NIO Technology Day: Self-developed chips, self-developed full-stack OS, next-generation smartphones, and more

The second NIO IN event of NIO concluded this afternoon. At this NIO IN event, NIO continued the main content of the previous NIO IN, introducing aspects such as intelligent driving chips, smart cockpit, intelligent chassis, and the second generation NIO Phone. What kind of growth has the seed planted at last year's NIO IN event seen over the past year? Let's take a look together. Shenji NX9031 In the opening speech, Li Bin raised the Shenji NX9031 and excitedly announced the successful mass production. Li Bin also demonstrated the capabilities of the NX9031 chip's ISP processor on-site, showing that under the same 800W camera, the results of NX9031's ISP processing are more refined. NIO claims that Shenji NX9031 is the world's first 5nm intelligent driving chip. Shenji NX9031 has over 500 billion transistors, a 32-core CPU architecture, 615 K DMIPS, self-developed high-performance ISP (6.5G Pixel/s, 26-bit width)... Li Bin proudly stated: "NX9031 is not just leading! I didn't say those four words hahaha." When the audience below shouted out "far ahead," Li Bin responded with a "yes!" SkyOS Tianhu Wang Qiyen, Vice President of NIO Digital Systems, stated that although cars have entered the AI era, in the history of automobiles, "there has never been a complete and comprehensive vehicle-wide operating system." So, what does a "complete vehicle-wide operating system" need? NIO's answer includes 7 aspects: High-bandwidth data throughput, ultra-low latency communication, support for large computing power and heterogeneous hardware, cross-domain integration, flexible evolution throughout the lifecycle, high reliability assurance, and systematic information and privacy protection. SkyOS builds a technical cluster of 1+4+N, covering multiple areas such as vehicle control, intelligent driving, cabin, and mobile interconnection. In this technical cluster, the virtualization system SkyOS-H, which is closest to the hardware, can manage programs, support cabin, and autonomous driving domain multi-operating system requirements, and provide a secure base. The four kernels (SkyOS-L, SkyOS-M, SkyOS-R, SkyOS-C), when combined, can meet the needs of different types of applications for the entire vehicle, and the middleware is the key to tightly integrating them. SkyOS-H SkyOS-H is a virtualization system oriented towards complex scenarios, high performance, and high reliability, used to manage and schedule the heterogeneous large computing resources of the entire vehicle, providing a secure isolated operating environment for various businesses. If the entire vehicle computing power is imagined as a large house, then SkyOS-H virtualization technology can divide the house into different functional rooms to meet various purposes and needs. This requires efforts in performance improvement. In terms of latency optimization in SkyOS, NIO has reduced the end-to-end execution time of real-time tasks by 50%, increased the throughput of multi-threading by over 40%, and increased the throughput of virtual disks by 15%. This can more efficiently utilize and manage the resources of the entire vehicle, providing customers with a better driving experience, making OTA updates faster, and improving the connectivity between the vehicle and the driver. ![](https://mmbiz-qpic.wallstcn.com/sz_mmbiz_jpg/UnA9a9gzqTBub9XZdRt19JAjcGyaKVrT0CzeMyNTLtUyqY7Lj6xz0QAibq1vDbPuKzo1J5IxIEakKHxiagCnOxdA/640? On four different kernels, NIO has made a lot of efforts. SkyOS-L Sky·OS-L is a lightweight OS for MCU low-power chips. Wang Qiyan stated that it supports all mainstream car MCU chip architectures and is the most widely deployed SkyOS kernel in vehicles. Reliability is one of the most important indicators of SkyOS-L, which needs to ensure that real-time messages of different priorities are delivered on time. In the on-site demonstration, compared to the commonly used car OS in the industry, SkyOS shows more consistent distances between bars representing messages of different priorities, demonstrating better system stability. SkyOS-M The microkernel architecture SkyOS-M is used in the central brain to mainly run vehicle control-related functions, including body, chassis, suspension, etc. One of the core advantages of this architecture is safety isolation. NIO adopts a modular approach to build the entire system, achieving two types of CPU isolation, two levels of memory isolation, five types of IO interface isolation, and 25 core service isolations, resulting in a stronger system robustness. At the same time, this architecture also has fast recovery capabilities. Through a 4-layer monitoring, 3-layer recovery security mechanism, and full-link up-down backup, it can achieve fast and reliable full-system self-recovery SkyOS-R SkyOS-R is mainly used in the field of intelligent driving, with strong end-to-end latency requirements, able to provide high performance, real-time optimization for a variety of rich application scenarios. NIO's goal is to establish an OS that can support various rich application scenarios and provide high real-time guarantee. Wang Qiyuan gave