The competition between smartphones has turned into a competition for shooting capabilities.
It is mainly reflected in two aspects. First, the number of cameras is superimposed. Mobile phone manufacturers develop self-developed image processing chips, and even cooperate with traditional lens manufacturers. The second is the market growth of CMOS image sensors (CIS). The proportion of mobile phones in the field is as high as 68%.
The CMOS image sensor is equivalent to the eyes of a machine. It converts optical signals into digital signals for the machine to “watch”. It is a key component to realize the camera function of mobile phones. Its performance directly affects the quality of photography, and its cost occupies 52% of the entire camera module. %, and is also widely used in security, industry, automobile and other fields.
Because of this, in the semiconductor field, CIS is regarded as a big cake in the segmented field. According to the forecast of the Yole Développement report, by 2026, the CIS market size will reach 31.5 billion US dollars (about 204.022 billion yuan).
Although the market is very large, the traditional sensor market has been basically eaten by Samsung, Sony and OmniVision for a long time, and it is difficult for start-ups to break through the shackles and enter the market. However, hard opportunities do not mean no opportunities. Vision chip technology is still evolving, and machine vision still has unsolved problems.
Recently, the domestic cutting-edge enterprise Ruisizhixin released a bionic vision sensor chip – ALPIX-Pilatus chip (ALPIX-P for short), this chip not only stands on the same starting line with Samsung and Sony, but also relies on its core technology advantages. , in the future, it is very likely to break the iron walls of major manufacturers such as Samsung and Sony in the field of visual sensing, bringing more choices to the industry.
ALPIX-Pilatus Fusion Bionic Vision Sensor Chip
After CIS, giants enter event cameras
A month ago, Sony Semiconductor released two new vision sensor chips. Different from traditional CIS, these two sensor chips are called stacked event monitoring vision sensors (EVS, Event-based Vision Sensor). The latter is modeled on the working principle of the human retina, each pixel is equivalent to a retinal cell, works independently, and is only sensitive to changes, also known as an event camera.
Not only Sony, but Samsung has also developed a series of chips to enter the event camera industry.
Looking back at the development of image sensors, CCD devices with high pixels and high manufacturing difficulties were the mainstream at first, and Japanese companies carried the banner. Until 1995, several Chinese international students founded OmniVision in Silicon Valley, using traditional CMOS technology for image processing. In the sensor, replace the CCD, thus breaking the monopoly of Japanese companies.
There are many reasons for CMOS to eventually replace CCD. On the one hand, although CMOS has poorer image quality than CCD, its power consumption and cost are reduced, which is suitable for the consumer electronics market.
According to the report of Strategy Analytics, a market research organization, OmniVision, Sony and Samsung have more than 80% market share in visual sensors. While CMOS technology maintains the main theme of the sensor market, the scale is also growing.
Since CMOS is so popular so far, why are the visual sensor giants cutting into event cameras?
There are two reasons. First, the integration of AI and CIS is the future trend. Second, traditional CMOS is only suitable for shooting static pictures with the human eye. In terms of machine vision, there are still many pain points.
Deng Jian, founder and CEO of Ruisizhixin, said that traditional machine vision faces four major pain points: poor motion capture, large data redundancy, high power consumption, and poor environmental adaptability.
This is mainly due to how the CIS itself works. CIS mainly integrates the photocurrent over time to realize exposure, and it takes a long time from exposure to digital-to-analog conversion to data transmission. The longer the time, the higher the energy consumption, and the lower the frame rate of the image. The long interval between frames results in the lack of information, which increases the difficulty of the machine’s judgment on the image.
At the same time, CIS outputs 8-12 bit information, which has a large amount of data transmission, high energy consumption, slow reading, large storage space requirements and high energy consumption for processing images.
“CIS also tries to solve these pain points, but based on its working principle of controlling all pixels to perform integral exposure at the same time through the shutter, the three parameters of frame rate, data volume and dynamic range are difficult to balance.”Deng Jian explained. For example, in high dynamic range shooting, multiple exposure algorithms are often used, that is, a high dynamic range image is formed by combining multiple frame images with different exposure times, which will inevitably lead to a decrease in frame rate, more data operations and greater system delay. , it is not applicable in scenarios that require short response times and capture rapid changes.
Comparison of bionic vision sensor and traditional CIS sensor parameters
Based on different working principles, event cameras can break the limitations of CIS in machine vision.Deng Jian said that the event camera imitates the working principle of the human eye. Each pixel works independently like a nerve cell, independently and continuously perceives the change of light, and converts the change of light into a digital pulse signal. If the light intensity does not change, it will not. output.
“To sum up, the event camera has the characteristics of high speed, small amount of data, low power consumption and large dynamic range, which correspond to the current pain points of machine vision one by one.” Deng Jian said.
The only independent event camera company in China, with core technology ahead of Sony and Samsung
It is precisely because the event camera can solve the problems that traditional CIS cannot solve, Samsung, Sony and other visual sensor manufacturers have entered the machine vision sensor that mimics the retina of the human eye.
According to Deng Jian, since the bionic vision technology originated in Switzerland, the main technical teams are mainly concentrated abroad, including Prophesee in France, insight and inivation in Switzerland. The two Swiss companies have been acquired by Sony and Samsung respectively. In addition to Ruisi Zhixin, Shanghai Chip Lun Optoelectronics is also involved in bionic vision technology, but at present, Chip Lun Optoelectronics has been acquired by Weil Shares.
“In the field of event cameras, Ruisizhixin is currently the only independent start-up company in China.” Deng Jian said.
Although it was founded in China, similar to several other event camera companies, the core team of Ruisi Zhixin is from Switzerland, and the core members have been engaged in the research and development of various high-end sensors, such as high-end image sensors, ToF, X-ray, neuron signal detection chips, etc. It has been less than 3 years since the establishment of Ruisizhixin, but its core team has entered the field of bionic vision as early as 2014 when it participated in the Swiss National Foundation project to design a retina-like chip implanted in the eye of the blind.
In terms of financing, at the beginning of this year, Ruisi Zhixin completed a Per-A round of financing of nearly 100 million yuan, which was jointly led by Hikvision and Yaotu Capital. Venture Capital and Tongchuang Weiye joined the investment, and the old shareholders Lenovo Venture Capital and Zhongke Chuangxing continued to raise their bets.
It is worth mentioning that Ruisizhixin has unique core technology, which has formed a differentiated pattern from Sony and Samsung from the very beginning.
“Why is the event camera, which has been developed for more than 6 years, still active in the field of industrial vibration measurement, but not in the field of consumer electronics?” Deng Jian said: “The existing technical concepts and methods of event cameras on the market are still based on twenty The technology from years ago has resulted in limited functions of the final designed and manufactured chip, and the scope of application is not wide.”
In fact, in a real shooting scene, machine vision requires not only fast prediction of event signals, but also fine judgment or reconstruction of image signals. Therefore, both event cameras are required to capture dynamic information, and traditional CIS is required to present detailed static information. information.
“The existing event cameras, due to their underlying design still using a relatively traditional method, have some problems, which are difficult to be compatible with the output image signal, and the pixel size is large and the noise is large.” Deng Jian said.
Because of this, RISIC integrates the bionic event camera function and the CIS function at the pixel level, replacing two cameras with one camera, and placing two sets of circuits in the same pixel, one for generating event signals and one for outputting. image signal. This technology was named Hybrid Vision Fusion Bionic Vision Technology by Ruisi Zhixin.
Hybrid Vision Fusion Bionic Vision Technology Principle
However, because the number of circuits in a pixel is ten times that of a traditional CIS, the design difficulty is also greatly increased, and at the same time, the designed chip pixel area may be large.
In order to solve this problem, Ruixizhixin adopts an advanced 3D stacking process, placing the pixel photosensitive part and the image signal output circuit on the upper wafer, and the event signal output circuit on the lower wafer, and then the upper and lower wafers are connected by copper interconnect technology. The circular circuits are correspondingly connected on each pixel, which can not only greatly reduce the pixel size, but also isolate the signal crosstalk between different circuits and reduce noise.
Entering consumer electronics, it is expected to ship in the middle of next year
Based on the unique core technology, Ruisizhixin has designed the ALPIX series chip products, and ALPIX-P is now officially released.
According to Ruisi, the newly announced ALPIX-P can quickly switch between the traditional image sensor mode and the bionic event camera mode. In image mode, ALPIX-P is a global exposure sensor with a maximum frame rate of 120 frames, which is completely consistent with existing traditional image sensors and fully compatible with existing mature vision systems. In the bionic event camera mode, ALPIX-P has the characteristics of ultra-high frame rate, high effective information ratio, and large dynamic range. It is expected to land in the VIoT field by next year.
However, the low-power IoT field is not the entire market target of Ruisi Zhixin. The ALPIX-Eiger, which is planned to be released at the end of this year, is suitable for small camera devices with ultra-high image quality, such as mobile phones and security cameras. Ruisi Zhixin said that the chip will further reduce the pixel size and increase the resolution.
“In fact, at this stage, we are not thinking about how to subvert the visual sensor industry, but just try to meet the gap in market demand as much as possible. As a start-up company, we will be down-to-earth and focus on the VIoT field in the early stage, and will expand to the automotive and consumer electronics fields in the future. “Deng Jian said.
The Links: 3HAC029105-001 4N5028
