At June's Automate trade show in Chicago, the industrial automation event drawing big players from Nvidia to Boston Dynamics, one thing became particularly clear: AI has an enormous footprint in vision systems right now. And, the hype is arguably at an all-time high.
Yet, standing at Chroma's booth as a component provider working across UV to shortwave infrared (SWIR) wavelengths, I found myself repeatedly explaining the basics of what an optical filter actually does. That experience crystallised something important about where our industry stands today.
The hidden cost of software-first thinking
What many system designers are starting to discover is that computing power remains a finite resource, regardless of your AI system's sophistication. The practical utility that gets lost in the rush to embrace software solutions is the pre-processing benefit that optical filters provide. Rather than relying solely on sensors and software to separate discrete light and sift through an overwhelming volume of information, a well-chosen filter converts that swimming pool of data into a tablespoon: something far easier for a computer to process, while freeing up processing power for other critical tasks.
This isn't a new problem. But, AI has definitely amplified it. System designers are typically mechanical engineering-based professionals who optimise products from a mechanical vantage point, without always having the foreknowledge needed to balance optical path integration requirements. The result is often a beautifully sleek enclosure paired with optics needing near-impossible specs and far more expensive than they needed to be – a situation that could have been avoided with earlier consultation.
When filters get considered too late
In my experience, optical filters simply aren't considered early enough in the design process. Part of this stems from the hype around AI's potential as a "do-everything" solution, which can shortchange the technology by wasting processing energy on tasks that simpler components handle far more efficiently.
The economics tell their own story. High-precision optics typically represent a fraction of a system's total build cost, often less than 1%, while delivering substantial gains in efficiency. Crucially, they achieve this without the prohibitive replacement costs of a damaged sensor.
Working in tandem with customers whose teams are primarily mechanically focused, our engineering and design teams handle machining and process engineering in-house to bring dedicated mechanical support alongside optical expertise. The real satisfaction comes from being involved at the prototype phase, right from the initial breadboard concept through the full production cycle, where we can anticipate potential issues and tailor genuinely economical solutions rather than over-specified ones.
The angle problem nobody accounts for
One of the biggest challenges in machine vision, particularly with wide fields of view required for real-world 3D perception, is the angle of incidence. As light moves off-axis, the spectral response of an interference film stack shifts toward the blue. Each edge of the band lags at a different rate, creating a non-uniform band collapse past a certain angle. This is a bedrock physics issue that cannot be engineered away, though collimating incoming light as much as possible before it reaches the filter yields the most accurate and reliable spectral response. In situations where it’s not possible to move the filter, there are filter design solutions that accommodate large cone angles depending on the requirements.
I recently worked with a customer who required a 30-degree half-cone angle on a 20nm bandpass filter in the visible spectrum, positioned at the front aperture of their instrument. At that cone angle, the band of a standard filter design collapsed to an unacceptable degree. If their system design had already been locked down, we would have been forced to use a more complicated, and expensive, coating design approach to mitigate the collapse. However, because their design was still open, we moved the filter deeper into the system where the light could be collimated first. This solved two problems simultaneously: spectral performance became near-perfect, and because the collimated beam only needed to illuminate a much smaller filter, we produced roughly ten times more parts per coating run, dramatically improving cost efficiency. That is the true value of involving optical expertise before system architecture solidifies.
SWIR applications and the drone effect
The proliferation of drone technology, alongside developments in autonomous vehicle vision, has pushed shortwave infrared applications well beyond their traditional home in aerospace and defence into agriculture, 3D mapping, and crop health monitoring via hyper- and multi-spectral imaging.
Working across this waveband, from 900 to 2,500nm, brings unique physical considerations. Achieving the same bandwidth at longer wavelengths generally requires thicker coating depositions. This increases both unit cost and physical stress on the substrate, creating genuine tension against the industry's simultaneous drive toward miniaturisation. And yet, sputtered SWIR coatings meet these challenges and can deliver on performance often only associated with VIS-NIR coatings.
The message for integrators
If there is one principle I wish every vision systems integrator understood before starting a project, it is this: whatever you can do to restrict incoming (photonics based) information to only what is strictly necessary is the single best thing you can do for your system.
Other advice includes:
- Prioritise optics early: Treat optical components as primary design considerations, not afterthoughts.
- Account for cone angles: Design optical paths that manage light geometry before it hits the filter (more on this in a future article from Chroma Technologies).
- Allow physical space: Leave sufficient room in your form factor for optics that require physical thickness to perform.
- Optimise the workload: Do not leave basic data filtering entirely to the sensor and software.
Optical components are often the unsung heroes working behind the scenes of systems people admire for entirely different reasons. Getting that message out and getting involved early enough to make a genuine difference, is an ongoing mission.
