Binary Patterned Web Inspection

Generally, this invention relates to an automatic electro-optical scanning inspection system for detecting and locating a variety of defect conditions in a binary patterned web in which the pattern consists of an array of optically perceptible intricately designed planar surfaces which are characterized by binary signals representative of individual preselected features thereof, and comprising: a radiation source for illuminating each subject of inspection; electro-optical means scanning said subject of inspection, detecting modulated radiation from the subject of inspection, and means generating, as data output, logic pulse signals responsive to the transitions between the presence and absence of individual features in each scan of the web; a computer pre-processor receiving the data output and composing a set of binary words incorporating the results of the electro-optical scanning together with the transverse direction location address of each said feature transition encountered; means delivering said set of binary words to a computer, said computer assigning machine direction scan location data bits to each said set of delivered binary words for identifying said set of delivered binary words with its machine direction location, comparing said identified binary word sets with their counterparts stored in computer memory defining the locations of said preselected features in an acceptable product derived from a master locational pattern, inclusive of preselected tolerance limits in both machine and transverse directions; and means responsive to the computer passing or rejecting the subjects of inspection.

As a typical example, in the manufacture of film chip carriers (FCC) for microelectronic circuitry, no commercially available systems are known that can automatically inspect the intricate patterns of the affixed film chip carrier leads for defects at speeds exceeding one carrier/sec in order to distinguish such defects as missing or oversized leads, or leads which are bent or out of registration.

Current film chip carrier visual inspection techniques are slow and painstaking, and are not entirely objective when grading the product. An automatic inspection system is essential for consistent detection and location of pattern irregularities in order to obtain high acceptable product yields.

Furthermore, existing holographic or optical mask matching techniques have a number of shortcomings. Among these are: (1) insufficient signal-to-noise ratios to permit detection of very small dimensional irregularities, (2) quantitative measures of goodness of fit are either impossible to achieve or lack accuracy, (3) measurement accuracy is dependent upon the precision of physical positioning of either the word piece or the master in order to obtain registration, or (4) inability to classify defects according to type.

To overcome these shortcomings, Applicant has found that requisite inspection precision can be obtained using a computer-controlled inspection system which includes: a wide spectral band width source of illumination, an image sensor (e.g., a line scan camera) to detect the light modulated by the product, a motion encoder to provide synchronizing pulses for relating each scan to an increment of product movement in the machine direction, a computer pre-processor to convert the scan signals to digital form and which can then be used to record and store the scan line or transverse locations of the carrier lead edge transition signals, a minicomputer (1) to effect registration of the scanned work piece with the coordinates of a master pattern stored in its core memory, (2) to compare the scanned pattern with the stored "blurred" master, i.e., the single line master lead pattern together with the allowable tolerance enlargements thereof, (3) to provide a record of the existence, type and location of the defective portions of the carrier lead, and (4) to effect product rejection action as regards scans not meeting the predetermined tolerances. Lastly a marking device, actuated by computer signal, is used to identify or destroy those chips not passing inspection.

The widths of the "don't care" (i.e., it is immaterial whether the item inspected contains or lacks the ingredient inspected for) tolerance zones corresponding to product specification tolerances about the film chip carrier lead pattern edges are pre-programmed into the computer in such a way that they are easily changed when specifications require and provide assurance that each carrier lead, no matter what its relative position within the pattern, will have its own tolerance band.

The following drawings constitute part of this specification, in which:

FIG. 1 is a schematic representation of the basic components of the film chip carrier inspection system,

FIG. 1A is an enlarged view of a sample strip of 16 mm film bearing the film chip carrier patterns,

FIG. 1B is a drawing showing the waveform generated by a typical scan across two film chip carrier leads,

FIG. 1C is a drawing illustrating a scan transverse of a typical oversized carrier lead,

FIG. 1D is a drawing illustrating a scan transverse of a typical carrier lead displaced in the machine direction,

FIG. 1E is a diagram illustrating registration procedures for the apparatus of FIG. 1,

FIGS. 2A and 2B, the latter of which is a horizontal extension of FIG. 2A, are schematic diagrams of the computer pre-processor circuitry,

FIGS. 3A and 3B, the latter of which is a vertical extension of FIG. 3A, are drawings showing the waveforms corresponding to signals located at points labeled in FIGS. 1, 2A and 2B and depict the operation of the computer pre-processor circuit,

FIGS. 4A through 4E are drawings showing the disposition of tolerance zones around typical leads,

FIG. 5 is a flow chart showing the software steps used to load from off-line a minicomputer memory with data points defining a master pattern,

FIG. 5A is a drawing showing points on a typical lead pattern selected for entry into computer memory,

FIGS. 6A through 6C are flow charts showing the software steps used to process on-line the scan information during the inspection sequence,

FIG. 6D is a view in enlarged scale of the interpattern inspection zone,

FIGS. 7A and 7B constitute a flow chart showing the software steps used to inspect and classify lead defects, corresponding to the dashed portion of FIG. 6C, and

FIGS. 8A through 8E are drawings showing schematic representations of typical lead defects.

For the preferred embodiment hereinafter described, a running web is under inspection, it being understood that the invention is, of course, equally applicable to the inspection of a non-moving web in which the distance between adjacent scans is determinable.

