AIA Logo

Member Since 1984


AIA - Advancing Vision + Imaging has transformed into the Association for Advancing Automation, the leading global automation trade association of the vision + imaging, robotics, motion control, and industrial AI industries.

Content Filed Under:

Semiconductor Semiconductor

Visual Inspection & Testing Visual Inspection & Testing

Vision Crystallizes Around Semiconductor Market

POSTED 02/06/2008

 | By: Winn Hardin, Contributing Editor

The semiconductor equipment market has always been a hungry consumer of machine vision technology. With each new cycle of semiconductor manufacturing equipment comes an opportunity to sell new quality assurance systems to OEMs and end users to help optimize their productivity and yield.

In the last few years, new drivers and disruptive events to the global semiconductor supply chain have added to the list of machine vision opportunities in the semiconductor and electronics markets. These drivers include – but are not limited to -- the move to finer-pitch devices, requiring new levels of resolution, speed and bandwidth to maintain throughput and profit margins, and ultra-thin film coatings leading to greater use of color lighting and cameras for inspection in a traditionally monochrome industry. High-value ICs are driving serialization throughout the supply chain, while new chip-scale packages (CSP) and regulatory changes such as Europe’s Restriction on Hazardous Substances (RoHS) are posing a host of back-end challenges that are often mitigated by vision systems quality controls (see: Vision Apps in Electronics).

Looking High and Low

‘‘When semiconductor equipment manufactures started using vision it was primarily for alignment, because they couldn’t manufacture wafers without it,’‘ explains Marilyn Matz, Senior Vice President, Vision Software Business Unit Manager for Cognex (Natick,  Massachusetts). ‘‘As manufacturers mature, they are moving to add value by integrating inspection into their machines. In addition, the demand for traceability to improve manufacturing yield continues to grow.’‘

‘‘It’s not just tracking whole wafers anymore,’‘ adds John Petry, head of Semiconductor and Electronics marketing at Cognex. ‘‘More and more data matrix codes are being put on individual IC packages as fabs build up the infrastructure to track components through the electronics assembly process, which includes shipping them around the world for packaging and test.  It’s starting with high-value ICs such as processors, not DRAMs. And while wafer marking is now standardized with SEMI OCR and T7 Data Matrix codes, IC marking is still wide open in terms of marking specs.’‘

‘‘A related trend is the wafer ID retrofit business,’‘ notes Cognex’ Matz. ‘‘Companies like Texas Instruments and Qimonda see the value of wafer ID on their new 300 mm wafer manufacturing equipment, so they’re upgrading 200 mm tools with new wafer ID readers. Upgrading older equipment with modern ID readers can raise read rates from 70 or 80% up to 99+%, and it’s much more cost-effective than buying a new prober or other tool.’‘

The increasing use of high-throughput wafer sorters and handlers is another driver that may increase the use of vision in fabs. ‘‘We have found, particularly in high-throughput equipment, that existing mechanical pre-alignment methods are too slow,’‘ adds Cognex’ Petry. ‘‘Adopting a machine-vision-based pre-aligner, such as our recently released In-Sight 1820, offers distinct speed advantages.’‘

Where mechanical pre-alignment might take several seconds to align a wafer, a vision-based solution needs only half a second to locate the wafer center and alignment notch. Because it doesn’t need to spin the wafer, vision-based pre-alignment results in less handling and higher throughput. It can also handle wafer types that challenge mechanical systems, such as thinned wafers mounted on glass substrates.

According to Henning Tiarks, Product Manager for camera manufacturer Basler Vision Technologies  (Ahrensburg, Germany), ‘‘what we’re seeing is that there are two types of cameras requested by the semiconductor and electronics industry, which are nearly the same market. One is the need for very high performance cameras such as our A400, where the customer comes to us looking for specific algorithms programmed on the Altera FPGA inside our high-resolution cameras up to 4 megapixels at 96 fps. Lower end applications using low to medium-resolution cameras like our Scout and Pilot camera families for wire bonding and alignment can use many different CMOS or CCD sensors. These applications are often handled through our integrators.’‘ 

Even analog cameras have their place on the low end of the resolution/application spectrum. ‘‘The semiconductor industry is a bit unique in the use of analog cameras,’‘ explains Bob Wolfe, Director of Sales at frame grabber vendor, BitFlow Inc. (Woburn, Massachusetts). ‘‘In terms of what we’re shipping – and not necessarily what we’re developing for that market – it appears that the semiconductor industry is still largely analog, but I think that’s changing as we speak. Price still appears to be very important to Asian semiconductor markets.’‘

To improve the capability of existing analog machines and vision system retrofits for low-resolution applications, BitFlow has developed the ALTA-AN family of frame grabbers. There are three members to the family, supporting one, two or four cameras. Each camera input has one AFE (Analog Front End). The AFE is a single IC that has all the signal conditioners – PLL, three A/D's, clamping, gain/offset and synch extraction. By placing these discreet components on a single piece of silicon, noise is reduced and the quality of the analog image is improved. Each AFE can independently support one RS170, progressive scan or RGB camera.

