Patent Application
20090251698
This invention relates to a method of identifying and determining the alignment of features on wafers or chips that are coated by a layer of transparent or partially opaque material known as over-bump applied underfill.
Flip chip technology is the fastest growing chip interconnect technology as it allows very large numbers of I/Os. Thus, the footprint of chips with low numbers of I/O's can be made very small. This is also true for associated packages such as chip-scale packages.
The major advantage of flip chip technology is that it can utilize the total chip area to make the I/O connections, while wire bonding uses only the chip periphery. Referring to
Such resin underfill material 14 can be applied by capillary flow, using a so-called no-flow process or by wafer-level applied processes. There are several wafer-level applied underfill processes, among them a Wafer-Level Underfill (WLUF) process which uses an over bump applied resin, that is then b-staged, followed by dicing the wafer to singulate chips and finally joining the chips with the WLUF layer to substrates. U.S. Pat. No. 6,919,420 Buchwalter et al which is incorporated herein by reference describes the WLUF process employed in connection with this invention.
U.S. Pat. No. 6,924,171 of Buchwalter et al., which is incorporated herein by reference, illustrates a method of flip chip joining of chips to substrates in accordance with the method of this invention.
There is the problem that an over bump applied resin layer can obscure the pattern of the electrical connect structure (C4s, copper studs, micro-bumps and similar structures) and other alignment marks making it difficult to align a wafer for dicing, a wafer for wafer to wafer joining, or a singulated chip for joining to its respective substrate or to another chip in 3-D applications. A singulated chip is a single chip produced by dicing.
To align wafer or chip, the over bump applied resin is a material which must be either transparent or translucent and the thickness of the layer must be thin enough so that the pattern is still visible. While it may be advantageous to use a thicker over bump applied resin coating, the thicker the underfill layer, the less visible are the connect structures or other alignment marks. Thus, alignment of wafer or chip under such conditions is a significant problem.
While it is often possible to align wafers and chips in which the connect structures are visible to the naked or aided eye by manual alignment, this process is undesirable in a manufacturing environment in which parts must be aligned quickly with high accuracy by an automated process.
Accordingly, a need exists for a process to recognize the exact or approximate location of one or more of the connect structures or other alignment marks which are at least partially obscured by an over-bump applied material such as used in the WLUF process. These and other needs are met by the recognition method of the invention described in the following. Other advantages of the present invention will become apparent from the following description and appended claims.
Heretofore, in the case of the surface of an over bump applied resin or over bump applied material in an application with features such as coated C4 balls the optical system has not been able to automatically recognize the diffuse location of the partially obscured features (C4 balls) and has been unable to identify the location thereof exactly or approximately.
In accordance with an aspect of this invention, a method is provided for determining feature position data to be used for alignment by employing identification of a location of a feature that is at least partially obscured by varying the depth of focus upon a feature on a work piece thereby enabling obtaining and determining a Signal to Noise Ratio (SNR) approximating a maximum value from an image captured by optical means scanning of a feature on said work piece. Preferably, the feature comprises an electrical interconnect structure such as a C4 solder ball or a copper stud bump; the feature is obscured by an over bump applied material. Preferably, employ a focus depth offset from a surface to obtain good images. Preferably, employ the results of a first alignment using a focus offset metric that can be used in subsequent alignment determination steps on either the same or subsequent samples in a same or a subsequent batch. Preferably, employ pattern recognition of a feature in best focus to determine location information as to an exact or approximate location of said feature; and subsequently use the location information for the alignment of the sample to another matched or designated pattern.
In accordance with another aspect of this invention, a method of alignment comprises determining feature position data to be used for alignment by employing identification of an approximate location of a feature that is at least partially obscured by a partially opaque material by varying focus upon a feature on a work piece; obtaining and determining a maximum Signal to Noise Ratio (SNR) from an image captured by optical means scanning of a feature on said work piece; whereby high contrast imaging of feature edges are obtained. Preferably, employ pattern recognition of a feature in best focus to determine location information as to an approximate location of said feature, and subsequently use the location information for alignment of the sample to another matched or designated pattern.
Preferably the above methods are used in combination with a mathematical determination of the location of the patterns without requiring prior learning of patterns in best focus to determine the approximate location of said features, and subsequently using this information for the alignment of the sample to another matched or designated pattern.
Preferably the above methods are used for the alignment of matching parts in such applications as wafer alignment for wafer to wafer bonding, or chip alignment for chip to wafer bonding or chip to chip bonding or chip to substrate bonding. Preferably apply the above methods to pick and place tools and/or for the purpose of wafer alignment for dicing.
In accordance with still another aspect of this invention, apparatus is provided including a computer system and software for providing alignment data to be used for alignment by employing identification of an approximate location of a feature that is at least partially obscured by varying focus upon a feature on a work piece thereby enabling obtaining and determining a Signal to Noise Ratio (SNR) approximating a maximum value from an image captured by optical means scanning of a feature on said work piece.
The invention is a process for recognizing the exact or approximate location of one or more partially obscured features or electrical connect structures such as exemplified by features such as C4 solder balls on a chip alternatively on a wafer which are at least partially coated with an over bump applied resin or material. The invention is based on a focusing routine thereby enabling an operator or a tool to obtain and determine a maximum or near maximum Signal to Noise Ratio (SNR) from an image captured by optical means. This high SNR image in turn enables fully automated alignment through either edge detection or image recognition methods. Alternatively, a SNR which is above a threshold value can be employed.
