determination criterion 201, “normal” / “abnormal” and “determination impossible” (ambiguous) Judgment is made. The specific determination is as follows.
If it is within 1σ, it is “normal”, and if one of the current sensor (CS1 to CS16) outputs given by j is 2σ or more, it is “abnormal”. This “abnormal” determination example is an example as shown in FIG. If it is between 1σ and 2σ, it is considered that the determination is impossible (ambiguous), and floor noise evaluation is performed.

In the floor noise evaluation, evaluation of the amplitude difference D _Sj for each current sensor and the variation in polarity (value ± polarity dispersion) in the equation (8) in the “determination impossible” range of the second determination criterion 201 is performed.

When all of the current sensor output (CS1 to CS16) outputs D _Sj are biased in the same polarity direction (+ or −), the same tendency offset (characteristic bias) is set to “normal”. This “normal” determination is a determination that the currents distributed to the bonding wires (B1 to B8) of the power semiconductor device 10 have the same tendency and that there is no bias. If there is a problem in the pulse response characteristic test, the determination is followed. On the other hand, when random polarity dispersion (+ and-are mixed) in each current sensor output, if the magnitude is about 1σ to 1.5σ, it is regarded as “normal”, and random polarity dispersion with a magnitude of 1.5σ or more. If it is recognized, the tendency of characteristic bias is not seen and it is difficult to consider it as a trend offset, so it is considered “suspicious”. An example of “suspicious” determination is a pattern as shown in FIG. Note that the magnitude determination threshold value 3 (1.5σ in this embodiment) in the case of random polarity dispersion for each current sensor output is analyzed to determine the defective state of the actual production line, and the optimum value is selected. There is a need.

  The advantage of evaluating the difference integrated value (c) for each current sensor in the second determination criterion 201 is that the sample differences shown in FIG. 9 (a) to (f) are spatially distributed for each output of the current sensors CS1 to CS16. The value of the integrated signal can be evaluated as it is, and the bonding wire bonding failure can be determined with the accuracy of each current sensor.

FIG. 12 shows an actual measurement example of the first determination criterion 200. The current sensor total difference integrated value of FIG. 11 (e) is plotted against a plurality of sample waveform signals, and the standard sample average of the equation (9) is plotted. each in determination threshold value 1 and the determination threshold value 2 (example as the center value of the existence probability of the standard deviation value a current sensor overall integrated value average D _AC value 1 [sigma [existence probability of approximately 68.2% width], 2 [sigma] [about 95 .4% width existence probability]).

In the data plot of FIG. 12, there are samples that exceed 2σ (threshold value 2) among the 25 standard samples, although they are normal products. Therefore, the σ value in FIG. 12 is merely an example, and it is necessary to select and set the optimum determination threshold 1 and determination threshold 2 in consideration of the yield of the power semiconductor device.

  FIG. 13 is a flowchart showing the processing steps of the bonding wire current magnetic field distribution inspection diagnostic method of the present invention. This shows the content processed by the GO / NO GO determination function block 133 of FIG. 3 and evaluated and determined by the equations (2) to (8), and the standard sample average signal data stored in the comparison data memory 132, Evaluation threshold 1 and evaluation threshold 2 according to the first determination criterion 200, evaluation threshold 1, evaluation threshold 2 and evaluation threshold 3 according to the second determination criterion 201, and the standard sample average signal in the feature point marking function block 131 of FIG. The integration calculation interval information set in the data is used. Hereinafter, each step will be described.

Step S100: Inspection sample signals (FIG. 10 (a)) [current measurement waveform] and standard sample average signal (FIG. 10 (b)) [reference] of each current sensor (CS1 to CS16) output of the sensor array module 20 of FIG. Waveform] Create a difference.
Step S110: Integration is performed in the integration calculation section Ti in FIG. 10 of each current sensor channel difference signal (see (c) in FIG. 11).
Step S120: Integration between each current sensor channel is performed (see (e) of FIG. 11).
Above series of processes Suttepu S100~S120 means that calculates a D _AS shown in (6).

Step S130: D _AS a according to the first criterion 200 determination threshold 1 (1 [sigma) and the determination threshold value 2 (2 [sigma]) the test sample signals using a (FIG. 10 (a)) [Current measured waveform on the integration calculation section (FIG. 3, the standard deviation value evaluation is performed within the region indicated by the region-of-interest marker in the feature point marking display G4 and the section indicated by Ti in FIG.
Step S140: If it is completely within the allowable range (within 1σ), the process proceeds to Step S150, where it is determined as “normal”. If it is completely out of tolerance (outside 2σ), the process proceeds to step S180 to determine “defective”. If the determination is ambiguous (between 1σ and 2σ), the process proceeds to deviation value evaluation for each current sensor based on the second determination criterion 201 in step S160. When “normal” or “bad” can be determined in the determination in step S140, the inspection is completed as soon as possible.

