Ron Mueller | CEO of Vision Markets of Associate Consultant of Smithers
Tetsuo Nomoto | Head of Sony Semiconductor Solutions Europe, Senior Vice President of Sony Europe B.V.
This talk introduces advanced sensing technologies enhancing imaging solutions. HDR imaging technology with temperature robustness improves recognition performance for automotive application. The combination of time-of-flight technology and machine learning improves the range accuracy of mobile application. In applications such as factory automation vision sensors are expected for capturing high-speed movement of target objects. Furthermore, non-Si technology that extends the detectable wavelengths to SWIR clearly captures information that has never been found by human perception, and will provides a predictive solution based on machine perception. The imaging technologies promise to accelerate the progress of sensing world by continuously improving image quality, extending detectable wavelengths, and further improving depth resolution and temporal resolution.
Benedetto Vigna | President, Analog, MEMS and Sensors Group of STMicroelectronics
With the pervasion of mobile consumer products requiring an ever increasing number of components and capabilities and advances in Artificial Intelligence, there is a growing need for image sensors with more features, higher performance and lower power consumption. Furthermore, these imagers have to be integrated along with other sensors via complex system integration.
STMicroelectronics photonics sensors are addressing these challenges with dedicated technology developments and continuous R&D activities. With a common CMOS manufacturing platform across a wide range of sensors, including MEMS-based motion and environmental sensors and ranging Time-of-Flight devices, and the addition of a strong microprocessors IP portfolio we have created a full ecosystem that can improve human-machine interaction with optimised sensor fusion.
The talk will cover current technology development and advances in the manufacturing and design of image sensors expanding to devices for a wide range of sensing applications.
Vladimir Koifman | Chief Technology Officer of Analog Value
Despite a significant investment over the last 5 years, most LiDAR companies fail to come up with competitive products. The presentation analyses difficulties in creating an automotive-grade LiDAR and typical mistakes that LiDAR companies make. The dead-ends and potentially working solutions are discussed.
Albert Theuwissen | Founder of Harvest Imaging
For several years deep trench isolation (DTI) is used as a technique to isolate pixels of CMOS imagers, mainly to decrease optical and electrical cross talk. But very recently the application and advantages of DTI are much broader than just reduction of cross talk.
By clever use of biasing and/or clocking the gates in the DTIs :
The talk will give an overview of the status of the DTI technology and the wide range of architectures in which the DTIs play a crucial role.
Yoel Yaffe | R&D Director of Samsung Semiconductor Israel R&D Center Ltd.
AI GANs may artificially generate realistic images and videos. A method to reliably identify authentic imaging content on the internet is needed. Conventional digital signatures are not effective since any change to the image invalidates them. We will present a practical yet effective solution to this problem which may be adopted globally with minor impact on the existing ecosystems and the image sensor architecture. In additional, the proposed solution would be extended to video clips.
Dr. Barry Behnken | Co-Founder & SVP of Engineering of AEye Inc.
As LiDAR systems adapt to increasingly challenging autonomous vehicle applications, both the technology itself and the metrics by which it is assessed must keep pace with new perception demands of the AV ecosystem. Within the context of “Agile LiDAR,” AEye will discuss the importance of shifting from the traditional metrics of detection range, frame rate, and resolution to the more operationally relevant metrics of classification range, object revisit rate, and instantaneous resolution.
Michael Kiehn | Director Sensor Development of Ibeo Automotive Systems GmbH
Automation of driving is one of the main topics on the research and development agenda of the automotive industry. On one side, automotive OEMS are working to achieve the step to level 3 automation of driving in general public traffic. On the other side, shuttle service providers are working on even higher levels of automation for limited use cases.
Environmental perception is a crucial part of realising automated driving. Today Radar and camera technology are established as sensing technologies for advanced driver assistance systems. Only recently, Audi expanded its sensor suite by a LiDAR sensor. Whereas LiDAR is not necessarily required for level 2 automation it is widely seen as mandatory for higher-level automation.
Today available LiDAR technology is based on mechanical scanning. This limits the robustness, durability, size and low cost potential of LiDAR sensors. Therefore, there is a demand for solid-state LiDAR technology. Many established and start-up companies came up with a broad range of solutions. Most of them are either MEMS scanning LiDARs or flash LiDARs. Both technologies have their advantages and their limitations. Ibeo’s 4D Solid State LiDAR actually combines the advantages of scanning with those of solid-state flash LiDARs.
Prof Douglas J Paul | EPSRC Established Quantum Technology Fellow of James Watt School of Engineering, University of Glasgow, U.K.
CMOS single photon avalanche detectors (SPADs) have been commercially available for a number of years and operate in a range of markets including lidar, mobile phones, autonomous vacuum cleaners and biological fluorescence imaging. The use of silicon limits the operation to below 1000 nm wavelength and the indirect bandgap also limits the detector efficiencies significantly above 900 nm. Whilst InGaAs SPADs have been available for many years operating out to 1700 nm, the high cost and low yield have limited the applications to predominantly military and scientific areas where high costs can be tolerated. Germanium is already used for a wide range of photodetectors and operates out to 1600 nm wavelength. Ge on Si SPADs produced using silicon foundry processes where the Ge absorber produces electron-hole pairs for multiplication in a silicon avalanche region will be presented with single photon detection efficiencies up to 38% at 1310 nm at 125 K operating temperatures. The efficiency and dark count rate as a function of area and operating parameters will be investigated along with demonstrations of reduced afterpulsing compared to InGaAs SPADs. The operation as a function of wavelength and temperature will be discussed along with the progress to 1550 nm operating at Peltier cooler temperatures. Examples of the SPAD use in lidar and quantum technology applications will be provided.
Raul Bravo | President, Co founder of Outsight
Hyperspectral remote sensing refers to remote spectral detection of light, reflected or scattered from a target. Each pixel of a hyperspectral imager can contain hundreds of spectral channels, as opposed to the traditional three-color RGB cameras. Hyperspectral cameras are limited in the accuracy of the spectral signal since any variation in the illumination spectrum translates into a misinterpretation of the target response. We'll introduce a new Active 3D Semantic Camera (hyperspectral).
Patrick Robert, Fellow Expert - Electronic Design, LYNRED
Session details to follow
Jörg Kunze | Team Leader R&D New Technology of Basler
New CMOS replacements for discontinued CCD sensors often differ in pixel size and thereby cause integration problems in existing applications. Adapting the pixel grid usually requires interpolation. Common interpolation methods create images with inhomogeneous pixel size and gain. This creates implausible EMVA 1288 results and may lead to visible artefacts. We present a novel image interpolation that fixes these problems, thereby performing fractional binning in the digital domain.
Ljubomir Jovanov | Researcher of Ghent Uni/imec
Due to the rapid development of CMOS sensor technology, the resolution of imaging sensors is constantly increasing. While this trend is favourable for many applications, such as surveillance, industrial inspection, medical imaging and security, it also poses numerous challenges for camera hardware and optics. In order to stay on pair with the increase of the sensor resolution, lenses have to be built with a higher precision, which also requires increase in the physical size and the price of the lens. The use of larger lenses is not always physically possible, especially in portable camera systems which leads to inevitable chromatic aberration artefacts. Moreover, due to the reduced size of pixel, the amount of light it receives is significantly smaller, which increases the level of noise dramatically. Finally, a vast majority of the cameras today are using one imaging sensor with spatial colour multiplexing arrays. This reduces the complexity and the price, compared to the three sensor camera, but necessitates the use of more complex pixel processing algorithms.
We propose a complete video processing chain for correcting lens aberration artefacts, noise reduction and de-mosaicking. The proposed solution is implemented on a GPU and is capable of processing Ultra HD video streams at 30 fps, while offering full flexibility of a software implementation.
David Schie | CEO of AIStorm
For the first time CIS structures can be used directly or in hybrid configurations as the building blocks of neural networks. AI-in-Imager charge based neural networks open new possibilities for image enhancement and high speed processing. In this talk we will discuss the technology behind these networks, as well as potential features enabled by this technology. We will discuss the limitations, especially controller, memory and mobile AI models and examine results from silicon.
Sung-Yun Park | Assistant Professor and Assistant Research Scientist of Pusan National University & University of Michigan
We present a prototype CMOS active pixel that is capable of simultaneous imaging and energy harvesting without introducing additional in-plane p-n junctions. The prototype pixel uses a vertical p+/nwell/psub junction that is available in standard CMOS processes. Unlike the conventional CMOS electron-based imaging pixels, where the nwell region is used as a sensing node for image capture, we adopted a hole-based imaging technique, while exploiting the nwell region for energy harvesting at a high fill-factor of >94%. To verify the feasibility, CMOS image sensors are fabricated and characterized. We successfully demonstrated that the energy harvesting can be achieved with a power density of 998 pW/klux/mm2, while capturing images at 74.67 pJ/pixel. The fabricated prototype device has achieved the highest power density among the recent state-of-the-art works and can self-sustain its image capturing operation at 15 fps without external power sources above ~60 klux of illumination.
Ramses Valvekens | Managing Director of easics
How to develop small, low-power and affordable AI engines that run close to your sensors?
Designers search for embedded AI solutions that integrate tightly with the sensors such as image sensors, LiDAR, Time-of-Flight, ... A flexible framework is used to automatically generate hardware implementations of the deep neural networks. This scalable AI engine for FPGA and ASIC is ready for the future. A talk for image sensor manufacturers that add AI in their products.