FIG. 1 shows the basic components of the system in which an intense, wide bandwidth light source, 1, such as a quartz-iodide lamp, is used to illuminate a moving 16 mm polymeric (e.g., a polyimide) web strip product 2. FIG. 1A shows a segment of this product in enlarged scale, and also illustrates the size relationship of a typical copper-lead pattern which has been repetitively embossed into the web substrate at an earlier stage in the manufacturing process, uch as described in U.S. Pat. No. 3,968,563, as well as the spatial relationship of the web sprocket holes 2c, 2c' to the carrier lead patterns. It is seen that, whereas the carrier web base (drawn in plan convention as a transparent material) provides support for the copper film chip carrier leads 2a, which are typically 15 mils (about 0.4 mm) wide at their widest points, the end portions of these leads are quite narrow and extend unsupported into the substrate-free central apertures 2d and 2b, provided for later reception of integrated circuit chips and lead frames, respectively, and thus are susceptible to displacement from their designed positions. It will be hereinafter shown how the inspector distinguishes this latter defect from the several other detectable defect classes.

The apparatus of this invention can utilize image sensors broadly as the electro-optical inspection means, and can rely upon either reflected or transmitted radiation for detection modulation. However, for purposes of explanation, and as shown in FIG. 1, a line scan camera 3, typically a Reticon.RTM. 512 photodiode model LC200, is used to detect reflected illumination through camera lens system 4, which contains the optical components necessary to: (1) magnify the image sufficiently to cover the extent of the camera linear diode array and (2) accentuate the contrast between the "on-off" copper and "off-on" copper transition points to enhance lead edge definition. Since the particular polyimide substrate described transmits only red wavelengths, and the Reticon.RTM. camera detects radiation into the blue wavelengths, a blue optical filter is preferably used with either the source or camera optical elements to suppress the unwanted red wavelength reflectance from the non-copper surfaces.

The web transport arrangement includes constant speed driver drums 5, 5' and idler drums 6 and 6', all of which are provided with teeth to engage the web sprocket holes and thereby positively advance the web through the inspector, and over inspection drum 7, which rolls freely. A band of black felt material secured over the full periphery of drum 7 serves two purposes: (1) it provides sufficient contact friction between the inspection drum and the web riding over it to rotate the motion encoder 11 shaft without slip and without marring the product surface, and (2) it provides a contrasting background for scans across the copper leads hereinbefore mentioned. It addition, a thin strip of radiation reflective material, such as adhesive tape, can be affixed to a peripheral portion of the cylindrical surface, in order to provide contrast for the sprocket holes in the event that these are used for image registration purposes as hereinafter described.

Motion encoder 11, typically a 36,000 pulse per revolution unit, Dynamic Research Corp. Model 32-10-B14 9000 E, is coaxially attached to the inspection drum 7, and, as it turns with the drum, produces motion pulse signals used to establish precise product position in the machine direction to an accuracy of about 1 mil. (0.025 cm.).

A computer pre-processor circuit 12 electrically connects the line scan camera 3, and motion encoder 11, with a minicomputer 14. As will be hereinafter detailed with reference to FIGS. 2A, 2B, 3A and 3B, the computer pre-processor circuit 12 converts the input signals depicted in FIG. 1B to a form needed for subsequent data comparison in minicomputer 14. Finally, conventional marking device 13 (not further described) operates upon receipt of a REJECT signal from minicomputer 14, to mark, stamp or destroy those film chip carrier patterns not passing inspection.

Although minicomputer 14 is shown as a component of the preferred embodiment, microprocessor (e.g., a microcomputer) units having programmable memories are equally adapted to this inspection task. However, in order to show specific flow charts needed to operate the system, these have been described with reference to a Digital Electronics Corp. PDP 8/E computer, which is a model widely used throughout industry.

An input-output element 15, such as a teletype, communicates with the minicomputer and permits the entry of program instructions, tolerances and pattern contour points and also reads out both the location and the type of specific defect conditions detected in a readable format. A recorder records minicomputer programs and results of inspection.

Before describing the components of the logic, timing and gating circuitry of the computer pre-processor block 12, a brief overview of the inspection strategy which is used is helpful, assisted by the showings of FIGS. 1 and 1B.

As the moving web strip substrate 2 passes through the camera field of view, the charge appearing on each of camera 3's individual photo-diodes 8 in linear array during any given scan represents either an "on", an "off" or a "partly on" copper point. FIG. 1B shows the camera diode linear array oriented transversely across the images of two typical leads A and B. By appropriate thresholding, the video data signal contains pulses whose time scale widths correspond to the spatial distance between the "full off-to-full on" copper or "full on-to-full off" copper transition points in the scan. These are denoted by dashed vertical lines in FIG. 1B.

It is evident that, since the interdiode spacing in the array is known, then the lead widths in the scan direction are also determined, provided that the magnification of lens system 4 is also known. Similarly, since scan spacing in the machine direction is known, the extent of the lead configuration in a direction normal to the scan direction is determinable. After using a registration procedure, hereinafter detailed, to locate the scan starting point consistently at a known position on each inspected pattern, the problem then remains only to compare, on a scan-by-scan basis, the addresses of the diodes signaling the transitions, such as No. 11, No. 17, No. 22 and No. 31 (FIG. 1B), against the reference addresses stored in computer memory for the corresponding scan across a master pattern.

