In this short report, we will review the current exploitation of machine vision in the electronics industry. We will do this both from the perspective of existing and emerging markets and from a scientific and technological perspective. Throughout, we will concentrate on one of the largest sectors in the industry: printed circuit board inspection.
The electronics and printed circuit board industry comprise three broad classes of company. The original equipment manufacturers (OEM) such as Siemens, Intel, Motorola, Nokia, and Ericsson; the large PCB manufacturers who specialise in large- scale production of electronic components, and the local PCB assembly subcontractors. In terms of volume, these three classes form a pyramid with the OEMs at the top and the local subcontracts with their high total production volume at the bottom. As PCB manufactures continues to adopt increasingly higher levels of integration and achieving higher and higher levels of component density, tolerances on assembly become tighter and tighter. This has caused an increased need for visual inspection, especially in companies at the top and middle of the pyramid.
There are three main functions of machine vision in printed circuit board assembly:
Solder paste inspection is an in-line inspection process and, hence, systems must be capable of achieving production rates in real-time, delivering area, height, and volume of solder paste deposits following screening and prior to placement of components. Philips (NL), SVS (USA), and CyberOptics (USA) are the three market leaders in this sector. Both the Philips Triscan and the SVS 8100 systems used scanning laser-based triangulation to yield the required measurements and, common to all triangulation techniques, occlusions have to be catered for by exploiting more than one light source. Deposits of solder paste are visually-complex objects comprising collections of specular microscopic balls of solder. The resultant reflective surface causes significant difficulties for vision systems and make it difficult to achieve the high tolerances on accuracy, repeatability, and reproducability required by the electronics industry. In addition, the need to deploy active illumination techniques places a severe load on the mechanical transport systems, a load which not all vendors are capable of bearing. The CyberOptics CyberSentry system, while slower than the other two systems, uses structured light to compute the local geometry and achieves acceptable repeatability and reproducability (R&R) measures.
MV Technology Ltd. (IRL) is the world's largest supplier of component placement systems, shipping between 45 and 60 units annually. This system owes its success to its ability to achieve the strict industry accuracy and R&R requirements through the use of very high quality high-speed mechanical sub-systems and high-speed imaging. The vision technology is based on grey-scale feature-based template matching using, naturally enough, proprietary algorithms to achieve the necessary robustness. The U.S.-based Theta Group also have a presence in this market.
In the area of post-reflow measurement, where the assembled PCB is inspected after the components are electrically and mechanically bonded to the PCB following a heat- induced phase transitions of the solder from solid to liquid and back to solid, four classes of systems can be distinguished. These are
In the case of inspection of solder fillets using reflected light, systems tend to offer only binary go/no-go functionality, mainly because of the complexity of the object being inspected, the difficulty in characterising good fillets, and the need to use complex variable-illumination inspection heads. Elimination of false reject errors pose some of the main difficulties. Market leaders in this sector include Control Automation (US); Shue (D), and Grundig (D). On the other hand, Nicolet (USA) and IRT(USA) lead the field in 2-D x-ray systems. In the case of 3-D x-ray laminography, the clear market leader is the US company Four Pi (which is owned by Hewlett- Packard).
Before completing this very short sample of PCB inspection systems, it is worth mentioning a number of other applications of machine vision in this general area. These include the use of vision systems on the component placement machines themselves. Typical functions in this area include component recognition, localization (position and orientation, typically in 2-D), and PCB fiducial mark (i.e. registration mark) identification and location.
Apart from the vision technology itself, to which we will turn our attention shortly, there are a number of discernible trends in the industry which have a strong bearing of the possible success of vision companies and on the vision techniques which will need to be deployed to achieve the required functionality and to remain competitive.
These include the increasing demand for solder paste inspection to be integrated in the screening process (as a control function, rather than a QA function), the need for 3-D solder joint inspection, and the very stong demand for statistical process control functionality in all measurement systems. In addition, there is a trend toward compliance with the SECS/GEM. (Semiconductor Equipment Communications Standard Generic Equipment Model) CIM standard interfacing protocol. But, perhaps most problematic of all, there are the trends in PCB component packaging and assembly themselves. At present, fine-pitch gull-wing surface-mount devices (SMD) are prevalent. However, bump grid array, ball grid array, and flip-chip components are becoming more and more common. Inevitably, visual inspection will have to track the introduction of these new manufacturing technology with the consequent need to integrate non-visible light imaging techniques with visible-light inspection to provide a completely integrated inspection system.
In addition to the introduction of new robust vision techniques to solve emerging inspection and control problems, there is a strong trend in the industry for the deployment of vision to effect in-line process monitoring and control in the manufacture of PCBs. As a consequence, machine vision systems increasingly have to be able to achieve demonstrated accuracy, repeatability, and reproducability performances with strict industry-standard statistical process control parameters. In turn, this creates an urgent need for the adoption of acceptable benchmarking, characterization, and testing strategies for industrial machine vision.
