Monday, August 19, 2019

Sunday, August 18, 2019

Thesis on Small Size Stacked SPAD Imagers

Edinburgh University publishes a thesis "Miniature high dynamic range time-resolved CMOS SPAD image sensors" by Tarek Al Abbas that explores the small pixel pitch SPAD sensors:

"The goal of this research is to explore the hypothesis that given the state of the art CMOS nodes and fabrication technologies, it is possible to design miniature SPAD image sensors for time-resolved applications with a small pixel pitch while maintaining both sensitivity and built -in functionality.

Three key approaches are pursued to that purpose: leveraging the innate area reduction of logic gates and finer design rules of advanced CMOS nodes to balance the pixel’s fill factor and processing capability, smarter pixel designs with configurable functionality and novel system architectures that lift the processing burden off the pixel array and mediate data flow.

Two pathfinder SPAD image sensors were designed and fabricated: a 96 × 40 planar front side illuminated (FSI) sensor with 66% fill factor at 8.25μm pixel pitch in an industrialised 40nm process and a 128 × 120 3D-stacked backside illuminated (BSI) sensor with 45% fill factor at 7.83μm pixel pitch. Both designs rely on a digital, configurable, 12-bit ripple counter pixel allowing for time-gated shot noise limited photon counting. The FSI sensor was operated as a quanta image sensor (QIS) achieving an extended dynamic range in excess of 100dB, utilising triple exposure windows and in-pixel data compression which reduces data rates by a factor of 3.75×. The stacked sensor is the first demonstration of a wafer scale SPAD imaging array with a 1-to-1 hybrid bond connection.

Characterisation results of the detector and sensor performance are presented. Two other time-resolved 3D-stacked BSI SPAD image sensor architectures are proposed. The first is a fully integrated 5-wire interface system on chip (SoC), with built-in power management and off-focal plane data processing and storage for high dynamic range as well as autonomous video rate operation. Preliminary images and bring-up results of the fabricated 2mm² sensor are shown. The second is a highly configurable design capable of simultaneous multi-bit oversampled imaging and programmable region of interest (ROI) time correlated single photon counting (TCSPC) with on-chip histogram generation. The 6.48μm pitch array has been submitted for fabrication. In-depth design details of both architectures are discussed.

Thanks to RH for the link!

Saturday, August 17, 2019

Adasky Explains Why Visual World is Confusing for AI

Adasky presentation at Autosens Detroit 2019 shows that sometimes visible light images can be confusing for AI. Thermal camera is claimed to resolve these ambiguities:

Friday, August 16, 2019

LiDAR News: Velodyne, Aeye, Livox, Robosense, Hesai

BusinessWire: Velodyne introduced the Puck 32MR, a cost-effective perception solution for low speed autonomous markets including industrial vehicles, robotics, shuttles and unmanned aerial vehicles (UAVs). The Puck 32MR has a range of 120m and a 40-deg vertical FoV.

We are proud to announce the Puck 32MR as the latest addition to our broad array of lidar products,” said Anand Gopalan, CTO of Velodyne Lidar. “Velodyne continues to innovate lidar technologies that empower autonomous solutions on a global scale. This product fills a need for an affordable mid-range sensor, which our customers expressed, and we took to heart. We strive to meet our customers’ needs, and the Puck 32MR is another example of our versatility within an evolving industry.

The Puck 32MR is designed for power-efficiency to extend vehicle operating time within broad temperature and environmental ranges without the need for active cooling. The sensor uses proven 905nm Class 1 eye-safe lasers and is assembled in Velodyne’s state-of-the-art manufacturing facility. The Puck 32MR is designed for scalability and priced attractively for volume customers, although the exact price is not announced.

Aeye publishes its corporate promotional video:

Autosens publishes Aeye presentation "RGBD based deep neural networks for road obstacle detection and classification:"

Livox explains how it has achieved the low cost of its LiDARs:

"Traditionally, high-performance mechanical LiDAR products usually demand highly-skilled personnel and are therefore prohibitively expensive and in short supply. To encourage the adoption of LiDAR technology in a number of different industries ranging from 3D mapping and surveying to robotics and engineering, Livox Mid-40/Mid-100 is developed with cost-efficiency in mind while still maintaining superior performance. The mass production of Livox-40 has reached hundreds daily.