The differences between the addresses of the inspected product transition points and those of the stored reference pattern, if occurring outside prescribed tolerance zones, can then be used to classify and locate a wide variety of defect conditions. For example, the defect of FIG. 1C would be classed as either an oversized or a translated lead in the transverse direction since the transition points on a series of scans across this lead indicate that, although the left-hand and bottom edges are within tolerance, the right-hand edge is outside. The dashed lines indicate the bounds of tolerance (or "don't care") zone referred to the edges of the master lead pattern stored in computer memory. In this example, the tolerance zone consists of one machine direction zone at the lead lower edge and two transverse direction zones, i.e., at the lead left- and right-hand edges. An example of a displaced lead in the machine direction appears in FIG. 1D, which shows the machine direction transition occurring outside the horizontally oriented tolerance (or "don't care") zone, whereas the lead left- and right-hand edges are within tolerance. The following table tabulates the identifying characteristics of some of the more typical film chip carrier lead pattern defects:

The algorithm used to determine these defect conditions is hereinafter described.

The computer pre-processor circuit 12, shown in detail in FIGS. 2A and 2B, contains gating circuitry, two random access memories (RAMs), and addressing logic necessary to interface camera 3, and motion encoder 11, to the computer 14. FIGS. 3A and 3B show typical waveforms at key locations within the circuit to aid in comprehending its operation.

This circuit serves to assign transverse direction locations to the binary data received from the scanning system and transmit this information to the computer at the completion of each scan where it is then correlated with the machine direction scan position.

The primary binary signal inputs to the computer pre-processor circuit 12 include motion pulses, waveform A; video data pulses, waveform B; array clock pulses, waveform C; light level signals, waveform D; and data enable pulses, waveform E. The motion signals, waveform A, are recurrent low-going pulses representing an increment of product travel in the machine direction, typically 1 mil (0.0025 cm). These pulses are received directly from the motion encoder 11. The remaining four signals are produced by the circuitry of the Reticon.RTM. line scan camera 3.

In those cases where a line scan camera is not used, persons skilled in the art could readily provide a separate system clock and gating scheme to replace the data enable and array clock signals, waveforms E and C respectively, which are conveniently produced by the Reticon.RTM. camera.

The video data signals, waveform B, show the binary signal output when the linear diode array scans the on-copper (logic high) and off-copper (logic low) regions of a pattern such as that shown in AA (FIG. 3A) representative of two quite different product patterns. Array clock pulses, waveform C, produced by a free-running clock internal of camera 3, are used to address the elements of the scanner array and provide the synchronizing signals needed to address RAMs 16 and 17 (FIG. 2B). The light level signal, waveform D, becomes logic high at the point where the signal amplitude of the scanned linear array first exceeds a preselected camera threshold level and returns to its logic low state when the camera 3 receives an external start signal, waveform F, occurring at the beginning of each motion pulse, waveform A. The light level signal, waveform D, provides the inspector operator with the knowledge that sufficient illumination exists for inspection of the pattern about to ensue. Finally, the data enable signal, waveform E, not only controls the length of the camera scan interval across the linear diode array, but is also used to gate the synchronizing array clock pulses C for addressing, on each scan, the bit locations in the two RAMs 16 and 17 corresponding to the photodiode locations in the array. As will be hereinafter described, the brief 226 .mu.sec computer communication time, or retrace interval, between data enable pulses, waveform E, is used to permit reading-in the loaded contents of RAMs 16 and 17 into computer 14 for address comparison at the conclusion of each scan. In this way, defects which exist are quickly registered and located.

For purposes of this description, the time interval between each motion pulse, waveform A, is separated into two adjacent time periods. The first time period, T.sub.0 to T.sub.1, consists of the scan interval across the product. Simultaneously, minicomputer 14 compares the data from the previous scan with similar numbers stored in its memory that represent the boundaries of a good part and its tolerance (or "don't care") zones. The second time period, T.sub.1 to T.sub.2, constitutes the computer communication (or retrace) time. This latter period is used to load the computer with the data accumulated during the immediately preceding time period, T.sub.0 to T.sub.1. Waveform U shows the sequence of computer operation during two successive product scan periods.

Each pulse in waveform A, produced by motion encoder 11, initiates the first time interval at T.sub.0 and simultaneously triggers a 1 .mu.sec one-shot multivibrator 21 (FIG. 2A) to produce a camera external start signal, waveform F, which in turn causes camera 3 to initiate a data enable signal, waveform E. Simultaneously, the leading edge of each motion pulse, waveform A, causes one-shot 22 and inverter 23 to produce a low logic level reset signal, waveform G, which in turn resets array address up-counter 24 (FIG. 2B) to zero. Simultaneously, reset signal G will pass through inverting OR gate 19 (FIG. 2A) and inverter 20 to reset RAM address up-counter 31, waveform M.

