3D Machine Vision in the Semiconductor Market

In the semiconductor industry the single biggest application for 3D machine vision-based systems has been for co-planarity measurements on leaded IC packages. Co-planarity is an issue with multi-leaded active SMD’s. Both 2D and 3D measurements should be made - co-planarity (3D) and lead alignment (2D) in the case of leaded-based packages and bump co-planarity (3D) and placement (2D). Several companies are now offering systems, versions of which can handle a family of IC packages and perform co-planarity (3D) and lead alignment or bump placement (2D).  Accuracies on the order of +/- 0.00025’‘ are required. Ideally the system should be adaptable to all IC packaging designs.

3D systems do compete with those that use multiple viewing 2D-based machine vision systems. The latter are generally used to inspect components on take up reels or after marking operations. The 3D systems are generally used on higher pitch components presented in trays.

Before packaging it has been suggested that there is a requirement for ‘pre-cap’ or ‘3rd op’ inspection.  This involves check for: bond location, bond size, bond shape, wire path/position, wire loop/height, lifted wires/non-stick on die or package, wire separation, wire tail length, wire diameter, crack/tear at bond, and crossed wires. The bond itself is a 3D artifact and the loop associated with the wire itself between the two bond sites is also important. This also requires interpretation of 3D data. It is not clear whether any of the known commercially available products to perform this inspection are based on 3D techniques. It is opined that 3D is inferred from processing 2D scene data.

Flying spot scanning laser-based instruments are being used to find defects on semiconductor wafers. One might suggest that this is a 3D systems approach. The semiconductor industry also has requirements to measure film thickness . While some tactile-based profilometers are used, where the finest sensitivity is required one of the main instruments used is based on ellipsometry principles.

Triangulation-based sensors are being used in applications in the semiconductor industry: cassette mapping, protruding wafer detection, wafer edge measurement, flatness detection. 3D-based machine vision techniques are also being used to inspect BGA and CSP packages for warpage.

Companies known to be suppliers of 3D-based machine vision systems for these applications in the semiconductor industry were asked to contribute to this article. They were forwarded a questionnaire and asked for their responses accordingly. Those contributing were: 

• Rajiv Roy, Sr. Director Strategic Marketing – August Technology
• David Banitt, CEO – Nano-Or Technology
• Robert Michaels, Vice President Sales & Marketing - RVSI
• Mr. Christian Kolbe, Business Development Manager Inspection Sensors, Siemens

What are some specific applications in the semiconductor industry that your company addresses with 3D machine vision technology?

[Rajiv Roy, August Technology] Bump Inspection

[David Banitt, Nano-Or Technology] There are various applications that the technology of Nano-Or can beneficially used for:

• Copper plated patterned and non-patterned wafer (pre-CMP metrology)
• Post CMP metrology – Dishing and erosion
• Shallow Trench Isolation (STI)
• Metallized layers characterization, thickness and roll-off
• EPI Layers
• Bare Silicon for starches, slips and warp
• Back side roughness after thinning
• Bumps
• Probes’ marks

[Bob Michaels, RVSI] One specific application would be to measure the height of each and every bump across an entire flip chip wafer to ensure 100% product quality while also providing valuable data relative to upstream processes. Another application would be to verify the coplanarity of all leads on an entire lot of fully assembled surface mount chips.

RVSI provides fully automated, three-dimensional metrology and defect detection to verify the dimensional and cosmetic integrity of wafers as well as fully assembled packages and sockets. Designed for and prevalent throughout high volume manufacturing (HVM), the equipment is accurate while accurate, easy to operate and reliable in accordance with the demands of HVM.

[Christian Kolbe, Siemens] Siemens applications of 3D-based machine vision include

• solder paste balls (height, shape, existence, position etc.) on substrates and FRT boards;
• surface characteristics (flatness, waviness etc.) of i.e. ceramic substrates;
• bumps (height, position, existence, coplanarity etc.) on chips (flip chips) and panels;
• discrete components (shape, volume etc.) i.e. potentiometer; silicon lenses on wafer

Can you provide a general description of the approach your products use to arrive at 3D image data?

