Interpretation of ON Semiconductor’s image sensor strategy—long history, comprehensive innovation

As the main enabler of intelligent sensing technology, what is the layout of onsemi in image sensors? Recently, at the Munich Shanghai Electronics Show, the author had the honor to interview Tao Zhi, Greater China Marketing Manager of Industrial and Consumer Applications of ON Semiconductor's Intelligent Sensing Department, and took stock of some of ON Semiconductor's products and layout in the field of intelligent sensing.

Long history and accumulation

Counting from the acquisition, ON Semiconductor is the oldest CMOS image sensor company in the industry, with more than 45 years of history.

There are obviously several main lines here, including Cypress, Truesense, Aptina, Sensl and their corresponding predecessors, all of which have had great success in the sensor industry. For example, Cypress's customized CMOS sensors are used in ARRI professional digital cameras. Movies shot with ARRI cameras, lenses and lighting systems have repeatedly won Oscars for Best Picture, Best Cinematography and other awards.

Not to mention Kodak, which has the most patents and inventions in the image sensor industry. For example, one of the foundations of image sensor technology, the Bayer array, was invented by Kodak scientist Bryce Bayer.

The same is true for Aptina. Eric Fossum, the inventor of the CMOS image sensor, founded Photobit, the first CMOS commercialization company, and was acquired by Micron, which later became Aptina.

Through successive acquisitions, ON Semiconductor has the largest patent portfolio in the industry, with more than 2,300 items, and also has the most comprehensive product portfolio in the industry. Currently, in addition to the mobile phone market, ON Semiconductor's products cover industrial, automotive, consumer, IoT and other fields. It is the number one CMOS sensor supplier in the global industrial and automotive markets.

Always at the forefront of the times

"What ON Semiconductor is best at is innovative work." Tao Zhi emphasized.

An example of ON Semiconductor’s most recent innovation is the AR8022 image sensor. The device's embedded high dynamic range (eHDR) capability and optimized near-infrared (NIR) response are critical for applications with harsh lighting conditions, such as security surveillance, body-worn cameras, doorbell cameras, and robotics. The sensor's low-power architecture and motion wake-up capabilities are designed to significantly reduce system power consumption.

The main role of HDR technology is to improve image quality. By using HDR technology, the color contrast and brightness in the image can be enhanced, making the image more vivid, vivid and realistic.

There are currently three main HDR technologies, including multiple exposures, large and small pixels, and super exposure.

Firstly, for multiple exposure technology, long exposure is used to capture details at low brightness, and medium and short exposures are used to capture images at medium and high brightness. In the case of conventional pixel design, a dynamic range of 120dB (3 exposures) or even 140dB (4 exposures) can be achieved through multiple exposure technology. However, the general multiple exposure technology has shortcomings: there will be motion artifacts for moving objects, especially for LED light sources. Sometimes it can be captured by a short exposure time, and sometimes it cannot, which results in unclear LED flickering. The problem. Typical pulse light sources are LED traffic lights, traffic signs and car tail lights.

The second is large and small pixel technology. There are both large pixels and small pixels on the chip of this type of design. Large pixels capture low-light information, and small pixels capture high-light information. Since large and small pixels are exposed simultaneously, this design avoids the problem of motion artifacts. However, this technology suffers from large pixel QE loss, as well as color problems caused by different QEs of large and small pixels, crosstalk, etc., requiring users to spend a lot of resources and costs for correction, which brings great challenges to users. Large and small pixel technology uses the ratio of large pixels and small pixels to sense light to achieve HDR. Current semiconductor manufacturing technology capabilities do not allow arbitrary expansion or reduction of the physical size of pixels, so the dynamic range that large and small pixel technology can achieve is also limited.

Under the current pressure of cost and miniaturization, to make a chip with the same resolution smaller, you need to use smaller pixels. The large and small pixel solution is more difficult to reduce the pixel size than the single pixel solution. Note: The large and small pixel technology is a patent of ON Semiconductor and was obtained by Micron in 2009.