In the course of each camera scan across the product, one-shot 25 senses the trailing edges (the "on-to-off" copper transitions) of the video pulses, waveform B, and produces edge transition pulses, waveform H. NAND gate 26 allows these pulses to pass to inverting OR gate 27 and inverter 28 only in the presence of data enable signal, waveform E. The resulting address pulses 36, 37 (or on the subsequent scan 36', 37') waveform L, are used to increment the RAM address up counter 31 provided the address 31 control signal, waveform Q, is not asserted at the control terminal of inverting AND gate 32 (FIG. 2B). In a similar fashion, after inverter 34 inverts the video pulses, waveform B, one-shot 33 senses their leading edges (i.e., the "off-on" copper transitions) and produces transition pulses, waveform I. NAND gate 35 allows these pulses to pass to inverting OR gate 27 and inverter 28 only in the presence of data enable signal, waveform E, to produce additional address pulses 41, 42 (or on the subsequent scan 41', 42'), waveform L.

Since this embodiment uses only two 16 word RAMs, 16 and 17, no more than 32 transition signals can be stored on each scan. However, there is substantially no limit to the number of RAMs which can be used, dependent on the nature of the inspected product and the number of anticipated transitions encountered in a given scan. For film chip carrier lead inspection, experience has shown that no more than about 20 transitions are typical during any given product scan.

Low logic level WRITE signal pulses 43-47 (and on the subsequent scan 43'-47'), shown in waveform N, enable RAMs 16 and 17 to record the address (diode position) of each of the combined transition pulses in waveform L. As will be evident hereinafter, pulses 47 and 47' which occur within the second time interval, are used to record address number 512 in both RAMs 16, 17 and thereby alert the computer to the fact that no more transitions have been recorded during the foregoing first time interval. Incidentally, should all 32 sequential memory locations be used, the number 512 is written into RAM address 31 on the occurrence of the 32nd transition pulse. The means used to generate the WRITE pulses comprise an inverter 51 (FIG. 2A) which inverts the high logic level data enable signal, waveform E, so that it can set RS flipflop 52 which, in turn, produces a control signal, waveform O, to operate NAND gate 53. Since the lower terminal of RS flipflop 52 is at a logic low level during the first time interval, inverting AND gate 54 allows RAM address pulses, waveform L, to pass to the second input terminal of NAND gate 53 via pulse delay unit 55. This element applies a 150 n sec delay to each of the address pulses in order to synchronize the operation of RAMs 16 and 17 with the signals from array address up-counter 24. WRITE pulses 47 and 47', waveform N, are based on the occurrence of the trailing edge of the data enable signal, waveform E, which is sensed by one-shot 56. The trigger pulse so formed, waveform J, passes through NAND gate 57, inverting OR gate 27, inverter 28 and then splits into two paths. The first path includes inverting AND gate 32 (FIG. 2B) and RAM address up-counter 31 which increments either RAM 16 or 17 to the next word position (provided the address 31, signal Q, event has not occurred). The second path includes inverting AND gate 54 (FIG. 2A), pulse delay unit 55 and NAND gate 53 which passes the 150 n sec delayed trailing edge WRITE signal N to RAMs 16 and 17 since the high logic level control signal still exists at the othe NAND gate 53 input terminal. However, the 300 n sec delayed portion of the trailing edge signal passes through NAND gate 61 and resets RS flip-flop 52 so that a low logic level control signal on NAND gate 53 effectively prevents additional transition pulses beyond the data enable trailing edge signals 47 and 47', waveform N, from being written into RAMs 16 and 17 until the leading edge of the next data enable pulse, waveform E, again sets RS flip-flop 52.

Thus far, each transition occurrence has been stored in one of the two RAMs 16 or 17 with its corresponding 10-bit diode address location. RAM 16 will accumulate the first 16 transition points; then, on the 17th count, inverter 62 (FIG. 2B) will invert the address signal to deactivate RAM 16 and simultaneously activate RAM 17 to accumulate the remainder of the transition counts up to and including a total of 32. Should additional transitions occur, these are superposed in address location 31, since NAND gates 64, 65 with inverting AND gate 66 effectively operate to disable any additional incrementing of address counter 31. In this event, an address 31 pulse would appear in the first time period and remain until the occurrence of the next RAM address reset signal, waveform M, at the end of the scan. The presence of bits in the SEL (Select Line) address input positions in either RAM serve to identify which of the two RAMs is providing the transition words when memory contents are transferred to 12-bit coupler 63, typically a DEC M1703 omnibus input interface unit, since the outputs of the two RAMs are wire-ORed together. RAM input positions labeled MSB (Most Significant Bit), LSB (Least Significant Bit), and MEM ADDRESS contain bits identifying the memory location of each 10-bit diode addressed transition signal. Bit 10, labeled END OF SCAN, is transmitted each time the 512th diode is addressed, and is used to indicate the last word in a given string of transition words and, in effect, tells the computer to disregard any additional transition pulses in the current string transmitted to it through 12-bit coupler 63. Since RAMs 16 and 17 provide for storage of 12-bit words, the remaining two bits are conveniently used to record light level and video data. By knowing the video data status, waveform B, one can determine if the current transition word represents an "off-to-on" copper or an "on-to-off" copper transition; whereas the light level bit obtained from waveform D can be used to indicate whether the sample is adequately illuminated.

