• Font Size:
  • A
  • A
  • A

Feature Articles

Machine Vision Applications Grow along with Semiconductor Complexity – Part II

by Winn Hardin, Contributing Editor - AIA



In this article, we’ll pick up from Part I from the raw die stage through packaging and final board assembly, looking at changes in manufacturing processes that represent new opportunities for machine vision.

Drivers for the semiconductor and electronics industry include the moves towards 300mm wafers; ultra-thin-layer wafer construction and IC designs; and increasing manufacturing costs leading to a need for higher yields and more product tracking and automation.

Changing Colors
Front-end applications are dominated by 2D vision systems using traditional lighting techniques, but as the process moves to wafer bumping, dicing, frame attachment and packaging, many vision systems need additional contrast, height information, and spatial resolution.

New wafer designs with ultra-thin nitride, oxide and polyimide layers, among others, change the way the light reflects from the surface of the wafer. ‘‘We’re finding that most of the Intels, ST Micros and other cutting-edge microprocessor manufacturers are putting coatings on so thin that it changes the appearance of the wafer,’‘ explains  Thomas Nepstad, European marketing manager for Vision Software at Cognex Corporation (Natick, Massachusetts).  ‘‘Depending on the thickness of the coating, the wafer surface can look blue, purple or dark green. You don’t just have the traditional silver and gold wafers anymore. Many of the lighting techniques that used red lights no longer work. That’s why we’ve moved towards using infrared [IR] LEDs to handle those cases. IR works well for both standard wafers, as well as wafers with ultra-thin layers.’‘

The drive for ever smaller packages has had a far-reaching impact, from bond pads to IC contacts.  ‘‘Both bond pads and solder paste are shrinking in size, moving toward 25 micron pitch, which is very small,’‘ explains David Michael, Cognex’s Director of Core Vision R&D and Semiconductor and Electronic Applications. ‘‘They’re also moving from all square pads to different shapes, like rectangles, octagons, and other strange shapes that fit inside the circuit patterns. These pose new challenges for vision systems that locate bond pads for probers, bonders, and screen printers.’‘ On the IC end, the number of I/O ports for each IC also has grown along with the complexity of their designs. To accommodate the additional contacts, manufacturers have moved to chip-scale packages (CSP) such as ball grid arrays (BGA) and column grid arrays (CGA). These packages place the IC contacts on the bottom of the chip rather than just along the sides. ‘‘All the new CSP packages are impacting semiconductor applications,’‘ says Keith Russell, CEO of Euresys (Itasca, Illinois). ‘‘As package sizes get smaller with higher lead/ball count, it requires higher resolution images and increased processing power.’‘

Further complicating the issue is the emergence of RoHS standards and lead free solder.  ‘‘As far as RoHS goes, there’s been a significant reformation of the solder paste,’‘ continues Cognex’s Michael. ‘‘People are using different pastes and making the balls smaller. Some of the pastes you can image by changing illumination; others you need to use color imaging to discriminate between different paste materials with lower contrast than lead solder paste. Lower contrast means that lead-free pastes have less distinct edges, and using color vision with high-power processing engines can give you the data you need to make accurate measurements. If you had a single type of substrate with a single type of paste, you could play a monochrome game with careful selection of the illumination, filters and cameras. Color gives you the opportunity to handle a wider variety of substrates and pastes, making it easy to handle change over from one product to another.’‘

Going 3D
Beyond contrast, scanning laser triangulation systems can provide accurate 3D measurements of the smallest semiconductor features. For instance, to create a contact between the wafer and the die frame, wafer bumps are machined on the die to tolerances in the tens of nanometers. To check the size and location of these ‘‘bumps,’‘ manufacturers turn from high-resolution imaging to 3D laser triangulation, in which a laser beam is scanned across the camera’s field of view and displacements from the optical axis are analyzed by image processing algorithms to yield height information.

Vision Industry Looks Forward to Going Green


As the world struggles with recurring energy crises and the specter of global warming, governments, corporations and institutions are pouring funds into developing renewable energies, including photovoltaics (PV). This is good news for the planet, but also good news for vision companies familiar with semiconductor manufacturing.


Photovoltaic (PV) cells are multicrystalline semiconductors that convert sunlight into electrical energy. Like traditional semiconductors, PVs are made using crystal wafer substrates that are patterned using lithographic techniques. Unlike traditional silicon chips, however, PV cells are brittle thanks to the underlying crystal lattice mismatch, and have larger features than cutting-edge microprocessors.


'The solar market is here and now, and growing,' explains Marilyn Matz, Senior Vice President, Vision Software Business Unit Manager for Cognex (Natick, Massachusetts). 'These processes don’t require as much precision…but they have the same requirements as traditional semiconductor and electronics manufacturing for alignment, inspection and traceability. Today, PV manufacturers are using vision systems to inspect the 'fingers' and 'bus-bar' electrodes printed on each PV cell to collect the electrons. These vision systems look for breaks and defects, and inspect the edges of PV cells for cracks and chips. Vision is also being used to align PV cells as they are placed into an aluminum frame to form modules and then assembled into panels.  2D ID codes are printed on modules and panels for traceability during manufacturing.'