Step S160: Each determination threshold value 1 (1σ) and each determination threshold value 2 (for the second determination criterion 201 set for each current sensor is set to the difference value D _Sj for each current sensor output expressed by the equation (8). 2σ) is used to evaluate the standard deviation value of each current sensor channel. The difference value D _Sj is already created for each current sensor in step S110. When the second determination criterion 201 is used, detailed evaluation can be performed within a variation range for each current sensor, but processing time is required.
Step S170: If it is completely within the permissible range (within 1σ), the process proceeds to step S150 described above, and “normal” is determined. If it is completely out of tolerance (outside 2σ), the process proceeds to step S180 described above, and is determined to be “defective”. If the determination is ambiguous (between 1σ and 2σ), the process proceeds to floor noise evaluation using the second determination criterion 201 in step S190.

Step S190: Each of the determination thresholds 1 (1σ) and each of the determination thresholds 2 (2σ) according to the second determination criterion 201 set for each current sensor by using the difference value D _Sj for each current sensor represented by the equation (8). ) And determination threshold 3 (1.5σ), and the degree of bias between 1σ and 2σ is determined to evaluate the nature of the floor noise.
Step S200: If the difference value D _{Sj for} each current sensor is random (in a state where it is divided into the + side and the − side) and is largely greater than or equal to 1.5σ of the determination threshold 3, the process proceeds to Step S210 and “suspect” judge. If the values are all biased to one of the positive and negative sides of 1σ and 2σ, or the random value is smaller than 1.5σ of the determination threshold 3, the characteristic offset tends to be small. The floor noise is considered, and the process proceeds to step S220 to determine “normal”.
Note that the determination threshold 1 (1σ in the present embodiment), the determination threshold 2 (2σ in the present embodiment), and the determination threshold 3 (present embodiment) of the second determination criterion 201 for performing the determination for each current sensor including the floor noise determination. In the example, 1.5σ) can be improved to a higher accuracy by reviewing the degree of variation in characteristics of the power semiconductor device.

As described above, the embodiment of the flowchart showing the processing steps of the bonding wire current magnetic field distribution inspection diagnostic method in FIG. 13 is an example of achieving high speed and high accuracy of determination. A flowchart of a software configuration in which the process directly proceeds to step S160 and each current sensor (CS1 to CS16) output is individually evaluated using the second determination criterion 201 is also possible.
Further, in the case of pursuing only high-speed determination, the first determination criterion 200 is provided with a determination threshold 3 (1.2σ as an example of a determination value) similar to the second determination criterion 201, and the floor noise of step 190 is determined. This can be realized by adopting the same configuration as the evaluation and configuring the flowchart so as to shift to the floor noise evaluation in step S190 when the determination is “ambiguous” in the deviation value evaluation determination in step S140.

  As described above, according to the present embodiment, an abnormality such as an uneven distribution of current flowing in a plurality of bonding wires that are wiring members of a power semiconductor device is detected in a non-contact manner by detecting a current magnetic field. In addition, it is possible to inspect, diagnose, and detect the cause of hidden defects or the like that may impair the long-term reliability of the device due to defective wiring, and eliminate defective products.

  The present invention relates to a power semiconductor device, particularly a power device widely used in power supplies, inverters, and the like of electric equipment and electronic equipment, as a technique capable of accurately determining bonding wire bonding failure during packaging of an IGBT. Failure during use of semiconductor device products can be reduced, and reliability can be significantly improved.

10 Power semiconductor devices (IGBT)
DESCRIPTION OF SYMBOLS 20 Sensor array module 40 IGBT test | inspection apparatus 50 Test current terminal 51 Test current 52 Reference position 53 Deviation position 100 Current magnetic field distribution analyzer 110 High-speed digitizer 120 Software analysis calculation processing part 122 High-speed digitizer control function block 123 Automatic range data Acquisition function block 124 Data acquisition function block 125 Acquisition data memory 126 Frequency component correction function block 128 Data selection function block 129 Data display control function block 130 Data storage / read function block 131 Feature point marking function block 132 Comparison data memory 133 GO / NO GO determination functional block 200 First determination criterion 201 Second determination criterion

Claims (12)