Wanted: High-Resolution and Speed

At the front-end of semiconductor manufacturing, prior to dicing the exposed wafer into dies for mounting to lead frames and final packaging, high resolution cameras help (see: provide precise alignment of photolithographic masks and quality checks for critical dimensions in the chip design. As feature sizes on microchips continue to shrink with each new generation of manufacturing equipment, the need for higher resolution quality assurance inspection systems increases. However, increasing manufacturing costs also mean that the entire process cannot slow down if vendors want to maintain margins.

‘‘The trend has been toward larger imaging arrays in cameras, to provide sufficient resolution over a large field of view,’‘ explains Marty Furse, CEO of GigE Vision™ camera maker, Prosilica (Burnaby, British Columbia, Canada), ‘‘but speed [frame rate] is also very important in order to achieve higher manufacturing throughput. As packages become denser, camera resolution requirements increase.

‘‘In some cases,’‘ continues Prosilica’s Furse, ‘‘to achieve a large field of view, lower resolution cameras are used and mechanically moved to create a large image made up of smaller ‘tiles’.  This method creates a high-resolution image, but this slows down the process and can introduce particulate contamination due to the mechanical movement. In general, it is more effective to use a high-resolution sensor that captures the whole image in a single shot. This generally requires a large sensor array, and low distortion optics.
Stitching together images from multiple cameras is also used instead of mechanically moving a single camera, but demarcation artifacts between cameras can be difficult to overcome. In this case, and the tiled case, optical distortion is also very problematic.’‘

To handle high resolution images at high frame rates, both Prosilica and Basler build GigE into their camera I/O. Basler’s Pilot piA2400, 5-megapixel camera, for instance, uses GigE to allow the camera to generate 12 fps.

‘‘Cost-performance trade-offs for both the camera and the optics is probably the biggest issue for an integrator who is designing a semiconductor inspection system,’‘ concludes Prosilica’s Furse. ‘‘The challenge is that many semiconductor applications require both a large field of view and high resolution, which generally means choosing a large sensor – but this means larger, more expensive optics, too.  To resolve this issue, there has been a trend toward using smaller pixel sizes in order to achieve high-resolution in a smaller sensor format allowing lower-cost optical solutions.  For example, Prosilica’s GC2450 has very high resolution in a relatively small sensor format; the GC2450’s 2/3’‘ 5-megapixel sensor allows for a wider range of high-resolution optics than is available for larger sensor formats while still offering a very high-resolution solution.’‘ Many camera vendors have developed special calibration routines that are integrated into their front-module assembly processes to further improve alignment between sensor and optic, guaranteeing users the most defect- and distortion-free image possible.

Finally, vision suppliers are also designing algorithms with the semiconductor industry in mind. As chip features become smaller, traces on an integrated circuit can vary considerably because it’s more difficult to hold the trace contour to an exact location. Despite the variation in the trace, however, the vision system must be able to locate the ‘‘mouse bites’‘ and other defects that can lead to short circuits and faults. Cognex has developed smart algorithms that can accommodate significant deformation of a feature while still locating and identifying critical defects.

‘‘These lessons don’t just apply to semiconductor manufacturing, but also to printing, etching and dispensing in other industries,’‘ adds David Michael, Cognex’ Director of Core Vision R&D and Semiconductor/Electronic Applications. ‘‘We have to understand the physics of solder printing and glue dispensing in electronics assembly, but it doesn’t mean the vision algorithms only apply to die bonding, for example. There are customers that are printing date and lot codes on pharmaceutical bottles that have similar requirements.’‘

‘‘A good analogy would be search in the early days of machine vision,’‘ explains Cognex’ Matz. ‘‘Pattern finding technology has evolved from using intensity-based models to geometric-based pattern matching using edge and feature information rather than region information. Geometric pattern finding is better able to handle scale, rotation, and non-linear intensity changes. Our latest algorithms increase the degrees of freedom that the algorithms can understand. Now the state-of-the art pattern finding algorithms can also locate features on non-planar surfaces, and warped or flexible surfaces.’‘

Once limited to gross alignment, wafer tracking, planarity and off-line trace inspection, front-end semiconductor applications for machine vision continue to grow as chip designs become more complex and manufacturing costs increase.  ‘‘The electronics and semiconductor markets are still one of the biggest drivers for vision technology,’‘ adds Basler’s Tiarks, and that’s not likely to change.

In Part II of this series on machine vision in the semiconductor industry, we’ll look at back-end and electronics board assembly applications, including developments in structured and unstructured lighting, the impact of color on the semiconductor industry, and particular challenges posed by RoHS and the latest chip scale packages.