It remains now to transfer the transition words stored in RAMs 16 and 17 to minicomputer 14 as controlled by 12-bit coupler 63. As hereinbefore mentioned, this is accomplished at the outset of the second time interval when initiated by the trailing edge of the data enable pulse, waveform E. In addition to generating RAM address pulse 512, waveform L, and WRITE pulse 47, waveform N, the data enable trailing edge signal initiates a low logic level read request command pulse number 1, waveform P, through one terminal of inverting OR gate 67 and inverter 68 to 12-bit coupler unit 63. The computer is programmed to delay reading the first transition word from memory by several microseconds to allow time for writing the last diode address 512 in RAMs 16 and 17 and resetting RAM address counter 31 such that a zero-bit appears in its LSB position. The first RAM address pulse to increment the RAM address counter 31 to the LSB position after reset is produced by having one-shot 71 form a pulse on the trailing edge of the reset signal for flipflop 52 after passing through inverter 72. Once formed, the pulse passes through NAND gate 73, inverting OR gate 74, inverter 75 and inverting OR gate 27, as hereinbefore described, to form address pulse 1, waveforms K and L.

After computer 14 completes reading the first transition word from RAM 16, it transmits a read done pulse 1, waveform R through 12-bit coupler 63 to one-shot 77 to initiate the next read request pulse 2, waveform P, and increment the RAM address counter 31 so that the next transition word is readied for transmission to the computer. This is accomplished in the following manner. After the first read done pulse, waveform R, passes from one-shot 77 through NAND gate 76, it travels along two paths. The first path passes the signal to a second terminal of inverting OR gate 74, which appears as pulse 2 in waveform S, and travels via inverter 75 through inverting OR gate 27 (FIG. 2A) to become RAM address pulse 2, waveform L, as previously described. The second path passes the read done pulse through inverter 81 and NAND gate 82 to the second terminal of inverting OR gate 67. The output signal of this gate passes through inverter 68 and forms the second read request pulse, waveform P.

This sequence of events continues until RAM address up-counter 31 has been incremented to address 31, which produces a high logic level address 31 pulse, waveform Q, at the output of inverting AND gate 66. Since this signal deactivates inverting AND gate 83, NAND gate 76 and inverting AND gate 32 (FIG. 2B), the read done signal pulse, waveform R, is no longer able to initiate read request commands or increment address counter 31. Consequently, no more transition words are read into the computer until the next computer communication time interval. The manner in which the transition words are used in the inspection system will be described next.

Implementation of an inspection algorithm for perfectly manufactured parts is straightforward and easily understood. Referring to FIG. 4A, which depicts a perfect lead, it is only required for the lead under test to have an "off-to-on" copper transition at identically point P.sub.1 and an "on-to-off" copper transition at location P.sub.2. Given a master drawing for a perfect part, theoretically it would only be necessary to enter the P values for all possible scans as a reference and then compare scans from the part being inspected against this reference. Any deviation in the P values of the part being inspected from the reference would constitute a defective part.

Such a simple algorithm is not practical for the inspection of film chip carriers or, in fact, manufactured parts in general. Normal manufacturing tolerances permit some variation in the lead geometry from the ideal. Excursions outside these tolerances constitute defects. Prescribed lead tolerances can be represented graphically as shown in broken line representation in FIG. 4B.

FIG. 4C shows what a manufactured lead can look like and yet still be considered acceptable with respect to the tolerances identified. Note that application of a tolerance to the ideal lead infers three inspection zones, where zone 1 constitutes an area where copper must not be found, zone 3 represents a zone where copper must be found, and zone 2 constitutes a "don't care" zone wherein it is immaterial whether copper is present or not present. The algorithms described for this invention utilize this three zone concept.

Implementation of the three zone concept requires modification of the P parameters described in FIG. 4A to take into consideration the tolerance zones. Tolerance zones can be easily implemented when scanning across leads oriented orthogonally to the scan direction, but require considerably more complexity in scanning near leads oriented in the scan direction.

In the case of lead edges oriented orthogonally to the scan direction it can be seen, by reference to FIG. 4D, that the goodness of fit of this lead can be evaluated by subtracting the tolerance value from the perfect lead location value, and verifying that the device being inspected has an edge transition value that falls between these two extremes. For the example shown in FIG. 4D, the edge value of the part being inspected, P.sub.L, must satisfy the inequality, where P.sub.1 is the perfect lead:

these calculations are straightforward and can be executed rapidly in a general purpose computer.

However, inspection of lead edges that are essentially parallel to the scan requires the manipulation of considerably more reference points. As shown in FIG. 4E, the scan occurring just prior to the perfect lead location encounters copper in the lead under test. Note that a perfect part for this scan would indicate no copper at this TD location. To determine if this copper presence is acceptable, it is necessary to examine reference scans up to a distance equal to the tolerance on either side of the present scan line. If the values of the reference scans allow either copper or no copper within this zone, then it is a "don't care" zone and the copper detected is acceptable. If, however, none of the values of the reference scans show copper, then it is a no copper zone. And lastly, if all of the values of the reference scans examined show that copper should be present, then it is a copper zone. Thus, to determine the acceptability of lead edges oriented in the direction of the scan, the computer would have to search through its memory all possible references in the tolerance zone to reach a decision on product acceptability.

With the present state of the art in computer technology, it is not possible to check the quality of lead edges oriented with the scan at speeds required for on-line inspection of film chip carriers (typically <1 second/device). An important element of this invention is the use of two algorithms, an off-line algorithm and an on-line algorithm which, in combination, permit accurate determination of the acceptability of all orientations of lead edges at on-line speeds.