In accordance with this invention a sample coated with an over bump applied resin is loaded into an apparatus that is equipped with a sample holder, a substrate holder and an optical recognition system capable of aligning a sample to a substrate. Subsequently, the optical system focuses on the surface of the sample using traditional techniques. Next, the tool scans down through the focus until the maximum or near maximum SNR is achieved on the image capture. The maximum or near maximum SNR is achieved when the focus reaches the level of the plane where the feature sidewalls, e.g. sidewalls of solder balls, are vertical. The limited depth of focus of the optical system images a “cross section” which in the case of C4 balls is a circle. The exact or approximate location of the feature (e.g. a bump or a C4 solder ball) is now identified by determining the center point of the feature (e.g. a circle) and comparing the center point location to the location where the center point must be located for perfect alignment. Data which represents the focus offset to the surface that most accurately identifies the location of the partially obscured structure is identified and stored. That focus offset data can then be applied directly to other samples to maximize throughput since the through-focusing step requires some time.
Using the sidewall focus level the location of the features (e.g. features of a bump or a C4 solder ball), independent of the quality and thickness of the over bump applied resin layer, can be determined with a very high degree of accuracy.
An example of a method for determining alignment includes image based pattern recognition which uses learning from a first run and comparison of the images of subsequent runs to determine the center point.
Referring to
Once the location of one feature (e.g. a bump or a solder ball) is determined, at least one other feature must be sampled (within the image field or through multiple image fields) to calculate a model of the sample alignment (X, Y, Theta with 2 points, higher order alignment parameters with increasing numbers of alignment locations).
The preferred embodiment of this invention is the use of the focus scan to determine the highest SNR on the image capture and therefore the best positional performance of the alignment system. The focus offset to the surface would be stored in memory and subsequent alignment sites would be imaged using this focus offset, maximizing throughput. This embodiment maximizes the performance through process variations.
In another aspect of the method of this invention use is made of the focus scan to maximize SNR in the image capture on all sites. This is the most accurate embodiment of this invention but comes with a throughput penalty which may preclude the use of this embodiment in high-volume applications.
Still another aspect of this invention adds a focus offset from an automated focus that results in a best-guess focus on the sidewalls of the connect structures. This method is sensitive to connect structure variations such as solder ball size and shape and is not as precise as the first two embodiments; this embodiment has the greatest throughput.
This invention does not require the invention of a new optical alignment or new focusing system. Instead it works with many of the currently available optical alignment systems.
In
In step A, the system loads the sample 38 coated with a layer of applied material into a holder comprising the stage 39 in the system 30 of
In step B, the system focuses the optical system on the surface of the sample using traditional techniques of moving the stage 34 in the Z direction.
In step C, the system scans the stage up or down through the focus obtained in Step B, until the maximum SNR is achieved on the image capture in the camera 40, and the processing device 44. The maximum SNR is achieved when the focus reaches the level of the plane where the sidewalls of the feature being examined are vertical and the limited depth of focus of the optical system images the cross section.
In step D, the system performs a test as to whether contrast diminished in the previous iteration of Step C. If the answer is NO, the system branches back to step C. If the answer is YES, the system proceeds to Step E.
In step E, the system reverses the direction of scan in the Z direction and returns to step C for another scan in the reverse direction from that in the previous iteration of Step C.
In step F, the system identifies the exact or approximate location of the feature by determining the center point of the feature and the system compares this location to a location where that point has to be for perfect alignment.
In step G, the system stores the focus offset to a surface that identifies location of the partially obscured structure most accurately and then applies it directly to other samples to maximize throughput since the through-focusing step requires time
In step H, the system samples another feature in the image field or multiple image fields to calculate a model of sample alignment (x, y, theta with 2 points, higher order alignment parameters with increasing numbers of alignment locations).
In
In step A′, load a sample of over bump applied resin (OBAR) or other material coated onto a wafer or chip into a sample holder in the system with an optical recognition system for aligning the sample to a substrate on a substrate holder.
In step B′, focus the optical system on the surfaces of the bump sample using traditional techniques of moving the stage 34 in the Z direction.
In step C′, scan the stage up or down through the focus obtained in Step B until the maximum SNR is achieved on the image capture in the camera 40 and the processing device 44.
In step D′, the system tests as to whether contrast was diminished in the previous iteration of Step C. If the answer is NO, the system branches back to step E. If the answer is YES, the system proceeds to Step E′.
In step E′, the system reverses the direction of scan in the Z direction and returns to step C for another scan in the reverse direction from that in the previous iteration of Step C.
In step F′, the system identifies the exact or approximate location of a bump by determining the center point of the bump and the system compares this location to a location where that point has to be for satisfactory alignment.
In step G′, the system stores the focus offset to a surface that identifies the location of the partially obscured structure of the bump most accurately and then applies it directly to other samples to maximize throughput since the through-focusing step requires some time.
In step H′, the system samples another feature in the image field or multiple image fields to calculate a model of sample alignment (x, y, theta with 2 points, higher order alignment parameters with increasing numbers of alignment locations).
One of the advantages of this method is that no specialized alignment marks are required since the interconnect features or similar existing features themselves can be used to identify their exact or approximate position. The above methods employ a mathematical determination of the location of the patterns without requiring prior learning of patterns in best focus to determine the exact or approximate location of said features, and subsequently using this information for the alignment of the sample to another matched or designated pattern.
Another advantage of this method is that it allows to automate the process of aligning parts to each other in order to increase the throughput in a manufacturing line.
The invention is for a process of aligning a part which exhibits a partially obscured bump, e.g. a C4 solder bump, a flat topped bump, etc.
The invention has been described in terms of preferred embodiments thereof, but is more broadly applicable as will be understood by those skilled in the art. The scope of the invention is only limited by the following claims.