In a bonding wire current magnetic field distribution inspection diagnostic method for a power semiconductor device in which each of a plurality of power semiconductor devices and a substrate are bonded with a plurality of bonding wires,
A plurality of current sensors for detecting a magnetic field generated by currents flowing through the plurality of bonding wires and outputting a current corresponding to the magnetic field as a current of each bonding wire are provided,
A feature point comparison between a reference current magnetic field distribution pattern previously collected by the current sensor and a newly collected current magnetic field distribution pattern is performed, and a difference between pre-designated feature points is evaluated. A bonding wire current magnetic field distribution inspection diagnostic method for a power semiconductor device, characterized by diagnosing an abnormality when recognized.   Sample difference integration that integrates the current magnetic field distribution pattern of each current sensor, including the time axis of the waveform, with the time axis for the current magnetic field distribution pattern that has been collected in advance and the newly collected current magnetic field distribution pattern. 2. A method of bonding wire current magnetic field distribution inspection and diagnosis for a power semiconductor device according to claim 1, wherein the signal is plotted on the spatial axis of the current sensor array, the occurrence of spatial deviation is evaluated and detected as an abnormal feature, and the abnormality is determined. .   The difference between the current magnetic field signal of each current sensor is obtained for each sampling period with respect to the current magnetic field standard signal as a reference collected in advance and the newly collected current magnetic field inspection signal, and the difference is integrated on the time axis. The bonding wire current magnetic field distribution inspection diagnostic method for a power semiconductor device according to claim 1, wherein abnormality determination is performed by comparing the difference integrated value for each current sensor and a predetermined criterion.   The power according to any one of claims 1 to 3, wherein correction is performed in advance so that all output characteristics substantially coincide with amplitude and frequency characteristics of output signals of the plurality of current sensors. Method for inspection of bonding wire current magnetic field distribution of semiconductor device.   The reference current magnetic field distribution pattern and the newly collected current magnetic field distribution pattern are respectively unnecessary magnetic fields existing in the inspection environment that significantly disturb the current magnetic field signal to be collected from the current magnetic field detected by the current sensor. 5. The bonding wire current magnetic field of the power semiconductor device according to claim 1, wherein a signal from which environmental noise (hereinafter simply referred to as “environmental noise”) as radiation noise is removed is used. Distribution inspection diagnostic method.   5. The original signal including environmental noise detected by the current sensor is used for each of the reference current magnetic field distribution pattern and the newly collected current magnetic field distribution pattern. 6. A bonding wire current magnetic field distribution inspection diagnostic method for a power semiconductor device according to 1. In a bonding wire current magnetic field distribution inspection diagnostic device for a power semiconductor device in which each of a plurality of power semiconductor devices and a substrate are joined by a plurality of bonding wires,
A plurality of current sensors for detecting a magnetic field generated by currents flowing through the plurality of bonding wires and outputting a current corresponding to the magnetic field as a current of each bonding wire;
A feature point comparison between a reference current magnetic field distribution pattern previously collected by the current sensor and a newly collected current magnetic field distribution pattern is performed, and a difference between pre-designated feature points is evaluated. An abnormality diagnosis means for diagnosing an abnormality when recognized is provided. A bonding wire current magnetic field distribution inspection diagnostic device for a power semiconductor device. The abnormality diagnosis means includes
Difference accumulation means for integrating the difference between current magnetic field signal waveforms of each current sensor including the time axis of the waveform with respect to the current magnetic field distribution pattern as a reference collected in advance and the newly collected current magnetic field distribution pattern along the time axis When,
An abnormality determination unit that plots the sample difference integration signal integrated by the difference integration unit on the spatial axis of the current sensor array, evaluates and detects the occurrence of a spatial bias as an abnormal feature, and performs abnormality determination 8. The bonding wire current magnetic field distribution inspection diagnostic apparatus for a power semiconductor device according to claim 7. The abnormality diagnosis means includes
A current sensor that obtains a difference between current magnetic field signals of each current sensor for each sampling period with respect to a current magnetic field standard signal that is a previously collected reference and a newly collected current magnetic field inspection signal, and integrates the difference over a time axis Another difference accumulation means;
8. The bonding wire current of the power semiconductor device according to claim 7, further comprising: an abnormality determination unit that compares the obtained difference integrated value for each current sensor with a predetermined determination criterion to perform abnormality determination. Magnetic field distribution inspection diagnostic equipment.   10. The output characteristic correction means for each current sensor for correcting in advance so that all output characteristics substantially coincide with the amplitude and frequency characteristics of the output signals of the plurality of current sensors. The bonding wire current magnetic field distribution inspection diagnostic apparatus for a power semiconductor device according to any one of the above.   8. The abnormality diagnosis unit, wherein the reference current magnetic field distribution pattern and the newly collected current magnetic field distribution pattern each use a signal from which environmental noise detected by the current sensor is removed. The bonding wire current magnetic field distribution inspection diagnostic apparatus for a power semiconductor device according to any one of 10 items.   In the abnormality diagnosis means, the reference current magnetic field distribution pattern and the newly collected current magnetic field distribution pattern each use an original signal including environmental noise detected by the current sensor. The bonding wire current magnetic field distribution inspection diagnostic apparatus for a power semiconductor device according to any one of 10 items.