A flow chart for the off-line algorithm is shown in FIG. 5. It permits one to enter into the computer the defining characteristics of the pattern to be inspected directly from a design drawing and in combination with stated machine direction tolerances. The end result is to generate a pattern reference data file that constitutes the location of copper zones, no copper zones and "don't care" zones for the machine direction tolerances. The off-line algorithm is executed only once for a given pattern and set of machine direction tolerances.

Preparation of the reference proceeds as follows: a drawing of a perfect carrier is overlaid with a grid of rectangles where each intersection in the grid represents one diode of resolution. From the drawing, equations are written that define each portion of a piecewise continuous function that describes the pattern edges. For example, in FIG. 5A the lead edges are best described using four linear equations of the form: ##EQU1## These equations define the lead edge transitions over a machine direction distance of 0-75. The equations are entered into the off-line algorithm along with the MD count interval over which they represent the respective lead edges. A new equation must be used each time a change in the position of a reference transition point in an adjacent scan line has not been predicted by the previous equation. An example of the raw data entered corresponding to the lead shown in FIG. 5A is as follows:

The off-line algorithm in turn manipulates these equations to produce the edge P values for the perfect pattern corresponding to a given MD count. For example, in the MD count interval 51 to 75, the algorithm would determine that the ideal pattern would begin copper at TD position 10 and leave copper at TD position 18.

Returning to the flow chart of FIG. 5, the operator next enters the scale factor between the design drawing measurements and the magnification of the inspector optical system. Next, the appropriate machine direction tolerance is entered. At this point, the off-line algorithm determines the machine direction "don't care" zones, compensates for system magnification and outputs a pattern reference file that describes the reference, "blurred" in the machine direction, as a function of the size of the MD tolerance.

The off-line algorithm could include the calculations necessary to take into consideration the transverse direction (TD) tolerance as well; however, it is more convenient to implement this in the on-line software. On-line implementation facilitates changing the TD tolerance as a function of position in the machine direction. In the case of film chip carrier leads aligned in the machine direction, tighter TD tolerances can then be applied to the narrow fingers in the center of the pattern, whereas looser tolerances can be applied to the leads at the extremities of the carrier.

The on-line algorithm provides the computer with a set of instructions which it needs to process the transition point data received from computer pre-processor circuit 12 after each scan. This is accomplished by comparing this data with the reference file data developed by the off-line algorithm, and, for differences falling outside of tolerance, producing error signals which can then be used to identify the defect condition and reject defective patterns.

FIGS. 6A and 6C show, in vertical progression, a set of on-line software steps providing effective flaw-detection and classification capability for the system. At the outset, after specifying the TD limits within which the sprocket hole must be detected, the first objective is to find the first sprocket hole having a length dimension in the machine direction that is within tolerance. The occurrence of this event is used to initiate two separate up-counting routines in the computer: namely, a sprocket hole count and a machine direction (MD) scan position count, both of which are incremented together with the occurrence of each motion pulse event. After both counters are initialized, the sprocket hole count is used to tell the inspection system when the next sprocket hole should be detected, and forms an MD registration reference, as hereinafter detailed, provided the dimensions of the immediately preceding sprocket hole fall within tolerance. The MD scan position count identifies the MD location of the scan line relative to the pattern.

Referring to FIG. 6D, the last of a series of N scans, after counter initialization, signals the commencement of the inspection of the next carrier lead pattern. After this event, the next series of L scans is used to determine whether the first detected lead of the new pattern is oversized or has been displaced in the machine direction or a Cu blob is present. Any of these conditions, identified as Error No. 1 and corresponding to defect condition No. 1 from the table supra, will produce a Reject signal. In a similar fashion, the failure of the system to detect the first lead within the next series of R scans signals Error No. 2, which corresponds to defect condition No. 2 (i.e., displaced, undersized or missing lead). Provided that no defects appear up to this point, and the first carrier lead edge has been detected within the maximum R scans, a wide TD tolerance value is used to detect gross defects at the lead extremities and establish TD registration points in a manner hereinafter detailed. A short delay of K scans is used to avoid the unevenness of the lead edge oriented in the scan direction when basing the initialization of the MD counter and registration measurement on the occurrence of the first scan which crosses the end of the first lead of the new pattern.

Referring back somewhat, the importance of determining accurately the position and dimensions of at least one sprocket hole per pattern in film chip carrier inspection is that sprocket holes are conveniently used as registration references throughout the fabrication steps of the carrier pattern, as detailed in U.S. Pat. No. 3,968,563. Consequently, a sprocket hole is an excellent reference point for the inspection of carrier lead patterns. However, there is no reason why any other consistent point on, or near, the pattern to be inspected cannot be used just as well.

Referring to FIG. 1E, which shows a perfectly general assumed lead pattern, registration inspection is conducted for individual product frames as follows:

X and Y axes are chosen from the reference pattern stored in the computer, conveniently at a point 0 in FIG. 1E, wherein Cu is first encountered in both the X and Y directions.

The sprocket hole counter is set to start counting at this zero origin.