In the previous sections, we noted that the electronics industry itself is evolving, with new PCB assembly and component packaging technologies being introduced. Consequently, it is difficult to predict what will be the required vision technologies in, say, three years time. However, one fact is clear: the industry continues to increased component densities and to increase pin-counts. Bearing in mind the trend toward in- line process control, we can make one reliable prediction: that very high-resolution image aquisition allied to very fast processing and analysis will be mandatory. Typically, both high-resolution line-scan and area-scan sensors will be needed and, increasingly, the use of colour image processing will be necessary. Furthermore, one can expect systems to exhibit a 1Gflop and 1Gbyte capability in order to deal with these speed and data requirments.
The increasing emphasis on pre- and post-reflow inspection of solder integrity will necessitate robust 3-D metrology (providing local object-centred measurements rather than viewer-centred range measurements) with a typical accuracy of 1 in 500. There will also be a requirement for reliable characterization and comparison of local shape of objects with complex geometry and reflectance properties.
One can also anticipate that the existing standard vision techniques, such as edge detection, segmentation, blob analysis, feature extraction (using, inter alia, 1-D and 2- D, grey-scale and gradient signatures), and classification, will remain key to the success of PCB applications. What will undoubtedly change, however, is that these techiques will have to exhibit demonstrated performances according to some benchmarking and characterization standard.
Moving away from the image processing and image analysis techniques to the equally important production and mechanical engineering factors, the availability and deployment of fast and accurate mechanical stages, with controllable illumination heads, is, and will continue to be, a critical aspect of successful systems.
Since this report is concerned with the competitiveness of industries which exploit vision technology as well as the scientific and engineering development of machine vision, it may be appropriate to spend a short amount of time to consider the critical success factors governing commercial and industrial success in the deployment of machine vision in the electronics industry. These factors are derived from the experience of a PCB inspection company which, over the last ten years, has successfully developed from being a university-based research group through a national R&D centre to being a highly-competitive company trading both in Europe and the USA and selling in to most of the major multinationals. Consequently, this section should be of interest both to RTD funding agencies, those responsible for technolgy transfer, and to vision companies themselves.
Unsurprisingly, there are two classes of success factors: those concerned with the underlying technology which enables machine vision to be deployed successfully to solve specific application problems and those concerned with the creation and support of a successful commercial venture. It may not gladden the hearts of the vision engineers and scientists, but the key message to emerge from their experience - and the experience of other companies in industrial machine vision - is that standard (and ageing) techniques are still the mainstay of successful systems but, perhaps more importantly and more remarkable, image acquisition, processing, and analysis itself accounts only for some 5% of a complete machine vision system; the remaining 95% derives from a mixture of mechanical and production engineering, targetted marketing, organizational and business management, and effective procurement strategies, such as outsourcing.
Let us begin with the 5%. Paradoxically, the key to success here has been the adoption of a 'low-tech' rather than a 'high-tech' approaches to solving vision problems. Significantly, this strategy is based on the crucial need for highly-robust - highly-accurate, highly-repeatable, and highly-reproducable - functionality. This results in the so-called and widely-adopted KISS approach to system development: Keep It Simple, Stupid. That said, it is of course fundamentally important to have strong expertise and deep experience of these 'low-tech' techniques. The KISS approach, while improving the likelihood of successful application of machine vision, inherently limits the scope of this application to the simpler problems. The consequence is that, if we are going to deploy more complex vision techniques to solve more difficult problems, then it is a necessary condition that they exhibit the requisite robustness. This in turn makes the adoption of a deep strategy for testing, characterization, and benchmarking of algorithms and systems an imperative (and this, of course, has been a recurring theme throughout this report).
The adoption of an adaptive approach to strategic activities whereby business opportunities are closely monitored and tactical actions are taken to seizing them. This results in a highly-focussed portfolio of expertise in specific application areas (e.g. measurement of device position or measurement of solder fillet profile).
The opportunity to outsource several distinct activities, such as the manufacture of the mechanical chassis of the inspection stations or the integration of system components such as X-Y stages and handling robots with controller hardware and software, to external market-leading companies. However, it is very important to maintain strict control of the quality of out-sourced components.
The existence of an incubation period where skill-sets and strategies are boot- strapped without having to remain in profit. In some instances, this is achieved when corporations inject capital into an embryonic division, or when university and national R&D centres are co-funded prior to full commercialization.
The availability of a small number of staff with a solid grounding in both the technology of machine vision and the discipline of entrepreneurship and business administration.
The existence on a highly multidisciplinary team with expertise in:
There is but one important conclusion which must be drawn. It is that the deployment of industrial machine vision in the electronics industry can be both successful and profitable. The long-term prognosis is excellent, given the present growth in electronic-based consumer and information technology products. However, there are significant challenges, the chief of which is the need to deploy real-time (i.e. line production rate) 3-D analysis at much improved resolutions but ensuring that these techniques exhibit the requisite (and stringent) accuracy, repeatability, and reproducability performance characteristics.
The author would like to acknowledge the very considerable contribution of Mr. Sean O'Neill, CEO, MV Technology, Dublin, Ireland, to the content of this short report.