Picture Perfect - Optimising Imaging Performance Parameters for Specific Application Criteria

Author : Mike Green - Editor, EPDT

02 January 2024

Figure 1: Samsung’s ISOCELL HP2 imager for high-end smartphones
Figure 1: Samsung’s ISOCELL HP2 imager for high-end smartphones

The ability to capture sufficiently accurate imaging data, even under difficult operating conditions and in application environments that are far from conducive, is of inestimable value to modern day life. The following article looks at the companies at the forefront of this segment, and the pioneering technologies they are responsible for developing.

There is undoubtedly a growing prevalence of image sensors, with them being employed across a wide array of different sectors. These go all the way from surveillance equipment, quality control systems in industrial automation installations and the machine vision hardware incorporated into automated guided vehicles (AGVs), through to the external cameras needed to support automotive driver assistance functions, smart building systems, medical instrumentation, plus consumer electronics products. 

It's probably no surprise to anybody that the global business for these devices is currently experiencing considerable enlargement. According to a recent report by Verified Market Research, a compound annual growth rate (CAGR) of 9.8% is projected between now and 2030 - resulting in a monstrous annual revenue of $41.4 billion by the end of that period. 

An abridged history
When I started out in the electronics industry, back in the late 1990s, there were basically 2 types of image sensor that engineers could procure - namely charge coupled devices (CCDs) and CMOS-based units - with each respectively having a sizeable share of the overall market. Not as mature at that stage, CMOS imagers were considered inferior and only really employed in the most cost-sensitive of applications, where performance wasn’t such a big concern. Conversely, CCDs (which had originally been conceived at Bell Labs decades earlier) offered clear performance benefits in terms of higher quantum efficiency and lower dark current - leading to less noisy renderings of acquired imaging data. In the years that followed CMOS devices gained greater dominance, with improved performance being witnessed with each new generational iteration, while still also keeping a significant price point advantage. Today CCDs have been marginalised to niche use cases (such as astronomical imaging and high energy physics research).   

CMOS supremacy and the breadth of opportunities addressed 
The volumes of CMOS image sensors now being produced is staggering, with IC Insights’ figures on this putting shipments at close to 10 billion units annually. With a plethora of applications being served by them, the key parameters that dictate whether they are specified/sourced will vary dramatically. What is required for taking eye-catching holiday snaps from a smartphone will be very different from the safety-critical demands of object identification to prevent potential vehicle collisions. 

Figure 2: Sony’s IMX735 for automotive deployment
Figure 2: Sony’s IMX735 for automotive deployment

Under some circumstances it will be resolution that is top of the list. While in others, it will be attaining a high dynamic range, minimal power consumption, keeping within budget constraints or ease of integration that proves most important. This text will give a brief overview of what the particular priorities are, along with examples of products that have matching attributes. It can only really scratch the surface though - and there will be alternative products with similar capabilities that it will be worthwhile investigating too.  

The consumer domain
There is no question that the uptake of imaging related activities by the general public, in combination with the countless outlets for sharing pictures that social media now provides, have been a major driver of the unprecedented CMOS image sensor sales seen in the last decade. Through the imaging technology they select for their product designs, consumer brands have a very effective way of gaining differentiation over their rivals.   
Aimed at incorporation into top level smartphone handset designs, the ISOCELL HP2 from Samsung (which was introduced at the start of 2023) boasts a 200Mpixel resolution. It is comprised of 0.6µm x 0.6µm pixel structures. The sophisticated Tetra2pixel pixel-binning technology employed means that the output from adjoining pixels can be aggregated together under low light conditions (either in pairs, or sets of 4). This means that resolution can be sacrificed in order to boost output. In addition, the dual vertical transfer gate (D-VTG) arrangement means that a photodiode has been embedded into each pixel. Then, via the voltage transfer gate, electrons are passed from the pixels to the logic layer. The upshot is the full-well capacity for each pixel is augmented significantly (by about 33%). 

Vehicle-based imaging 
We are already reaching a point where even mid-range car models can have a multitude of external and in-cabin cameras fitted. These include front view units for collision prevention, parking sensors, lane trackers, occupant monitoring devices, etc.

Setting a new benchmark with respect to automotive image sensor resolution, while also adhering to ISO 26262 functional safety requirements, the Sony IMX735 was released in mid-September. Comprising more than 17.4Mpixels, this AEC-Q100 Grade 2 compliant 13.7mm diagonal format device enables object recognition to be conducted over a longer range than is possible with competing devices - so there is more time in which situations can be reacted to. Another key facet is the 130dB dynamic range, which means that high definition images can be derived even under poor lighting conditions. By having a horizontal row-by-row pixel signal output (rather than a vertical line-by-line arrangement) it is more straightforward to synchronise with the vehicle's LiDAR scanning equipment. 

Figure 3: The 8Mpixel OX08D10 image sensor from Omnivision
Figure 3: The 8Mpixel OX08D10 image sensor from Omnivision

Omnivision's 8Mpixel OX08D10 image sensor, which was released just a few days earlier, is likewise highly deserving of note. It is based on the company's proprietary 2.1µm x 2.1µm pixel TheiaCel technology, which features a capacious light capturing reservoir along with a dual conversion gain methodology (where captured photon signals are measured twice at 2 different gain values). Consequently, the device exhibits industry-leading low light capabilities. Lateral overflow integration capacitors also mitigate the LED flicker issues associated with traffic lights, roadside digital signage and other illumination sources. 

Smart homes/buildings 
As we start to see heightened degrees of automation within domestic settings, image sensors are expected to play a significant role here too. onsemi’s Hyperlux LP devices (which debuted in the autumn) are intended for security cameras, people counting functions, smart doorbells and biometric-based access systems, as well as having potential in an entertainment context too (such as extended reality headsets). The stacked architecture means that high resolutions can be supported (up to 20Mpixels) while taking up very little board real estate. The wake-on-motion function allows power budget to be conserved until full operation is called for. Another useful feature is its Smart ROI capability. Here, just the relevant sections of the scene may be captured. Other sections can be ignored, so as to save bandwidth and minimise the current draw involved.  

Factory automation          
Industrial usage calls for image sensing technology capable of rapid responsiveness (before situations arise that could result in production batches having to be abandoned or the prospect of staff being endangered). Support for high rates of data throughput and low latency operation are thus mandated.  

The 12MPixel 1.1” optical format GMAX3412 imaging devices from Gpixel are aimed at machine vision applications. They make use of global shuttering (so as to be able to readout captured images in one go and avoid distortion issues). In 10-bit mode, with all 16x LVDS channel pairs being read, 128fps frame rates can be supported. Even when in 12-bit mode, 60fps is attainable. 30fps can be realised in 12-bit mode via the 4x MIPI D-PHY channels.

Also using a global shutter arrangement, Sony’s IMX900 is a 3.2Mpixel stacked CMOS imager is meant for use in robotic implementations (component picking-and-placing, production line tending, etc.). Compactness also endears it to AGVs and drones too. It has a unique structure that allows increased aperture size (by placing the memory elements away from the photosensitive pixels and embedding them in the signal processing). The wider and thicker pixels that this presents means more incident light can be captured and near infrared (NIR) performance is also improved.

Contact Details and Archive...

Print this page | E-mail this page