Instead of using expensive laser emitters or immature MEMS scanners, Mid-40/Mid-100 adopts lower cost semiconductor components for light generation and detection. The entire optical system, including the scanning units, uses proven and readily available optical components such as those employed in the optical lens industry. This sensor also introduces a uniquely-designed low cost signal acquisition method to achieve superior performance. All these factors contribute to an accessible price point - $599 for a single unit of Mid-40.

Livox Mid-40/Mid-100 adopts a large aperture refractive scanning method that utilizes a coaxial design. This approach uses far fewer laser detector pairs, yet maintains the high point density and detection distances. This design dramatically reduces the difficulty of optical alignment during production and enable significant production yield increase.

BusinessWire,, Quartz: Velodyne files patent infringement complaints against Robosense (AKA Suteng) and Hesai to the US District Court for the Northern District of California. Velodyne said the two Chinese companies have “threatened Velodyne and its business” by copying its LiDAR technology.

"Although styled in some instances as “solid state” devices, the Robosense Accused Products include, literally or under the doctrine of equivalents, a rotary component configured to rotate the plurality of laser emitters and the plurality of avalanche photodiode detectors at a speed of at least 200 RPM. For example, the RS-LiDAR-16 device rotates at a rate of 300 to 1200 RPM. The laser channels and APDs are attached to a rotating motor inside the LiDAR housing."

Thursday, August 15, 2019

Artilux Presents its GeSi 3D Sensing Tech Featuring 50% QE @ 1550nm

CEChina, EEWorld: After first unveiling its GeSi sensors in October 2018 and presenting them at IEDM 2018, Artilux formally announces its new Explore Series for wide spectrum 3D sensing. By operating at longer wavelength light than conventional solutions, the new Explore Series delivers exceptional accuracy, reduces the risk of eye damage and minimizes sunlight interference, enabling a consistent outdoor-indoor user experience. The breakthrough is based on a new GeSi technology platform developed by Artilux in cooperation with TSMC.

Through a series of technological breakthroughs, ranging from advanced material, sensor design to system and algorithm, Artilux says it has redefined the performance of 3D sensing. Current 3D sensing technology from other companies usually works at wavelengths less than 1µm, typically 850nm or 940nm. These wavelengths have two major drawbacks: firstly, poor outdoor performance due to interference from sunlight, and secondly a potential risk of irreparably damaging eyesight with misused or malfunctioning lasers, since the human retina can easily absorb laser energy at such wavelengths. Existing attempts to extend the spectrum to wavelengths above 1µm have suffered from poor QE, falling drastically from around 30% at 940nm to 0% for wavelength 1µm and above.

Artilux and TSMC integrate GeSi as the light absorption material with CMOS technology on a silicon wafer. It eliminates existing physics and engineering limitations by significantly increasing QE to 70% at 940nm, as well as further extending the available spectrum up to 1550nm by achieving 50% QE at this wavelength. Combined with modulation frequency at 300 MHz and above, it delivers higher accuracy and better performance in sunlight and greatly reduces the risk of eye damage – all at a competitive price. It is now ready for mass production.

Erik Chen, CEO at Artilux, says "Intelligence is shaped by connecting the dots among experiences originating from how we perceive our surroundings. At Artilux, we believe intelligence and technology evolution start with observation and connections, and by exploring and expanding the frontiers of advanced photonic technology for sensing and connectivity, we aim to be an integral part of the infrastructure for future AI and technology evolution to be built upon. The Explore Series is named after this pioneering spirit and sets out to unleash the full potential of wide spectrum 3D sensing."

The new wide spectrum 3D sensing technology will enable numerous new opportunities from short range applications such as 3D facial recognition, with sub-mm depth accuracy in both indoor and outdoor environments, to mid-long-range applications, including augmented reality, security cameras, as well as robotics and autonomous vehicles.

The first wide spectrum 3D ToF image sensors of the new Explore Series, with multiple resolutions and ecosystem partners, will be announced in Q1 2020.

Pictured below is said to be the real 3D image captured by Artilux’s sensor at 1310nm:

Corephotonics Files Another Lawsuit against Apple on 10 Patents Infringement

AppleInsider: Israeli-based Corephotonics, recently acquired by Samsung, files its 3rd lawsuit accusing Apple in infringing on its 10 patents:

9,661,233 "Dual aperture zoom digital camera"
10,230,898 "Dual aperture zoom camera with video support and switching / non-switching dynamic control"
10,288,840 "Miniature telephoto lens module and a camera utilizing such a lens module"
10,317,647 "Miniature telephoto lens assembly"
10,324,277 "Miniature telephoto lens assembly"
10,330,897 "Miniature telephoto lens assembly"
10,225,479 "Dual aperture zoom digital camera"
10,015,408 "Dual aperture zoom digital camera"
10,356,332 "Dual aperture zoom camera with video support and switching / non-switching dynamic control"
10,326,942 "Dual aperture zoom digital camera"

The full text of the lawsuit has been published on Scribd:

e2v Promotional Videos

Teledyne e2v publishes a number of promotional videos on its image sensors:

Wednesday, August 14, 2019

More Autosens Videos: Sony, Valeo, Geo, ARM

Autosens publishes more video presentations from its latest Detroit conference in May 2019.

Sony's Solution Architect Tsuyoshi Hara presents "The influence of colour filter pattern and its arrangement on resolution and colour reproducibility:"

Valeo's Patrick Denny presents "Addressing LED flicker:"

GEO Semiconductor's Director of Imaging R&D Manjunath Somayaji presents "Tuning ISP for automotive use:"

ARM Senior Manager of Image Quality Alexis Lluis Gomez presents "ISP optimization for ML/CV automotive applications:"

Tuesday, August 13, 2019

Omnivision's CTO Presentation on Automotive RGB-IR Sensors

Autosens publishes Omnivision CTO Boyd Fowler presentation "RGB-IR Sensors for In-Cabin Automotive Applications."

Another Autosens video by Robin Jenkin, Principal Image Quality Engineer, NVIDIA updates on P2020 standard progress.

Huawei Demos Fingerprint Sensor Under LCD Display

Gizmochina: Huawei presents a prototype of IR camera-based fingerprint sensor under LCD display. Most of other under-display sensors work with OLED displays. The demo posted on Twitter says:

"This is an LCD screen fingerprint technology based on the infrared camera solution, which can basically ignore the LCD panel type and unlock the speed for 300ms."

Samsung Officially Unveils 108MP Sensor for Smartphones

BusinessWire: Samsung introduces 108MP ISOCELL Bright HMX, the first mobile image sensor in the mobile CIS industry to go beyond 100MP. The new 1/1.33-inch sensor uses the same 0.8μm pixel from the recently announced 64MP. The ISOCELL Bright HMX is the result of close collaboration between Xiaomi and Samsung.

The pixel-merging Tetracell technology allows the sensor to imitate big-pixel sensors, producing brighter 27MP images. In bright environments, the Smart-ISO, a mechanism that intelligently selects the level of amplifier gains according to the illumination of the environment for optimal light-to-electric signal conversion, switches to a low ISO to improve pixel saturation and produce vivid photographs. The mechanism uses a high ISO in darker settings that helps reduce noise, resulting in clearer pictures. The HMX supports video recording without losses in field-of-view at resolutions up to 6K (6016 x 3384) 30fps.

For ISOCELL Bright HMX, Xiaomi and Samsung have worked closely together from the early conceptual stage to production that has resulted in a groundbreaking 108Mp image sensor. We are very pleased that picture resolutions previously available only in a few top-tier DSLR cameras can now be designed into smartphones,” said Lin Bin, co-founder and president of Xiaomi. “As we continue our partnership, we anticipate bringing not only new mobile camera experiences but also a platform through which our users can create unique content.

