Text︱ Wang Shuyi
Picture︱ OPPO, network
At the 2021 Future Technology Conference (OPPO InnoDay 2021), OPPO grandly launched its first self-developed Chip Mariana X (MariSilicon
This chip adopts the application-specific architecture (DSA) to make the ultimate optimization of image processing. The internal integrated ISP bit width reaches 20 bits, competing products such as Qualcomm Snapdragon 1 (ISP bit width is 18 bits) , achieving the highest bit width of the current mobile phone image processing unit, and the high bit width ISP brings high dynamic range. Judging from the live demonstration of the Mariana X 4K video processing effect, compared with the current flagship model, the effect has been significantly improved.
Unlike traditional mobile phone image processing algorithms that operate in the RGB domain or YUV domain, Mariana , achieving lossless processing. According to test results provided by OPPO, compared with traditional methods, Mariana X can bring an additional 8dB signal-to-noise ratio improvement in RAW domain processing; in the image sensor processing link, Mariana X supports RGBW Pro mode. It realizes the separation processing of RGB and W channels, maximizes the characteristics of each pixel, and releases the potential of the RGBW matrix. When the size of the image sensor remains unchanged, compared with the traditional method, the signal-to-noise ratio can be improved by 8.6dB. The resolution is also improved by 1.7 times.
Specifically speaking of computing power, Mariana A lot of optimization work has been done to optimize the power consumption of this chip, making its energy efficiency ratio reach 11.6 TOPS/W, which is also an industry leader.
OPPO revealed that Mariana X uses TSMC’s 6-nanometer process and has completed mass production preparations. It will be launched in OPPO’s new generation Find
Starting from tens of billions in three years
In recent years, the dominance of mobile phone function innovation has been transferred from the main chip company to the complete machine factory. Only direct communication channels with large-scale end users , can we understand what is really needed in the market. The disadvantages of main chip companies in this regard have become more and more obvious, and the demand for OEMs to develop their own chips has become stronger. With the successful cases of Apple and Huawei, there are many domestic mobile phone manufacturers who are eager to try. .
However, as mobile phone functions become more and more complex, the development of the mobile phone main processor (SoC), which is the core component, is becoming more and more difficult. Currently, flagship mobile phone processors use the most advanced technology currently available, with more than 10 billion integrated transistors. It takes thousands of engineers to work together for more than two years to complete, and only the design files are submitted to the wafer foundry for manufacturing. The cost of tape-out is as high as tens of millions of dollars. In addition to IP purchase, tool and equipment investment, patent licensing and other expenses, from concept to formal mass production, the research and development cost of a flagship mobile phone processor chip is no less than that of building a car. It’s best not to think about investing billions of dollars.
OPPO said that the 50 billion invested in three years is not entirely spent on the chip field, but assuming that one-third is invested in self-research of chips, at least this investment level is correct. Flagship mobile phone chip research and development It's a banknote shredder. Without tens of billions of investment in three years, the products developed are basically unusable. Such investment is worse than no investment.
At this press conference, Chen Mingyong said: "Mariana MariSilicon Make self-developed chips.
”
OPPO founder and CEO Chen Mingyong
The adventure of OPPO’s self-developed chips
During the Mariana X launch event and interview Jiang Bo, senior director of OPPO chip products, was on the verge of choking and was very excited. “Our team building time was very short, and the entry point was very steep. Frankly speaking, it was very difficult, and every member of the team was under great pressure from the beginning. "," Jiang Bo said, "We recruit all experienced talents. If we make a mediocre product and practice integrating the team, it will be relatively easy." However, MariSilicon High, even today's top SOC cannot achieve it (note: refers to the ISP bit width). ”
Jiang Bo, senior director of OPPO chip products
Jiang Bo said that the OPPO chip engineering team works very hard and is under tremendous pressure every day. When the chip lights up, it is naturally very exciting. . Moreover, the team has only been established since 2019. The first one used a 6-nanometer process, but Mariana The efforts put in by the personnel and the risks taken by the decision-makers can be imagined.
Jiang Bo gave an example. The maximum speed of the MIPI (Mobile Image Transmission Interface) interface IP available on the market in 2019 can only support up to 3.5Gbps, but due to the high Mariana Gbps MIPI interface. The performance of the analog interface module requires more process adaptation, and simulation usually cannot cover all scenarios. If the 6-nanometer process has not been done before, the common practice in the industry is to first use Shuttle, that is, with multiple people Opportunity to purchase test engineering samples from the fab for tape-out). Submit the design to the fab for tape-out to test the match between the design and the process. After the test sample is brought back for testing, the design will be adjusted based on the test results to ensure that the design and process match. Later, we will officially tape-out, which is called full mask tape-out.
