Stephen Mounsey looks at some of the smallest cameras in the industry, discovering the limits of what's on offer at each size
Miniaturisation is ubiquitous in electronics, with a constant drive in the industry towards ever-smaller components and devices. The imaging and machine vision market is no exception. Just how small a camera can be produced depends in part on its functionality; a VGA resolution camera for simple, rolling-shutter video streaming can be made a lot smaller than a high-resolution, global-shutter GigE Vision camera, for example. The miniaturisation of cameras for imaging and machine vision is often aided by techniques and technology developed for mass-market applications, such as mobile phone cameras.
German manufacturer IDS took this approach to producing its XS model miniature camera for imaging applications. Steve Hearn, head of UK sales at IDS’ distribution partner Stemmer Imaging, explains that using chips from the consumer or IT markets allows the manufacturers to take advantage of huge production volumes with associated economies of scale, making the price per component very appealing. The XS camera is a 26 x 23 x 24mm device, featuring an 8 Megapixel sensor, autofocus (with an integrated, motorised lens), steady-shot, and face-finding, all of which is built into the single camera chip supplied by Sony. ‘It’s essentially technology that’s been taken out of a mobile phone,’ says Hearn. ‘This is a high-end camera with a lot of intelligent features at a very low cost. The key to IDS’ project was to get something that was consumer-driven, turn it into an industrial product in a magnesium industrial housing, and then provide the correct USB interface with software that would enable a system integrator or an OEM to turn the device into something usable.’
Being a consumer product in an industrial housing, the device’s functionality is not specialised for machine vision applications; it has a rolling shutter, rather than the global shutter usually found in machine vision, and it is not triggerable. Applications for the camera, says Hearn, include producing simple, high-resolution photographs for ID cards. ‘The face-recognition functionality can extract a face, or provide coordinates for it in the image,’ he says. In order to deal with more specialised applications, however, developers must move away from consumer technology.
Portuguese firm Awaiba develops and produces some of the most miniature camera modules currently available, with the smallest measuring just 0.7 x 0.7 x 1mm. Martin Wäny, chief executive of the company, explains that the driving market for cameras of this size is endoscopy. Consumer camera components are not available at this size, and so the company developed its chips, but consumer-driven innovations were also used in the design: ‘We used the latest generation of semiconductor packaging technologies, which were developed for mobile applications – silicon via packages, and wafer-level lenses for example,’ explains Wäny. ‘These permit us to bring the whole packaging assembly down to the micrometre scale. Obviously we also use quite aggressive CMOS-process technology in order to put the full system on a chip with a footprint of less than half a square millimetre.’
Awaiba already has a standardised camera of the 0.7mm size on the market, offering 250 x 250 pixels, and it will soon release a 300,000 pixel VGA resolution device in a 2 x 2mm package. The tiny cameras offer frame rates of up to 50 frames per second. ‘The main driver for these developments is the endoscopic market, but it also has a trickle-down effect for machine vision sensors,’ he says. ‘Inside packages and optics that we made mainly for the endoscopic markets, we are now beginning to put sensors with slightly lower resolution but with global shutter functionality. An endoscope has a rolling-shutter sensor, as it only requires a lot of pixels in a small area, but for machine vision, a global shutter with a small sensor can be very attractive.’
One of the challenges of miniaturisation, according to Wäny, is that of achieving connectivity to such a small device. ‘Once the module is so small, it is difficult to make electronic contacts with it. We only use four wires, but connecting four wires to an 0.7 x 0.7mm device is not easy,’ he says, noting that the company has put considerable development effort into finding a reliable and cost-effective manufacturing process. ‘We can’t tell you too much about our process, but we basically had to develop our own micro-soldering techniques to attach very fine wires to a very fine package,’ he says.
With these innovative manufacturing techniques in hand, Awaiba would have little trouble making its cameras even smaller still, but current development efforts focus instead on improving the resolution of the devices and on introducing global shutter functionality for machine vision applications. ‘We do not believe that there is a drive to make these modules any smaller. Already the size is such that if you drop the camera on the floor you have a very hard time finding it, even without a carpet! Even for medical applications, [0.7 x 0.7mm] is about as small as anybody asks for. Eventually, as science fiction turns into reality, when we have micro-robots floating through our bodies doing this and that, there may be a requirement for much smaller cameras, but current prototypes in micro-robotics are at the millimetre scale, despite being called nanotechnology.’
The Awaiba camera uses a data transfer protocol developed by the company, which is designed to have a very low complexity on the chip, while being able to cope with cable lengths of up to 2.5m. While 2.5m cable lengths might limit the usefulness of the camera in some machine vision applications, Wäny does not see this as a problem: ‘Obviously the medical market is driving development; machine vision is too small a market to drive anything if we’re honest. Machine vision is a market that has to piggyback on other technology drivers,’ he says. Even so, Awaiba intends to use the technology developed for medical markets in some machine vision applications: ‘We have some interesting applications for small machine vision sensors in pharmaceutical manufacturing. There are many processes involving automatic dispensers, working to a formula such as “20 drops of this” and “500 drops of that” and so on. Normally, the dispensers give one drop at a time, but sometimes they do not, and one should know whether a drop has come out or not,’ says Wäny. It would, he says, be impractical to put a large camera with a full C-mount lens next to each of these small dispenser nozzles, and so a miniaturised solution could be useful.
True machine vision on a tiny scale
For many industrial applications, only a true machine vision system will suffice; a suitable interface, standard lens mounts, high resolution sensors, and a triggerable global shutter are all desirable. These systems have also been successfully miniaturised, and one such example already on the market is the Ace GigE Vision camera from Basler. Henning Tiarks, team leader of product management within the components division of Basler, describes the Ace: ‘The outer dimensions are 29 x 29mm; all the analogue cameras that have been on the market for many years have had this footprint, but the Ace is a digital camera. All of the electronics needed for a digital camera have been designed to fit in this very small housing.’
A machine vision camera, and particularly a machine vision camera with a high-speed interface such as GigE, requires many specialised components, explains Tiarks. ‘We needed to find ways to fit all of these electronic parts into the camera; in miniaturised cameras, we have to be particularly careful with heat, and so we design for low power consumption. The other thing to consider is interference between circuits inside the camera,’ he says, adding that this is an especially important task so as to get acceptable sensor performance from small cameras. ‘In most cases you will find some misbehaviour of the camera in the sensor itself, meaning that it influences the image quality. We are very careful about where we put each part, and about how to handle it, how to absorb certain voltages, and so on.’ Avoiding interference requires careful design, alongside an iterative approach to improving that design. ‘There were some optimisation loops done in development, working out how the camera behaves with various internal layouts. We can always see interference in terms of the image quality; we measure image quality by way of the EMVA 1288 standard, and then we look at the next step in optimising the design,’ says Tiarks. Optimisation, he says, is done by replacing components and redesigning the layout of tracks on the PCB. It’s an iterative but guided process, always working towards a goal of meeting the EMVA 1288 standard for image quality. ‘In most cases the performance of the miniature camera is comparable or better to generation one or generation two GigE cameras.’
Minimising the effects of heat in a small camera requires a low-power design. Basler has co-developed a low-voltage Power over Ethernet (PoE) transformer alongside a manufacturer of power supplies. ‘This device is specially designed for the low power consumption of the camera, which is around two watts. Other devices that use a PoE transformer, such as a network switch for example, will typically run as high as 48W.’ Low power contributes to thermal management in the camera, but excess heat must still be removed efficiently: ‘The whole camera has a thermal contact, and a special heat dissipation concept. This allows direct heat dissipation to the housing, especially to the mountings,’ says Tiarks, explaining that in order to manage temperature inside the camera, the thermal pathway is planned right down to the mounting screws used to fix the camera to a surface.
Basler’s Ace is a fully-featured machine vision camera measuring 29 x 29mm, and its design required careful management of heat and interference. Image courtesy of Basler.
Although the camera itself is small, it still makes use of standard machine vision lenses, something which Basler may change in the near future. ‘People have begun to wonder whether it is possible to miniaturise the whole system, meaning the camera, the lens, and the cables. Right now we have a small camera, but we still use a lens that is bigger than it,’ says Tiarks. The next generation of cameras may still offer a C-mount for standard lenses, but will also allow shorter, more compact lenses to be attached, using the M12 format for example. ‘To give a feel for this, a typical C-mount extends 5cm from the camera, whereas an M12 extends approximately 1cm,’ he adds.
Knowing the industry
Before the introduction of digital machine vision cameras, the standard applications in factory automation were done with analogue cameras, mainly because they were cheap, small, and easy to integrate, and many applications are still using these analogue cameras for the same reasons. ‘We now go after this analogue market,’ Tiarks says: ‘Many automated industrial processes require a camera to be mounted on the working head of machine – on pick-and-place machines for example – and in these applications the size and weight of the camera is of greater importance than in static applications.’ As such, he says, small GigE cameras are a particularly attractive alternative to analogue cameras in such applications. Another advantage of the miniature GigE Vision camera is its long cable length (up to 100m), giving the interface an advantage over FireWire cameras which, although available in small sizes, are limited to 4.5m maximum cable lengths. The analogue systems Tiarks hopes to replace use cables of 10-15m.
Tiarks states that the Ace camera is also cheap enough to compete with comparable analogue cameras, as well as USB interfaced cameras. ‘USB is very popular, particularly in medical and biomedical applications,’ he says. USB cameras can be made smaller than GigE cameras as they make use of highly integrated but non-specialised standard processing chips. The GigE Vision camera also has a physical layer chip that coordinates the interface, and the PoE electronics. ‘If a USB chip is running faster than the computer’s bus is able to support, then you will lose image data. This is the reason that USB was never really successful in factory automation processes,’ he says, adding that this is not a problem in applications such as digitising a microscope output. Microscopy is a market Basler hopes to get into. Tiarks believes that the camera’s performance advantage, coupled with the extended cable length of GigE Vision and the fact that almost all computers have an Ethernet port, will make the Ace an attractive offering for such applications. This is just one more example of a machine vision developer pioneering a new solution to a commercial need, and we can expect to see many more such developments come through miniaturisation in the future.