Samsung is continuously pushing for innovations in pixel and logic technologies to engineer our ISOCELL image sensors to capture the world as close to how our eyes perceive them,” said Yongin Park, EVP of sensor business at Samsung. “Through close collaboration with Xiaomi, ISOCELL Bright HMX is the first mobile image sensor to pack over 100 million pixels and delivers unparalleled color reproduction and stunning detail with advanced Tetracell and ISOCELL Plus technology.

Mass production of Samsung ISOCELL Bright HMX will begin later this month.

Friday, August 09, 2019

Himax Updates on its 3D Sensing Efforts

GlobeNewsWire: Himax Q2 2019 earnings report updates on the company's 3D sensing projects:

"Himax continues to participate in most of the smartphone OEMs’ ongoing 3D sensing projects covering structured light and time-of-flight (ToF). As the Company reported earlier, in the past, its structured light-based 3D sensing total solution targeting Android smartphone’s front-facing application was unsuccessful due to high hardware cost, long development lead time, and the lack of killer applications. Since then, the Company has adjusted its structured light-based 3D sensing technology development to focus on applications for non-smartphone segments which are typically less sensitive to cost and always require a total solution. The Company teamed up with industry-leading facial recognition algorithm and application processor partners to develop new 3D sensing applications for smart door lock and have started design-in projects with certain end customers. Separately the Company is collaborating with partners who wish to take advantage of its 3D sensing know-how to automate traditional manufacturing and thereby improve its efficiency and cost. A prototype of the cutting-edge manufacturing line is being built on Himax’s premises and it believes this project can represent a major step forward in its alternative 3D sensing applications. Himax is still in the early stage of exploring the full business potential for structured light 3D sensing technology but believe it will be applicable in a wide range of industries, particularly those demanding high level of depth accuracy.

On ToF 3D sensing, the Company has seen increasing adoption of world-facing solutions to enable advanced photography, distance/dimension measurement and 3D depth information generation for AR applications. Very recently, thanks to ToF sensor technology advancement, some OEMs are also exploring ToF 3D for front-facing facial recognition and payment certification. As a technology leader in the 3D sensing space, Himax is an active participant in smartphone OEMs’ design projects for new devices involving ToF technology by offering WLO optics and/or transmitter modules with its unique eye-safety protection design.

Thursday, August 08, 2019

Samsung 108MP Sensor to Feature in Xiaomi Smartphone

NextWeb, DPReview: At a briefing in China, Xiaomi announces it is to become the first manufacturer to use Samsung’s new 108MP ISOCELL sensor. The maximum resolution of that sensor will be 12,032 x 9,024 pixels:

Thanks to TS for the pointer!

Wednesday, August 07, 2019

IHS Markit: Hynix CIS Sales to Rise 10x in 3 Years

IFNews: IHS Markit reports that multi-camera smartphones and ToF solutions adoption benefits most image sensor manufacturers but in different degree. Hynix CIS sales are forecasted to jump 10x from 2018 to 2020, while other companies are expected to grow by 20-50%:

Under-display selfie cameras is in development by major smartphone camera makers. However, the technology needs more time to solve the issues of display transparency and uniformity in the camera area:

Meanwhile, TF Securities analyst Ming-Chi Kuo forecasts a large growth of under-display fingerprint sensors in smartphones:

Tuesday, August 06, 2019

IRay Gets Closer to Diffraction Limit with 10um Pixel Microbolometers

Yantai, China-based IRay unveils its 10um pixel microbolometers, said to be the first in China. The first product has 1280 x 1024 resolution and features:
  • Vanadium oxide pixels
  • Integrated 14b ADC
  • 30fps speed
  • Better than 40mK NETD
  • Small pixel optics and algorithms

Thanks to YN for the pointer!

Monday, August 05, 2019

LiDAR News: Quanergy, Ouster, AI Spoofing

Quanergy publishes a video explanation of its LiDAR approach:

Ouster article "The Dead Bug Problem" emphasize the importance of larger lens:

"...lidar sensors with smaller optical apertures are less resilient to obscurants. Opaque or refractive obscurants, like a raindrop, can deflect laser light, attenuate the lidar signal and thereby reduce range in that pixel. With larger beam aperture, signal strength is only partially attenuated instead of completely blocked, and the point cloud would show minimal impact."