“Due to time pressure, we have completely canceled the iteration process, which means that IP colleagues face a very big problem. pressure. "Jiang Bo said that due to the high ability, experience and sense of responsibility of the analog design engineers, the MIPI IP integrated in Mariana capabilities.
Mariana The main processor has been simplified a lot. With HiSilicon's strength, it only started to turn around after several generations of iterations. Previously, OPPO's chips were regarded as the shortcomings of Huawei's mobile phones, with a higher starting point, higher initial investment intensity, and a longer iteration cycle. It doesn’t need to be as long as HiSilicon, but how to build a team of thousands of people into a true iron army with a unified mind and unified beliefs, and how to truly unleash the combat effectiveness of these chip people will be the test faced by OPPO executives
When dealing with the research and development of basic technologies such as chips, you must be patient and mentally prepared for the possibility that you may not be able to produce a successful product even after three or five generations of iteration. Fortunately, judging from Chen Mingyong’s words, he does. Be mentally prepared. Chen Mingyong said: "Mariana is the deepest trench in the world, which represents the difficulty of self-developed chips. No matter how many challenges lie ahead, we will persevere and never relax.
”
The dominance of mobile phone innovation belongs to the OEMs
Before Apple, Huawei, and later Tesla and OPPO, it has become a trend for OEMs to develop their own chips. In fact, complex functions The dominance of product definition has been swinging between chip companies and complete machine manufacturers.
When semiconductor technology was just industrialized, there were very few independent semiconductor companies, and most of them were system manufacturers (i.e., complete machine manufacturers) who set up semiconductors internally. Departments, typical representatives such as Motorola (later spun off from Freescale), Philips (later spun off from NXP) and Siemens (later spun off from Infineon), etc. The chips generated by the semiconductor departments of these companies are only for their own use. In theory, they can also They can provide optimized designs for their own products.
However, as the semiconductor industry gradually matures, the investment required to update and iterative technologies continues to increase, and it is difficult for a single company to maintain the corresponding investment, so the semiconductor departments of these system companies have begun to expand. To become independent and start selling its products to all potential customers on the market, it is necessary to increase the versatility of the chip, at the cost of increased costs, and it is difficult to optimize performance for a single customer application - unless customized development is done for that customer. However, these costs were acceptable at the time. Therefore, from the 1990s to the 2010s, the mainstream electronic product R&D model defined the main participation and indicators for chip companies. The complete machine manufacturers would do it based on the chip functions and performance provided by the chip companies. Development and complete machine requirements are realized through communication with the product manager of the chip company.
But with the emergence of smartphones, the situation has changed. First, the shipment volume of a single manufacturer meets the scale requirements of complex chip development. Judging from the data given by Canalys, the top five mobile phone manufacturers will each ship more than 100 million units in 2020. Only if the annual shipment volume reaches tens of millions or hundreds of millions can they meet the prerequisites for developing mobile phone processors with advanced technology—— The profits from mobile phone chip output can cover hundreds of billions of annual R&D costs.
Furthermore, modern smart devices represented by mobile phones have a broad and deep function expansion direction, and only first-line complete machines are available. Only large manufacturers can accumulate enough user feedback and technical pre-research. No matter how many system designers a chip design company has, it will not be able to establish a design feedback mechanism similar to that of the first-tier complete machine manufacturers that can achieve sufficient shipments. This is also true. This is the system advantage that Jiang Bo repeatedly emphasized in interviews with media such as TechSugar.
"OPPO is a terminal manufacturer and understands the pain points of users and the value of user scenarios. We will adjust the corresponding hardware architecture according to application scenarios to improve energy efficiency and increase effective computing power. "Jiang Bo said that it is not difficult to pile up NPU computing power on mobile phones, as long as the area is enlarged. However, to achieve the chip energy efficiency ratio to the level of Mariana Thanks to the close cooperation of colleagues.
Mariana There are many difficulties to overcome, but as long as OPPO is determined to improve its core technology, it must have the perseverance to "stick to the green hills and never relax", never forget to help, and give its own team enough time to iterate and mature, and the chip technology and terminal capabilities will be better. It will undoubtedly exert great power in future market competition.
Finally, I will end with a paragraph on my personal social media account: