Introducing MariSilicon X, the new OPPO image chip to improve the quality of photos and videos

Introducing MariSilicon X, the new OPPO image chip to improve the quality of photos and videos

Introducing MariSilicon X, the new OPPO image chip to improve the quality of photos and videos

News unit : Company Oppo on Tuesday, December 14, 2021 (Azar 23, 1400) on the first day of the INNO Day 2021 event apart from Air Glass smart glasses also unveiled a new achievement in the processing department of this company, namely the new MariSilicon X neural processing unit. This chip, which is made with TSMC’s 6nm technology, is expected to be used to improve the functions related to artificial intelligence, especially to improve the quality of photography and videography of advanced Oppo phones. MariSilicon X is also a leader in RAW image processing and its internal ISP supports HDR in 20-bit format. The next generation of Oppo’s Find X series will be the first phone equipped with this chip.

MariSilicon X is a complete chipset like Snapdragon 8 Gen 1 is not Google Tensor and only a neural processing unit equipped with a video signal processor (ISP) which, by being placed next to other hardware parts of the phone, removes an important part of the processing load in the field of camera and imaging from the main chipset of the device. In this chipset, a neural processing unit (NPU), an image signal processor (ISP) and a memory unit are placed together, which are used for artificial intelligence image processing.

The current video signal processor (ISP) This chipset enables 4K HDR video recording at 30 frames per second and supports 18 trillion int8 operations per second or 72 trillion int4 operations. To clarify the meaning of these numbers in the real world, Oppo used an interesting example where the noise reduction algorithm based on artificial intelligence in the phone Find X3 Pro equipped with a processor Snapdragon 888 operates at a rate of only 2 frames per second and consumes 1,693 milliwatts of power, while the same algorithm MariSilicon X operates at a rate of 40 frames per second and uses only 797 milliwatts of energy. This example shows 20 times better performance and almost half the energy consumption.

This chip features 20-bit HDR image pipelines that enable direct application of AI algorithms on RAW data. According to Oppo, this option results in an 8dB improvement in SNR (signal-to-noise ratio) compared to processing that is typically applied to the image after RGB and YuV encoding.

Another interesting point of this chipset belongs to RGBW sensors in which, this chipset provides separate processing for RGB pixels on the one hand and white pixels (W) on the other hand. This capability once again leads to an improvement of 8.6dB in the SNR index compared to the normal mode of combining pixels. Oppo calls this feature RGBW Pro, which the company says offers 1.7 times higher quality in texture processing.

In a nutshell, the ISP in this chipset Oppo video processing And the photo based on machine learning is applied directly to the RAW and raw data coming out of the sensor, as a result of which the image noise is reduced and we get a better output. A large part of this happens thanks to the MariSilicon X memory subsystem; Where the image data is not removed from the aforementioned chipset and by developing a custom memory architecture, high-speed cache memory is provided for the neural processing unit. The NPU unit is also equipped with a dedicated external DDR memory, which provides significantly higher speed and negligible latency with a bandwidth of 8.5GB/s compared to conventional data storage and access systems.

MariSilicon X as before It was mentioned that it will first appear in the high-end Find X series phones of Oppo company, which is postponed to the first quarter of 2022.

Previous IDC report on the global tablet market in the fourth quarter and the whole of 2021 - Apple is the top supplier on a quarterly and annual scale
Next Bloomberg: Huawei phones enter the market with other brands to avoid sanctions

No Comment

Leave a reply

Your email address will not be published.