As transverse counting ensues, a convenient lead, such as the middle lead L.sub.2 of FIG. 1E is employed as reference, its left- and right-hand edges being determined in a transverse sense over the full length course of scan duration, then averaged by dividing by two so that the vertical center line determined by successive transverse scans is verified. This gives the X value of the actual product pattern in terms of diode counts; i.e., the actual C.sub.3 extent, for which the reference C.sub.3 of, typically, 200, is known.

Similarly, after the continued scanning encounters the right-hand edge of a sprocket hole S.sub.1, the reference value of C.sub.1 (typically equal to 60) being known, a sensed determination for the corresponding transverse spacing gives a product value of C.sub.1 which, when added to the determined product value of C.sub.3, gives a total ascertained photodiode count which is the actual transverse dimension from the sprocket right-hand edge to the center line of lead L.sub.2.

To insure that the diode array overlaps the sprocket holes' right-hand edges during every frame scan, the scan is deliberately initiated well to the left of the right-hand edge of the first sprocket hole S.sub.1 examined. (This overlap value is filed in computer memory, so that it is assumed to remain in the same relative position through successive frames, unless, of course, the sensed value differs as scanning proceeds.)

Now, unless the scan-determined center line minus the overlap is within the corresponding reference registration total [(C.sub.1 +C.sub.3)=the allowable transverse direction registration tolerances], the product is rejected as outside transverse direction standards and inspection of the pattern terminates. This defect condition is labeled Error No. 5 in FIG. 6B.

Should the TD registration be within tolerance, all of the reference file edge transition points of the lead pattern are updated by this difference. Then, after shifting to the specified narrow TD tolerance value, the system continues to inspect the pattern in accordance with the algorithm shown in FIGS. 7A and 7B, indicated generally as the dashed box in FIG. 6C, described infra. For scans crossing sprocket holes, the system identifies those which are either incorrectly punched or missing and produces Error signals Nos. 6, 7, 12 or 13 as a result of an inspection routine which uses an algorithm analogous to that of FIGS. 7A and 7B. The existence of any of these errors will discontinue the inspection sequence and cause the system to locate the next valid sprocket hole and resynchronize the two counts before pattern inspection resumes.

Provided that the sprocket hole passes inspection, the sprocket hole count is updated in order to ensure that the vertical registration of the pattern in the machine direction is based on the most current knowledge of this reference position.

Vertical registration determination in the machine direction is effected by starting machine direction counting after a predetermined number of K scans are completed after Cu is first encountered in the course of MD scanning and then comparing this count with the sprocket hole count at the end of the pattern. Thus, the difference between the sprocket hole count, which is based on the position of the most recent valid sprocket hole, and the machine direction count is the vertical offset of the product with respect to the reference. If this off-set exceeds the reference value plus vertical tolerances, the product is rejected as outside machine direction standards, Error No. 4 (FIG. 6C). Consequently, it is not necessary to effect mechanical registration in order to inspect for registration acceptability in the X and Y directions.

Referring again to FIG. 6D, at the completion of MD registration, a series of E scans are used to determine the presence of Cu within the region between the lead ends on the carrier just inspected and transition zone S, which is provided to facilitate the assignment of defect-related signals occurring in the interpattern region to one or the other of the adjoining patterns.

The appearance of an Error No. 3 signal indicates either a discontinuity in the last inspected lead or the presence of a Cu blob. The width of transition zone S is specified by allowing S scans to occur before commencing the inspection of the next pattern at the end of the series of N scans.

Although not shown, the occurrence of an Error signal at any stage in the on-line algorithm causes the system to halt further inspection of the pattern and return to either point B or C (refer to FIG. 6A) in the algorithm to await the occurrence of the next pattern or sprocket hole, as the case may be. However, the algorithm may be modified to allow more than one Error signal to occur in any given scan or group of scans in order to further classify the nature of the defect. This will become clear when the representative lead defects shown in FIGS. 8A-8E are classified according to the inspect pattern routine of FIGS. 7A and 7B.

The object of the inspect pattern routine of FIGS. 7A and 7B is to identify the occurrence and nature of lead pattern defects by making certain comparisons between a 12 bit binary data word and a corresponding 12 bit reference word. A typical word list for the scan crossing the pattern shown in FIG. 8A is as follows:

Note that the format of the data word consists of the following information: 9 bits of TD position data, and 3 auxiliary bits (Light Level, End of Data and Video Data). The reference word format is comparable, except an MD "don't care" bit replaces the Light Level bit in the former. This bit normally appears in those reference words which define lead transition points adjacent to lead edges oriented parallel to the scan directon as derived in the off-line program. A portion of a lead end copper zone which would be described by reference words containing MD "don't care" bits is shown as a dashed rectangle in FIG. 8A. In this figure, as in the remaining FIGS. 8B-8E, TD "don't care" zones separate the hatched alternating Cu and no-Cu inspection zones and have widths which are determined by the size of the TD TOL (tolerance) values. Where the single subscript on D.sub.i or R.sub.j refers to the order of the occurrence of the data word or reference word in the list, there is used the double index subscript R values to represent the major and minor extremes of the TD "don't care" zone about the R.sub.j values in order to facilitate understanding the flow chart of FIGS. 7A and 7B.

Returning to FIG. 7A, the algorithm begins by establishing the left-most point on the scan where inspection must commence, initializes the i and j indices, and adds a D.sub.0 (Data)=0 word to the data word list.