[David Banitt] The 3DScope2000 uses an original technology, developed by Nano-Or based on interferometry principles but without the traditional reference mirror. By using a proprietary designed optical manipulator, the wavefront returning from the measured sample is interfering with itself, thus being fully immune to vibrations in the Z-axis. A CCD camera type of design allows the sensor to use any imaging system while maintaining the nm resolution and accuracy. Thus flexibility to accommodate various applications, measurement speed and robustness (no moving parts nor scanning) make the 3DScope2000 an attractive machine vision 3D sensor.

[Bob Michaels] Laser triangulation is the foundation for RVSI’s MicroMapä 3D technique. A low power laser mounted on a precision x-y translation table fires a collimated laser beam in a series of pulses, effectively blanketing the features of interest with a multitude of laser ‘‘dots.’‘ Each individual dot provides an accurate (x,y,z) coordinate in true 3D space. Collectively, these dots render an exact special replica of the surface contour, allowing all critical measurements to be made and compared against allowable dimensional tolerances for subsequent sorting of failed product. This non-contact approach is preferable for its accuracy, breadth of measurements, speed and immunity to handling damage associated with other 3D techniques. 

[Christian Kolbe] The SISCAN sensor technology is based on the physical principle of confocal microscopy. Hence, it is a height measurement technology rather than an image acquisition system (like cameras), in that, the SISCAN sensor generates ‘‘true’‘ 3D information as opposed to 2D information from an image acquisition system that are being computed into 3D data. Confocal microscopy utilizes a point-like light source (laser light) that is being projected into a focal point and reflected from the object's surface through a pinhole into a photo detector. If the focal point hits the surface of the object the entire reflected light will pass through the pinhole, otherwise parts of the reflected light will be deflected by the pinhole. Therefore, if the focal point hits the surface the intensity of the reflected light will be maximal. The intensity peak renders the position of the focal point and thus the height value for the surface spot in question. There are many different ways to move the focal point through a height measurement range in z-direction. Moving the measurement object up and down is one way. Moving the focal point itself is another. This can be done by using i.e. a rotating Nipkov disc (with pinholes of different diameter) or moving the lens up and down with a piezo.

In any case, for industrial application requiring high throughput these methods are too slow. The patented SISCAN technology utilizes a little mirror attached to a tuning fork vibrating with a frequency of 4 kHz. The laser beam is being passed via this oscillating mirror which prompts the focal point to move up and down resulting in a measurement rate of 8000 exact height values per second as the surface detection is processed twice per one period of the sine curve's descent and rise. The SISCAN MC64 sensor operates with 64 such channels in parallel.

[Rajiv Roy]   Rapid Confocal Sensing Technology is the general basis of August’s 3D systems and with the acquisition of STI Inc came technology for laser triangulation and high-speed white light interferometry.

What are critical 3D machine vision system performance criteria?

[Bob Michaels] Capability, speed and changeover are the most critical performance factors. In order for a 3D vision system to be useful for metrology, it must furnish accurate and repeatable measurements for all applications under all conditions. As high volume manufacturing (HVM) is extremely intolerant of any factors that directly diminish productivity, high throughput and quick changeover are mandatory performance criteria. 

[Christian Kolbe] Critical factors include

• Resolution, accuracy suitable for the application at hand
• Measurement speed/throughput
• Repeatability, reproducibility

[Rajiv Roy] Critical factors include 

• Throughput  
• Accuracy/Repeatability      
• Cost     
• Tool matching

[David Banitt] Critical factors include

• Accuracy
• Speed
• Robustness
• Ease of integration in in-line systems
• Automation compatibility.

What changes have been taking place in the technologies that are the basis of 3D machine vision systems used in the semiconductor industry that has resulted in improved performance?