Matz is quick to point out that the PV market is still nascent, and may never reach the size of the microchip and electronics markets, but she concludes that PV represents a significant opportunity for the machine vision industry. Today, that opportunity is spread evenly around the industrialized nations of the globe. 'The highest demand for PV is in Europe thanks to government mandates, but the manufacturing infrastructure is global.'

‘‘We used to sell lasers to check a printed circuit board [PCB] for warpage prior to population, but today, it’s really about BGA and solder paste inspection,’‘ explains Nick Cadieux, General Manager at lighting and optics manufacturer, StockerYale link (Montreal, Canada). ‘‘Most of the time, the BGAs are inspected on a tray.  BGA chips have to be manufactured to very specific size tolerances for the connectors to work. And of course, you have to measure the solder balls on the board before you populate it with the BGA. On the solder paste front, the volume of solder paste is critical to avoid short circuits or open circuits between the chip and the PCB.’‘

According to StockerYale’s Cadieux, red laser diodes still dominate the structured light projectors used in semiconductor manufacturing, however, other wavelengths are available at lower output powers. Typically, laser line scanners used in semiconductors operate between 20 and 100 mW, but customers are looking for 500 mW solutions to increase the scanning rate, while still providing enough light for accurate measurements.

But more than color and power, beam characteristics are critically important to semiconductor and electronic laser triangulation. ‘‘A 500 mW line is useless if the beam shape is wrong,’‘ Cadieux explains. ‘‘The line needs to be very thin across the width and uniform across the entire line length.’‘ StockerYale recently introduced a Telecentric Projector for semiconductor applications that uses a proprietary optic to generate a highly stable and uniform line up to 60 mm. ‘‘It’s easy to have a thin line over a short length, but if you want to cover 1 to 2 inches per scan, the Telecentric Projector is important. The longer the line, the faster you can go, as long as you have enough power.’‘

End users and integrators should look at the beam stability, as well as the expected lifetimes when choosing a laser source, adds Cadieux.
Camera manufacturers are also optimizing their products for laser triangulation. Basler Vision Technologies’ (Ahrensburg, Germany) high-end cameras include a field programmable gate array (FPGA) that contains many common algorithms for preprocessing images at the camera, alleviating bandwidth needs while expediting image analysis. ‘‘On a custom basis, Basler can program the FPGA for customer specific algorithms, such as 3D measurement with laser triangulation,’‘ explains Henning Tiarks, Product Manager at Basler.

What to Look For
As new manufacturing technologies and processes come online in the semiconductor and electronics industries, customers are increasingly turning to machine vision to keep yields high and margins up, even as the old applications evolve and new applications emerge. ‘‘The relative importance of electronic and semiconductor applications in our revenue has completely changed in the past years. Packaging inspection applications have become relatively less important, while display inspection and AOI (PCB inspection) applications have grown tremendously. This has also contributed to more stable (less cyclic) revenue numbers,’‘ notes Euresys’ Russell.

Despite technical changes to how semiconductors and electronics are manufactured, the machine vision integrator’s approach needs to remain – in general terms – the same. ‘‘The main guideline for an integrator when designing a semiconductor inspection system is the customer’s requirements,’‘ explains Euresys’ Russell. ‘‘They have to take into consideration speed, accuracy, cost constraints and the part to be inspected. With this information the integrator will specify the appropriate camera, frame grabber, camera optics, lighting and imaging libraries.’‘

The challenge becomes obtaining the expertise to select the right vision tools from a growing toolbox of options. The good news is that the machine vision industry – which has benefited so much from the increase in microprocessor speed – is able to return the favor by producing new lighting, cameras, optics, image processing boards and software capable of keeping up with the world’s most exciting technical industry. 


Vision Apps in Electronics

Examples of back-end semiconductor manufacturing steps that often include some form of vision system for inspection or alignment include:

  • Wafer Level Testing uses vision to guide probes to each die for specific tests, or global wafer process test structures to characterize the electrical characteristics of the entire wafer.
  • Die Bonders use cameras to detect alignment marks, guide robots to place die into packages, and check die registration after placement in the package.
  • Wirebonding uses machine vision to locate metal pads on the die for automated wirebonding. Speed is a key factor in wirebonding, so the speed of the camera is also important.
  • Ball-Bonding depends on machine vision to make sure the metal bumps on the surface of the die are registered with respect to the package to ensure extremely accurate mating between the metallic balls on the die and the pads on the package. As package density increases, accuracy becomes increasingly important – so camera resolution also becomes very important.  But since speed is also critical, it is necessary to balance the need for speed against the need for high resolution. Screen printers use vision to align a stencil (or screen) with a circuit board, pushing paste through holes onto the board to hold components for reflow soldering.

Electronic board manufacturing makes extensive use of machine vision with pick and place machines to populate PCBs with discreet components, as well as component position, solder placement, and visually check for defects on finished boards.

*The author would like to thank Marty Furse, CEO of Prosilica (Burnaby, British Columbia, Canada) for his significant contributions in creating this list of common machine vision applications for semiconductor and electronics manufacturing.











There are currently no comments for this article.

Leave a Comment:

All fields are required, but only your name and comment will be visible (email addresses are kept confidential). Comments are moderated and will not appear immediately. Please no link dropping, no keywords or domains as names; do not spam, and please do not advertise.

First Name: *
Last Name: *
Your Email: *
Your Comment:
Please check the box below and respond as instructed.

Search AIA:

Browse by Products:

Browse by Company Type: