- This article is filed under:
- » View All
Camera-on-Chip Winds of Change Felt in Machine Vision Market
by Winn Hardin, Contributing Editor - AIA Posted 06/26/2007
Automobiles, cell phones, and portable electronic devices are all driving the development of camera on chip (CoC) modules. These CoC modules typically have a complimentary metal oxide semiconductor (CMOS) sensor with on-chip, or hybrid image processing electronics. A low-cost lens is typically epoxied to the sensor and the module is ready to go.
While attractive in concept as the smallest, most integrated, and cost effective vision system in existence, this low cost approach cannot meet the needs of all but the most rudimentary absence/presence machine vision applications. Vision applications depend on high quality images to give software the best chance at extracting meaningful data, often with extensive help from optical and lighting design. However, while strictly defined CoC modules are not finding their way in any significant numbers to traditional machine vision applications -- as with the microprocessor and LED light, the vision industry is benefiting from levels of silicon integration pushed by the development of CoC, and the final realization of long-promised capabilities of the CMOS image sensor.
STMicroelectronic’s (Geneva, Switzerland) recently released the VS6724, a fixed-focus CoC module with 2 megapixel (Mp) sensor, on-chip JPEG compression electronics, pan and tilt mode shift electronics, flashgun support and various special effects for image display. But despite its UXGA resolution and being able to operate at 30 fps, the module is mainly targeted at cell phones and web cams.
On paper, the VS6724 would seem to be the ultimate in an integrated vision system. However, even after the low-cost, fixed focus lens dictated by the cell phone market is removed from the equation, vision systems still need more than CoC’s can offer to be successful.
CMOS imagers traditionally suffered from lower image quality than their more expensive and complex charge coupled device (CCD) architecture imagers. The quality issues were often described in terms of non-uniformity and artifacts caused by the standard semiconductor manufacturing process but, according to Cognex’s (Natick, Massachusetts) product marketing manager for factory automation, Conner Henry, ‘‘With the increase in cell phones and digital cameras, CMOS [imaging] technology is making leaps and bounds towards CCD quality.’‘
‘‘A main focus is to build cameras with not only [sensors] on board, but microprocessors for image analysis,’‘ Henry said. ‘‘As CMOS volume increases, the cost decreases. A couple of years ago you would have looked at a CMOS chip and said, ‘you can only do a few things with that’ and it was because of quality. Now CMOS offers a more cost effective solution than CCD, has a higher frame rate – which has always been CMOS’ strength – and better signal to noise ratio with clearer images and fewer artifacts. CCDs also peak out in the red, while CMOS can extend further into the infrared (IR), opening the door for some interesting applications if you understand the properties of the CMOS imagers. Some companies are using CMOS imagers as their flagship products instead of the traditional low-end products.’‘
On the imaging side, the CMOS sensor market has seen both maturation and differentiation, according to Greg Bell, vice president of sales, and Kevin Mayer, chief technology officer (CTO), at Lumenera (Ottawa, Canada).
‘‘One of the benefits of the new CMOS [imager] world is that they’re all working on ways to differentiate themselves,’‘ explains Lumenera’s Bell. ‘‘New features, capabilities, functions. One vendor might offer high speed; another offers ultra wide dynamic range, while others are experts at sub-sampling, or sub-windowing, or adding compression electronics.’‘
CMOS manufacturing processes allow for the easy integration of a memory cell, or buffer cell, directly next to or behind the light-sensitive active layer. This allows for global shuttering of the sensor array, as well as true windowing where pixels outside of the ROI are not read and do not waste clock cycles, as do most CCD windowing solutions. This is the heart of the frames per second ‘speed’ attributed to CMOS sensors.
Traditional wisdom has said that as CMOS designs and manufacturing process mature, microprocessors, controllers, digital signal processors (DSP), field programmable gate arrays (FGPA), and any number of computational units and filters would be integrated on the CMOS chip, reducing power consumption, overall BOM (bill of materials), and cost.
However, the truth is that physical parameters governing the twin worlds of CMOS electronics and imagers are quite different, and are likely to postpone the realization of advanced processing electronics on the same die as the sensor.
Fitting Electronics to Sensors
‘‘Are we trying to put processing on chip?’‘ asked to Dr. Edwin Roks, Vice President and General Manager of DALSA Professional Imaging (Eindhoven, Netherlands). ‘‘Yes, but then it’s always a question of optimizing performance and cost. Whether we go with a single chip or two chip solution [sensor, microprocessor], it depends a lot on functionality and consequently the optimal process feature size and area. If you have strict specifications on your imager, then it is often better to put processing on a separate chip. It depends on what processing you’re using.
‘‘The feature size is a key parameter to the fill factor as large as possible. But as you go to more advanced lithographies, such as standard digital electronics, the area reductions scale faster than the voltages, so you get higher electrical fields that have a detrimental effect on the image quality. So you typically use more relaxed feature sizes for lithography on sensors, which is good for both cost and leakage, but not necessarily in line with the perfect feature size for the electronics.’‘
In addition to power reductions, speed is listed as one of the biggest benefits to very large scale integration (VLSI) of sensors and processing electronics, but according to industry experts, bandwidth between multi-chip solutions is rarely a problem.
Basler Vision Technologies (Ahrensburg, Germany) recently introduced a 4 Mp CMOS array that can operate at speeds up to 96 fps, with plans to take it to 200 fps in the near future. The new Basler Sprint line scan series will offer 2k, 4k and 8k linear arrays at up to 140 kHz line rates even at 4k resolution, or approximately 4 times faster than conventional solutions.
‘‘This dual line CMOS sensor architecture of the next generation is not only extremely fast, it also provides an outstanding SNR (signal to noise ratio)’‘ explains Basler’s Lars Hansen, product manger for vision components, ‘‘we separate the active pixel surface with transfer registers behind, allowing us to have 100% fill factor and thus high quantum efficiencies of up to 60%. This makes the sensor a little thicker than a standard BGA (ball grid array). We bump bond a ceramic chip directly behind the sensor, which holds the sensor controlling electronics. So a sensor module was created which is optimized in terms of stability and heat dissipation. This in turn further reduces the heat-induced noise. The modern CMOS technology will continue to gain more and more market share for various applications where speed is not the only factor – where the image quality matters – and is a serious alternative to CCD sensors says Lars Hansen.
CoC Winds of Change
Basler is among the vision sensor suppliers offering the highest levels of light sensing and computational integration, but they’re not the only company benefiting from lessons learned from building compact camera modules. Cognex improved the processing circuitry on their two-chip CMOS imager solution, to move the DVT 515 vision sensor from only presence/absence applications to robot guidance, and high speed inspection on bottling lines.
Unfortunately, not all effects of the growing consumer imaging market will be advantageous to the vision market. The drive towards higher resolution CMOS imagers for consumer applications is also shrinking the pixel sizes, which runs against machine vision’s need for maximum dynamic range and fill factor.
‘‘Cell phones and cameras are driving CMOS sensors, but we’ve actually seen some sensors discontinued that were very good sensors for industrial markets because the volumes weren’t there,’‘ says Lumenera’s CTO, Kevin Mayer. ‘‘The industry is worried that some manufacturers may get out of the industrial market all together, but at the same time, automotive markets are driving CMOS imagers towards more industrial characteristics.’‘
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.