[Christian Kolbe] Changes I have observed include

• Increased resolution
• Accuracy through the utilization of lasers with different wave length
• More precise motion platforms (x/y table etc.) using i.e. air bearings
• Increase of measurement speed for higher throughput through combination of different measurement technologies in one inspection system (2D cameras combined with 3D measurement sensors)

[Rajiv Roy]

• Improved sensor technology - faster cameras.
• Improved x86 Processor speeds
• Improved and miniaturized imaging options - lens, laser source
• Improved staging options - lighter faster stages

[David Banitt] I have not seen much of a change or a breakthrough in the 3D technologies in the last period. There were possibly engineering related improvements that result in better performance and possibly some new applications. The Nano-Or technology does provides such capabilities that can be translated to new types of systems and applications so far not addressed in a cost effective way. The main features that allow the use of the technology in a machine vision configuration are:

• Immunity to vibration in the Z axis
• Flexibility to adopt various imaging systems (CCD camera concept)
• Minimal environmental conditions for proper operation
• Self contained design to allow simple integration into system

[Bob Michaels] Continual speed and reliability enhancements at the component and subassembly levels have allowed for faster overall throughout and extended uptime of the overall 3D vision systems. There is no reason to expect this trend to stop affording even higher throughput and tool utilization going forward.

Where do you see breakthroughs coming in the technologies that are the basis of 3D machine vision systems used in the semiconductor industry that will result in further improvements in the near future - next three years?

[Rajiv Roy]

• Delivery architectures - Current architecture of stage with wafer handler and dedicated sensor is too limiting. August is working on innovative architectures that allow modularity and flexibility of deployment. This will also improve Cost of Ownership per wafer.
• Throughput will always be important, we are driving 3D sampling rates to >32MHz z-elevation points at better than 0.2micron z-accuracy with <5micron X and Y elevation point.
• Sensors that cover all different materials and reflectivities.

[David Banitt] Have not seen such unfortunately.

[Bob Michaels] Lasers, AO (acousto-optical) modulators, PSDs (position sensitive detectors) and image processors are all seeing improvements that directly benefit the overall 3D laser vision system. While the technical enhancements of each are rather detailed, suffice it to say that they all manifest themselves in capability and speed improvements that can be realized by the customer using the tool on the production floor.

[Christian Kolbe] 

• The utilization of laser light with different wave length will reduce the size of the focal spot (below 1 micron) and, thus, render higher lateral resolution
• Different material for the tuning fork will allow for oscillation with higher frequencies resulting in even faster data acquisition
• New techniques will make the creation of multichannel sensor (128 channels) possible. again, higher throughput.

Are there market changes in the semiconductor industry that are driving the adoption of 3D machine vision?

[David Banitt] As dimensions shrink in the X and Y, they also shrink in the Z-axis. This applies both to higher requirements for flatness and surface quality, as well as the reduction in size of the bonding bumps. Stacking of dies, one on top of the other is a way to keep manufacturing cost low and still adding functionality. The handheld devices are becoming more complex and additional functions are required, while the size has to be kept small.

[Bob Michaels] We are seeing a rising demand in the requirement for dimensional verification of height, which by definition requires 3D vision. The perpetual shrink trend in semiconductors--imposed by the end products which they serve--has resulted in increasingly more stringent height specifications. Mobile electronic products, for example, are becoming perpetually slimmer and, therefore, thinner chips are needed. Only true 3D vision can ensure that the tight tolerances for thickness are being consistently met.

[Christian Kolbe] Yes, as miniaturization increases and material properties (i.e. reflectivity, roughness of surface etc.) are very diverse, ‘‘traditional’‘ methods (that is 2D camera image acquisition) are less applicable (to detect the height of a gold bump relative to a golden surface based on a 2D gray scale image is very difficult. Also, ceramic substrates with their light absorbing, light scattering rough surface are very difficult to illuminate so that good 2D camera images can be acquired). True 3D non-contact measurement technologies are required. Furthermore, more and more micro-applications are being manufactured industrially in larger volumes (see automotive electronics). Many of these applications are safety sensitive and thus required quality control. Too small and too delicate to use tactile 3D measurement systems and produced in large numbers, these application require fast optical 3D measurement sensors/systems.

[Rajiv Roy] Yes. Drive to <50u solder bumps.

How will 3D machine vision systems have to change to meet emerging applications in the semiconductor industry?

