Shortwave infrared (SWIR) imaging has existed for decades in applications ranging from military and defense to satellite imaging, anti-counterfeiting, and cultural heritage analysis. However, up until recently, SWIR cameras have not seen widespread adoption in the industrial space due to the costs and complexity associated with manufacturing indium gallium arsenide (InGaAs) image sensors. But now image sensor breakthroughs have led to lower-cost, compact SWIR cameras with extended spectrum capabilities, and these cameras are proving exceptionally effective in machine vision applications ranging from food and beverage to waste recycling.
SWIR Sensors Light the Path Forward
In the past, creating SWIR image sensors with smaller pixels than those in today’s industrial CMOS sensors has proven challenging with conventional pixel-level bump bonding. In this technique, a certain bump pitch is needed to bond the InGaAs photodiode layer to the silicon readout circuit layer. Sony’s new SenSWIR technology uses a Cu-Cu (copper-copper) direct bond interconnect that directly connects pixel chips and logic circuit chips of stacked CMOS sensors using copper terminals, which eliminates the need for specialized connection areas, so image sensors can be made with finer pixel pitches and smaller pixels.
Sony’s hybrid SWIR technology was delayed in 2021, during the COVID-19 pandemic, but it’s now hitting the market in mass. Many machine vision camera companies now offer smaller, high-resolution, lower-cost SWIR cameras, which offer sensitivity in the 400 to 1700 nm range.
Elsewhere in the SWIR image sensor space, onsemi recently purchased SWIR Vision Systems, a company that develops SWIR cameras based on colloidal quantum dot (CQD) thin film photodiodes fabricated on silicon readout wafers. These nanoscale semiconducting materials allow CQD-based cameras to image in the 400 to 1700 nm range while also offering extended range up to 2100 nm. With the acquisition, onsemi plans to combine its silicon-based CMOS sensors and manufacturing expertise with CQD technology to offer “highly integrated SWIR sensors at lower cost and higher volume.”
SWIR camera prices have been dropping dramatically, but with this deal, costs will likely drop even further, giving machine vision consumers a wide range of SWIR camera choices at an affordable price point. In fact, these advancements and lower entry costs into SWIR imaging have led to a tremendous response from the machine vision community. SWIR imaging is now among the fastest-growing segments in the machine vision market. Market research company Yole Group forecasts that SWIR imaging will a have 28% compound annual growth rate, increasing from $89 million in 2022 to $395 million by 2028 as a result of advancing technologies, steady demand, and shrinking costs.
Breaking Down Imaging Barriers
Another novel feature of the new Sony and onsemi SWIR sensor technology is that both sensors now cross the 1 µm/1000 nm imaging barrier, offering typical usable response from 400 to 1900 nm. Previously, traditional InGaAs SWIR sensors were not responsive in the visible range, offering a usable range of around 1 µm/1000 nm. Similarly, silicon-based CMOS sensors lost response by 1 µm, which led to the term “1 µm barrier” within the machine vision and imaging space. The ability of these new sensors to deliver continuous response across the visible and SWIR spectrums has led to the expansion of multispectral imaging applications in the machine vision market.
Multispectral imaging is driving several new and emerging imaging and machine vision applications, including agricultural applications. For example, the cameras can be used to define similarly colored green foliage to separate weeds from the target crop, to identify fruit within foliage, to pick rocks or other objects from soil, or to judge moisture content. Elsewhere, multispectral imaging applications are gaining steam in waste recycling, where the cameras can pick out specific types of material from the debris stream, such as high-density polyethylene (HDPE), where hyperspectral imaging options may be too costly and data intensive.
LED Manufacturers Respond
Even prior to the latest sensor developments, SWIR systems have proved quite useful in the machine vision space for applications where imaging beyond the visible wavelength is required. In food inspection, for example, SWIR cameras can identify damaged or bruised fruit beneath the surface or inspect food products through plastic packaging. Elsewhere, SWIR cameras are useful in applications such as inspecting fill levels in nontransparent containers, identifying moisture in packaging, gauging water content in plants, anti-counterfeiting, wafer and solar cell production, and much more.
For systems integrators experienced in machine vision technology, SWIR is essentially no different than visible imaging systems in terms of deployment. Today, with cost barriers falling, the demand for and deployment of SWIR cameras is increasing rapidly as the technology offers a useful, novel tool for the machine vision toolbox. While the proliferation of SWIR technology has been led by image sensor developments, lighting companies must also follow suit.
While companies have been quicker to market with lenses suitable for SWIR imaging systems, lighting manufacturers have been slower, primarily due to limits in SWIR LED technology and the corresponding high costs. Today, however, as technology progresses and demands increase, LED manufacturers are adapting to the shift in the market. In addition, newer SWIR LEDs produce more light output and less heat for every watt of power put in while also increasing the absolute intensity of their SWIR light output.
While SWIR LED lighting has been a custom or semi-custom option for most lighting manufacturers, several companies in the market offer standard products in SWIR wavelengths, and the number of options has increased. SWIR lighting options on the market today range from 1050 to 1550 nm, along with multi-zone lights that combine different wavelengths in a single housing.
Multi-Light Controllers Drive Multispectral Forward
Multi-light controllers are also necessary to build the next wave of multispectral machine vision systems. Multispectral imaging involves multiple images taken at unique wavelengths to show the properties of a given object in that narrow band response. Inspection applications may be as simple as software analyzing individual images, treating them as monochrome inspections. More commonly, however, the sequence of images is fused through a weighted filter into a single multispectral image, or images are added, subtracted, or differentiated against a reference image to produce desired results.
To streamline the process of programming and acquiring these image sequences, a newly available lighting controller offers programmable sequencing and a four-channel constant current LED driver in a single package. The controller can pair with matching multichannel lighting available in linear, ring, or dome light formats, each of which offers three of four channels. With a TCP/IP interface, the controller can make programming any type of imaging sequence intuitive for any level of user.
With these new technologies, users can select wavelengths from 420 nm to 1650 nm to individually populate each available channel as required for a given application. These developments, along with the latest SWIR cameras, mean that machine vision users today have more multispectral imaging technology at a lower entry cost point than ever before. The future of multispectral imaging is indeed bright, at many more wavelengths.