The first decision block asks if there is an overlap in the bounds of two adjacent TD "don't care" zones. For example, where R.sub.n denotes Reference, is R.sub.11 .gtoreq.R.sub.02 ?. If this condition is not satisfied, a TD "don't care" zone would be superposed over a Cu or non-Cu inspection zone and inspection of this zone is not required. The algorithm then bypasses this inspection zone and goes on to the next. However, since FIG. 8A shows that the edges of the TD "don't care" zones are adequately separated, inspection of the first non-Cu zone can continue.

The next block asks whether the data point value is less than the associated reference value plus TD tolerance. If not, an Error No. 9 signal is produced. FIG. 8B shows the case where the right edge of an oversized lead extends into a no-copper zone. The algorithm interprets this as an Error No. 9 defect.

The next block asks if the next data point value is greater than the associated next reference value minus TD tolerance. FIG. 8C shows the case where the left edge of an oversized lead extends into a no-copper zone, which would initiate an Error No. 8 signal. Note that a provision is made to increment the data points through the routine until D.sub.i+1 >R.sub.j, or until a data overflow condition is reached in order to compensate for the large number of data words which might appear between two adjacent reference words. Such a condition arises, for example, if a number of small bits of copper are detected in a "don't care" zone, as these generate data words, but would still not constitute rejectable defects.

The next test determines whether the Video Data bit in the data word corresponds to that in the reference word. If not, a reversal of the copper with the non-copper inspection zones is indicated, and the carrier fails inspection.

Provided neither D.sub.i+1 nor R.sub.j+1 are the last words in the list, both indices are incremented by one unless an MD "don't care" bit is present. In this case, the data words are incremented by one, but the reference words are incremented by two. The reason for this can be seen by reference to FIG. 8A. Assume inspection of the edges of the first hatched no-copper zone between points R.sub.02 and R.sub.1 has just been completed. Note that there is no need to inspect the dashed MD "don't care" zone area 90 around the lead tip. It follows that the algorithm can skip to point R.sub.2 in order to begin inspecting the left-hand edge of the next no-copper zone. Note that the data words are not incremented in a like degree, since an essential data point D.sub.3 would be lost in the event of a missing lead in the MD "don't care" zone, (i.e., data words D.sub.1 and D.sub.2 would not be present). However, on the next loop through the algorithm, only the data word is incremented to satisfy the condition D.sub.i+1 .gtoreq.R.sub.j+1 - TD TOL. Note that in the event D.sub.i+1 =D.sub.End occurs before R.sub.j+1 =R.sub.End, the inequality D.sub.i+1 .gtoreq.R.sub.j+1 - TD TOL will always be satisfied, since D.sub.End is a very large value. In this case, the algorithm produces an Error signal to indicate that an invalid data word has been detected in an inspection zone.

TD tolerance value changes between inspection zones are easily accommodated by use of the decision block shown in the return path of the flow chart.

As an example, FIG. 8A shows that the TD tolerance about R.sub.End has been increased by 2 units over that which was used in the preceding inspection zone in accordance with the tabulated value.

FIG. 8D shows a leftwardly displaced lead, which initiates an Error No. 10 signal. This defect condition can be described as a combination of Error No. 8, followed immediately by Error No. 9. Although not shown, the algorithm of FIGS. 7A and 7B can easily be modified to combine sequentially the Error signals which occur in adjacent inspection zones during any given scan in order to classify other classes of defects. For example, FIG. 8E shows a rightwardly displaced lead causing an Error No. 11 signal which can be identified by the combined occurrence of Error No. 9 twice in succession.

The end of the scan across the lead is signified by the end of scan reference bit R.sub.End being set in the reference word. The remaining data words are examined for proper values until the last data word, D.sub.End, is identified by its most significant bit being set. These remaining data words are tested first to see if they represent a transition within the "dont't care" zone formed around R.sub.End and, if not, are tested to see if they occur before End Inspect. All "on-off" or "off-on" copper transitions which take place prior to or at the End Inspect point will cause an Error No. 14 to be registered. An example of such a defect is shown in FIG. 8A where a Cu blob forms a bridge between a no-Cu and a "don't care" zone. In this example, data word D.sub.6 will be flagged as an Error No. 14, but data word D.sub.5 will not, since it occurs in a "don't care" zone.

The last decision block (FIG. 7B) determines whether the edge transition which occurs beyond the End Inspection point marks the terminal edge of a copper bridge with its initial edge signal occurring in a "don't care" zone or whether it is a non-disqualifying defect. If the last data values are acceptable, a good scan has occurred and the algorithm returns for the next scan.

Although this invention treats the specific case of the inspection of film chip carrier patterns, the computer implementation of "don't care" zones around the edges of a pattern to be inspected on a scan-by-scan basis using an image sensor (e.g., line scan camera, image dissector, TV camera tube) provides a technique whereby numerous types of patterned products might be inspected for flaws, such as precision parts stamped from sheet metal or molded from plastics, printed paper or fabric rolls, or the inspection of photomasks and integrated circuits.

The following is a complete computer program word description for a Film Chip Carrier Inspector of the design hereinbefore described: ##SPC1## ##SPC2## ##SPC3## ##SPC4##