Last Updated on December 15, 2021 by Anu Joy
Oppo unveiled the MariSilicon X, a neural processing unit (NPU), at its Inno Day event on Tuesday. The Chinese smartphone company claims that the chip will greatly enhance the quality of photos and videos.
It seems that the company is trying to inch its way into the semiconductor space with this new chip. It will be fabricated using TSMC’s 6nm process. The next-gen Find X series, to be announced in Q1 2022, will be the first phones to use this chip.
MariSilicon X: Separating ISP and NPU from the Main Chip
The MariSilicon X combines the tasks of an NPU (neural processing unit) and an ISP (image signal processor). The main task of an NPU is to implement all the necessary control and arithmetic logic to execute machine learning algorithms. They greatly increase the efficiency of artificial intelligence (AI) applications. ISP, on the other hand, is a specialized digital signal processor (DSP) used for image processing in digital cameras and other devices.
What sets it apart from the SoC on your regular smartphone, is that the NPU and ISP are a part of a separate ASIC. This means that it has a dedicated memory, which makes image processing faster, theoretically. Oppo claims that it can perform 18 trillion operations per second (TOPS) with high efficiency owing to the NPU’s dedicated memory. The chip is supposedly capable of 8.5 GB/s data transfers.
MariSilicon X Image Processing
The chip is said to support 20-bit, 120db dynamic range in images. It also boasts of a multi-tier memory architecture which makes real-time RAW processing possible. This, in turn, is purported to capture “stunning 4K AI Night Video with live preview”. Jiang Bo, OPPO’s Senior Director, claimed that its “cutting-edge imaging NPU” promises to make mobile imaging systems more powerful and create an extraordinary experience for users.
AI can use the NPU to implement algorithms for noise reduction, better color reproduction, HDR, improved detail, and dynamic range. The MariSilicon X will also be able to utilise the full potential of Oppo’s RGBW sensor since it supports the separation of the white channel from the RGB.
Discussion about this post