The third is super exposure, which ON Semiconductor uses in the Hyperlux series. By adding additional circuits, when the pixel capacitance is saturated, the overflow charge continues to be transferred to the external capacitor storage, thereby achieving an ultra-large full well charge, and the dynamic range can reach 110dB for a single exposure, and 150dB for two exposures. This approach has no LED flicker issues and motion artifacts, while also providing very good low-light performance.

The eHDR technology is called embedded High Dynamic Range, which can be understood as HDR with increased processing capabilities. Usually a multiple-exposure sensor will send the data of each exposure to the back-end processor for synthesis, and 3 exposures will need to send 3 frames of data. According to calculations, if traditional HDR data transmission methods are used, 4K 3 exposures require more than 10Gbps, which poses a huge challenge to SERDES and cables.

The multiple exposure synthesis of eHDR is completed inside the sensor. Through the internal intelligent fitting and compression function, the 20-bit data of 3 exposures can be compressed into 12-bit data, and then transmitted to the back-end platform, which greatly reduces the system cost. of data bandwidth. If the traditional HDR data transmission method is used, 4K 3 exposures require more than 10Gbps, which poses a huge challenge to SERDES and cables.

AR0822 uses eHDR (embedded High Dynamic Range). Multiple exposure synthesis is completed inside the sensor. Through the internal intelligent fitting and compression functions, the 20-bit data of 3 exposures can be compressed into 12-bit data, and then transmitted to The back-end platform greatly reduces the data bandwidth of the system.

AR0822 supports a variety of multiple exposure synthesis linearization fitting functions, DLO (Digital Lateral Overflow) and SCMAX (Smooth Combination Max), which are called intelligent fitting. This mode reduces the brightness critical area during multiple exposure synthesis. noise.

In addition, multiple exposures also have problems with motion artifacts and LED flickering. The AR0822’s eHDR also adds a motion compensation function to alleviate these problems.

What’s more worth mentioning is that in addition to eHDR ^TM , AR0822 also has a WOM (Wake on Motion) function: AR0822 can run at an extremely low frame rate. When motion is detected in the ROI area of ​​the screen, it will wake up the SOC. Restores sensor operation to normal mode.

Broad platform-based product portfolio

Tao Zhi said that CMOS sensors are used in different fields and have completely different requirements for various characteristics. According to the broad categories, it is first divided into human eye recognition and machine vision. The products corresponding to the two applications are completely different. For machine vision, what is needed is objective, accurate, non-deformable, and good motion recognition imaging that serves the algorithm. For human eye recognition, you need to consider color, image quality, etc., which is more subjective.

However, Tao Zhi also said that there is no complete boundary between these needs. For example, global shutter technology is mostly used in industrial applications, but this technology may also be needed in the motion capture process of human eye recognition.

For industrial applications, the requirements for sensors are also different. Even for the same application, the different requirements of the external environment are also very different. “The CMOS market is very complex and needs to consider various factors including pixels, low-light effects, external ambient light intensity, noise control, cost, power consumption, temperature, vehicle regulations, anti-intrusion detection, etc. This requires a thorough understanding of the product and market. With a deep understanding, we can define products with dual advantages in architecture and cost,” Tao Zhi emphasized.

To this end, ON Semiconductor's solution is to launch a series of products based on a certain technical architecture through a platform approach, and customers can make choices based on actual requirements.

Strong ecosystem

The complexity and division of labor of the optical system require the construction of an ecosystem and close cooperation with partners for debugging. At the Mucha exhibition, ON Semiconductor demonstrated a variety of solutions jointly developed with partners.

In addition, in order to lower the development threshold, ON Semiconductor and third parties have developed reference designs or modular products, combined with ON Semiconductor's high-efficiency SDK, to accelerate customer innovation in the field of intelligent sensing.

Fab Right Strategy

In the field of image sensors, ON Semiconductor has always adopted the foundry method. Now, with the company's new Fab Right strategy, a Fab for the production of image sensors will be established.

Tao Zhi said that the shortage of semiconductors in the past few years has also affected CMOS image sensors. Now several friends are actively expanding production capacity to cope with future demand growth. As ON Semiconductor, we will adopt a combination of OEM + self-owned factories and walk on two legs to better balance the return on capital investment, risk level, supply cycle, cost and other aspects. In addition, after being out of stock a few years ago, the supply situation has become an important consideration for customers when selecting models.