[Bob Michaels] The continual demands from customers for more cost effective solutions are in turn forcing suppliers of automated 3D vision equipment to constantly better their value proposition. RVSI responds to these challenges by extending capability to accommodate new applications; raising throughput on each subsequent generation of equipment to increase productivity; and reducing our manufacturing costs to lower system pricing. Collectively, these actions offer the better cost-of-ownership that customers expect. For example: A system that offers a 50% improvement in throughput can alternatively be thought of as being 50% less expensive.

[Christian Kolbe] Faster, more accurate, smaller (to be able to integrate into production equipment), versatile and flexibly applicable (through utilization/combination of different technologies).

[Rajiv Roy]  Systems will be flexible, modular, scalable and able to integrate into process tools.

[David Banitt] There are the usual requirements of speed, resolution and accuracy, but the cost effectiveness of implementation of the 3D measurement techniques in a MV application, the ability to measure a large variety of materials and shape pose significant challenges to these technologies.

As a supplier of 3D machine vision systems for the semiconductor industry what are some challenges you face in marketing such systems?

[Christian Kolbe] Two factors make marketing to the semiconductor industry challenging. One, the cyclical ups and downs of the semiconductor market demands a maximum of flexibility. This can become especially challenging for companies dealing with capital equipment or smaller companies whose business focus lies entirely on the semiconductor market. In terms of marketing this means that you have to have the right solution ready at the right time for the right price. Secondly, the semiconductor industry is extremely technology driven which obviously reflects into all business related to it. Products serving the semiconductor industry, thus, have to meet high technical requirements while staying cost competitive.

[Rajiv Roy] Our challenges are to make 3D inspection more affordable to our customers.

[David Banitt] Being a small company, the main challenge is not only to prove that the technology and system that you have does deliver what it is supposed to, but you need to have a strong customer support and financial strength to prove that you can support the user for long periods.

[Bob Michaels] The big swings in the global semiconductor market are especially dramatic at the capital equipment level. Consequently, reacting to the sudden demand for equipment in a timely fashion after a long downturn can present a challenge. Especially in the midst of a ramp, our customers are under extreme pressure to get product out the door, so naturally the equipment suppliers feel time-to-market deadlines as well.

What are your thoughts on the future of 3D machine vision in the semiconductor industry?

[Rajiv Roy]   As August applications span the length of the Semiconductor Manufacturing process, August has been exposed to applications ranging from inspection after Lithography, CMP and cleaning to bump and to the Test floor. August sees newer applications emerging. We are relentlessly driving to explore ways to bring affordable solutions to the manufacturer that will enable them to real-time monitor the process and improve their yields.

[Bob Michaels] 3D machine vision is very much alive, well and growing¾and will be for the foreseeable future. As highlighted in the Semiconductor Roadmap (that extends out to the year 2007), the gravitation toward flip chip ensures that the need for accurate 3D height metrology will be not be disappearing anytime soon. On the contrary, automated inspection that may have typically only required some form of 2D vision now calls for the addition of height metrology¾enabled only by 3D vision.

[Christian Kolbe] There will definitely be a demand for 3D inspection systems in the future. Especially, as micro sensors become to be a commodity product widely used in aerospace, automotive, etc. industries, QA for these safety sensitive components will become to be an issue.

What advice would you give to a company investigating the purchase of a 3D machine vision system for a semiconductor industry application?

[Bob Michaels] That’s simple: Settle on an industry leader. 3-D vision is inherently non-trivial and follows a long and precipitous learning curve. Perhaps that is why many vision equipment suppliers have either avoided the pursuit of 3D altogether, or have entered and subsequently left the market. RVSI had under its belt over a decade of 3D vision experience before entering the semiconductor market, and now has over thirteen years of additional knowledge devoted specifically to 3D metrology of semiconductors. That expertise is invaluable and is what customers have come to recognize as a prime selection criterion.

[Christian Kolbe] Make sure that you buy technology that not only suits the requirements of the application at hand, but which also holds the potential to be utilized in pre-production and production environments should your application reach volume production. You want to be sure that what you specified during the R&D phase holds valid when being in full production.

[Rajiv Roy]  Ask for experience and installed base, roadmap for improvement in throughput and displayed commitment to the market.