After two days at Image Sensors Europe 2026, one message stood out: the race for better pixels is giving way to a new competition, building smarter, more integrated vision systems. From AI-driven image pipelines to the growing demands of automotive and industrial applications, the industry is entering a phase where performance is defined not just by the sensor, but by the intelligence around it.
There was no single “headline breakthrough” moment. Instead, what emerged was a convergence of ideas across mobile, automotive, industrial, and scientific imaging. Taken together, they point to a shift that feels both gradual and profound: imaging is moving from a component-level discipline to a system-level one. As highlighted by Marie-Charlotte Leclerc of ST MicroElectronics,
“The next big step is not just better sensors—it’s better systems.”
That sentiment echoed across multiple talks. For years, progress has been measured in quantum efficiency, pixel size, and dynamic range. Those metrics still matter but, in truth, they are becoming hygiene factors rather than differentiators. Increasingly, performance is defined by how the sensor interacts with the rest of the pipeline, ISP, SoC, and now AI models.
This was especially evident in discussions around hybrid shutter architectures, in-sensor HDR, and sensor–AI co-design. The traditional separation between capture and processing is dissolving. Neural ISP approaches, for example, are now capable of replacing entire blocks of the imaging pipeline, learning how to reconstruct images directly from raw sensor data, as discussed with Tom Bishop from GlassAI. This is where the industry may be underestimating the scale of change. If ISP functionality becomes software-defined, the centre of gravity shifts from semiconductor differentiation to algorithmic differentiation. That has implications not just for engineering, but for business models.
At the same time, the relentless push for smaller pixels driven largely by the smartphone market appears to be reaching physical limits. Over the past decade, pixel pitch has shrunk from around 2 microns to as little as 0.5 microns. But several speakers questioned whether that trajectory still delivers meaningful benefits. One particularly striking observation from Joren Hoet, CEO of eyeo, highlighted the mismatch between optics and scaling:
“We’re trying to fit a one-micron photon into a half-micron pixel—it’s not a winning game.”
The uncomfortable reality is that the industry has known this for some time. What’s changing now is that it’s being said more openly. Pixel scaling is no longer a straightforward roadmap, it’s a compromise between physics, optics, and marketing.
As a result, innovation is shifting elsewhere. Pixel binning is now standard practice. Larger apertures and folded optics are compensating for light loss. And perhaps most significantly, software is stepping in where physics pushes back. Using AI to reconstruct detail that the sensor alone cannot capture. If mobile imaging is where scaling pressure is most visible, automotive and industrial vision are where the strongest growth narratives are emerging. Across multiple presentations, automotive was repeatedly identified as a key driver, not just in volume, but in complexity.
Modern vehicles are rapidly increasing their camera count, moving from a handful of sensors to potentially a dozen or more, covering everything from parking assistance to in-cabin monitoring. As automation levels rise, so too does the demand for higher resolution and reliability. As noted in Alastair Attard’s talk from UTAC Group,
“As automation increases, camera counts go up dramatically.”
What’s particularly interesting here is that automotive is forcing the industry to mature. Unlike smartphones, where failure is inconvenient, in automotive failure is unacceptable. That changes everything from design margins to qualification processes, and it’s quietly reshaping how sensors are developed.
Industrial vision tells a similar story, albeit less visibly. From food sorting and logistics to battery inspection and robotics, machine vision is already embedded in countless processes. Yet deploying these systems remains slow and resource-intensive, often taking months or even years. That challenge is now being addressed through simulation and digital twins. Tools based on platforms like NVIDIA Omniverse are enabling engineers to design and validate vision systems virtually before deploying them in the real world, as demonstrated by Jorg Kunze of Basler AG. If this approach delivers at scale and that remains a big “if” it could be one of the most commercially impactful developments discussed at the conference. Reducing deployment time is arguably more valuable to many end users than incremental sensor performance gains.
Alongside these application-driven trends, there is also a clear broadening of what “vision” means. RGB imaging is no longer enough for many use cases. Multispectral sensing, depth imaging, and event-based vision are all gaining traction, particularly in robotics, automotive, and security. The underlying driver is AI. Vision is increasingly the primary data source feeding intelligent systems, from autonomous vehicles to industrial robots. As discussed with Harish Venkataraman from META,
“Vision is becoming the eyes of the system.”
And yet, there is a subtle but important shift here. It is not just that vision feeds AI, it is that vision systems themselves are being redesigned for AI consumption, not human viewing. That distinction is easy to overlook, but it has profound implications for sensor architecture and optimisation.
Beyond the sensor itself, several talks highlighted the importance of packaging, manufacturing, and supply chain considerations, areas that often receive less attention but are increasingly critical.
Packaging, in particular, is evolving rapidly. The transition from ceramic to laminate and moulded packages is enabling smaller, cheaper, and more scalable designs. But this comes with new challenges around reliability, especially in automotive environments where failure is not an option. At the same time, the global supply chain remains a point of concern. Despite improvements since the pandemic, dependencies on specific regions, particularly in East Asia persist. And as stated during the panel discussion on day 1,
“You cannot build an image sensor if one component is missing.”
This is not a new insight but it is one that is now being taken more seriously. The shift toward regionalisation and dual sourcing is less about efficiency and more about resilience. And that shift is likely to stay.
At the other end of the spectrum, scientific imaging continues to push the boundaries of what sensors can achieve. Astronomy applications, for example, demand ultra-low noise, long exposure stability, and extremely large detector formats, requirements that far exceed those of consumer devices. While these applications may represent a small portion of the market, they often serve as a proving ground for new technologies that later find their way into mainstream products. It is a useful reminder that innovation does not always start where the volume is.
Stepping back, the most important takeaway from the conference is not about any single technology, but about direction. The industry is moving toward a model where sensors are no longer isolated components, but integral parts of intelligent systems. As reinforced again by Harish Venkataraman,
“It’s not about if sensors become more powerful—it’s about how they work with AI to understand what they see.”
For those working in machine vision, that shift has significant implications. It changes how systems are designed, how performance is measured, and ultimately, how value is created.
Better pixels still matter. But smarter systems, and the intelligence around them, are what will define the next phase of imaging.
Editor’s Takeaways
- The real battleground has shifted – Sensor performance alone is no longer enough; differentiation is moving rapidly toward system integration and software.
- AI is not an add-on, it’s redefining imaging – From neural ISP to edge processing, the pipeline is being rebuilt around machine consumption, not human viewing.
- Pixel scaling is slowing into a physics wall – The industry knows it, and is now openly pivoting toward computational solutions and optical innovation.
- Automotive is setting the new standard – Reliability, redundancy, and system-level validation in automotive are quietly reshaping expectations across all markets.
- Deployment, not just performance, is the bottleneck – Simulation and digital twins could be as disruptive as any sensor breakthrough if they reduce time-to-market.
- Supply chain resilience is now strategic – Not just a post-COVID concern, but a long-term design and sourcing consideration for the entire ecosystem.