University of Michigan and University of California, Irvine project "Adversarial Sensor Attack on LiDAR-based AV Perception" claims to find a way to fool Baidu Apollo AI-based autonomous driving platform into perceiving an obstacles in close distances to the front of a victim AV:

"In this work, we perform the first security study of LiDAR-based perception in AV systems. We consider LiDAR spoofing attacks as the threat model, and set the attack goal as spoofing front-near obstacles. We first reproduce the state-of-the-art LiDAR spoofing attack, and find that blindly applying it is insufficient to achieve the attack goal due to the machine learning-based object detection process. We thus perform analysis to fool the machine learning model by formulating the attack task as an optimization problem. We first construct the input perturbation function using local attack experiments and global spatial transformation-based modeling, and then construct the objective function by studying the post-processing process. We also identify the inherent limitations of directly using optimization-based methods and design a new algorithm that increases the attack success rates by 2.65× on average. As a case study, we further construct and evaluate two attack scenarios that may compromise AV safety and mobility. We also discuss defense directions at AV system, sensor, and machine learning model levels."

ON Semi Updates on its Imaging Business

ON Semi Q2 2019 earnings call updates on its image sensor business:

"In ADAS applications, our momentum continues to accelerate. We are seeing strong interest from customers in our broad portfolio of automotive image sensor products. Recall that we are the only provider of automotive image sensors with complete portfolio of 1 megapixel, 2 megapixel and 8 megapixel image sensors. The breadth of our portfolio has enabled us to secure many design wins from leading global OEMs and tier-1s."

Sunday, August 04, 2019

TSMC Offers 1550nm Sensing Platform

TSMC publishes its LiDAR and HDR sensing automotive portfolio mentioning 1550nm NIR+ Gen2 imaging platform:

"TSMC's pixel-level stacking technology enables higher resolution, HDR and Global shutters that are imperatives in fully autonomous vehicles. Our high QE(Quantum Efficiency) NIR technology enhances vehicle night vision. NIR+ at visible wavelength and NIR+ Gen2 at longer wavelength are being added to the portfolio, to make 940nm/1550nm Lidar sensor feasible."

Saturday, August 03, 2019

Friday, August 02, 2019

Techinsights Smartphone Imagers Review - Part 4

Techinsights publishes part 4 of the state of the art of smartphone imagers review, the omitted part from Ray Fontaine's presentation at IISW 2019 in June "Non-Bayer CFA, Phase Detection Autofocus (PDAF)."

Thursday, August 01, 2019

Omron Proximity Sensor

PRNewswire: OMRON launches the Light Convergent Reflective Sensor "B5W-LB series" for industrial applications. The sensor uses OMRON's unique lens structure combining four types of aspherical lenses for stable detection with a wide sensing range regardless of colors and materials such as shiny, black, and transparent surface:

Samsung Foundry Reports CIS Sales Growth

Samsung Q2 2019 report has an interesting part that the company's CIS foundry business is growing. Also, Samsung high resolution smartphone sensors and large pixel sensors show the growth:

SeekingAlpha earnings call transcript gives few more details:

"In the second quarter, as Chinese smartphone continues to intensify the competition for the camera spec, demand for multiple cameras and higher resolution, big-pixel, image sensors increased more, and we have achieved solid results.

Noticeably, demand for high value-added products is also expected to continue to increase because customers still want to differentiate product by adopting the products such as 64 megapixel image sensors and EUV 7-nano, APs, et cetera.

Looking ahead, we plan to expand our lineup of our 5G chipset solutions and image sensors to address demand for the high specs in the smartphone market and we will also expand our mid to long-term business scope by diversifying our product offerings through the development of the 3D and FoD sensors and automotive IoT chips. terms of revenue of our entire System LSI business, image sensors account for about one-third.