an example of optimizing terminal processing calls to illustrate NIO's efforts to improve real-time guarantee. The scheduler determines which external events can be processed first. By optimizing a large number of call returns, the size of the scheduler's preemptive window has been increased, allowing tasks with higher priority and implementation requirements to be completed more quickly. SkyOS has a substantial performance advantage over Linux, 113 times higher under light loads and 20 times higher under heavy loads, providing more confidence for intelligent driving construction. SkyOS-C Wang Qiyuan stated that SkyOS-C is one of the most important kernels and the OS closest to the user. This kernel carries most of the functions of intelligent cockpits and serves as the engine for on-board AI applications and intelligent experiences. NIO has built native intelligent operations in SkyOS-C to efficiently support applications based on large models, including AI intelligent experiences such as NOMI SkyOS Middleware Middleware is the key to achieving layered decoupling, cross-domain integration, and is the foundation for supporting full lifecycle upgrades and information security. NIO has built more than 40 middleware, abstracting over 1600 atomic capabilities, provided to various domain functions in the form of services and APIs for invocation. In other words, developers can find the tools they need in a toolbox filled with tools. With such complexity, NIO needs a framework to support the management and invocation of vehicle atomic capabilities, and that is the SOA architecture. Self-developed SOA Framework The SOA framework is like a kitchen, flexibly selecting and matching ingredients to make dishes according to the requirements of different customers. Wang Qiyan stated that this SOA framework can provide reliable remote calls, high-bandwidth communication, precise access control, and flexible deployment. Banyan 3.0.0 AI Intelligent Chassis Since the release of the AI intelligent chassis six months ago, over 150 million road bump events have been identified, covering over 94% of counties and cities nationwide. The chassis of Banyan 3.0.0 will officially upgrade from 4D Comfort Chassis to 4D Global Comfort Navigation, and can be activated without turning on navigation. Cabin entertainment and audio content Banyan 3.0.0, in collaboration with QQ Music, introduces premium AI sound quality, unlocking a 7.1.4 panoramic sound experience, currently supporting the playback of millions of mainstream songs. In terms of tuning, NIO has introduced its self-developed expert equalizer Pro EQ, supporting fine adjustment of frequency response curves from 20 Hz to 20,000 Hz This year, NIO will use NOMI GPT for the first time to achieve AI-generated EQ. NOMI can analyze the style of a song and comprehensive information about the singer to generate a set of balanced curves in real-time. It can also express personalized adjustments to favorite vocal styles. On the K song 2.0 platform, NIO introduces the first AI intelligent accompaniment experience based on a native application. As soon as the user starts singing, the AI can immediately eliminate the original vocals and switch to accompaniment mode. In addition, NIO is equipped with AI voice filters to help repair breath, lip, and tooth sounds. Flux Light Application Platform Li Tianshu stated that in the current lack of native in-car applications, the Flux Light Application Platform can directly generate flash applications based on the web, providing high-quality content and services with an experience close to native applications. Taking Apple Music as an example, users' accounts, playlists, music libraries, and mobile apps can achieve complete synchronization and can be controlled through NOMI for playback. 4. NOMI Evolution In NIO's plan, they will evolve NOMI in three directions: more human-like interaction, becoming a capable assistant on its own, and becoming a warm emotional partner. To achieve this, NIO will: Further iterate the human-vehicle dialogue paradigm, achieve a one-second response time through self-developed end-side wake-up engine, increase interaction efficiency by 100%, and be usable even in offline mode. In the face of multi-person communication, NOMI can simultaneously position, separate, recognize, understand, and respond to dialogues of 4 users, and make smart summaries and replies based on the order of arrival and execution. In terms of multiple instructions, based on large-scale model training for sentence segmentation and understanding, NOMI can support continuous delivery of dozens of instructions without pause in a single sentence, and can execute them. Furthermore, NOMI can not only recommend detailed parking information and real-time security to users, but also have the ability to act as an "encyclopedia," automatically generate and edit travel videos, and more. NIO has also introduced the NOMI Emotional Engine. Supported by the NOMI GPT cognitive central model and multi-agent architecture, this engine has three cornerstone abilities: situational intelligence, exclusive intelligence, and emotional expression. In simple terms, NOMI can adjust cabin environment, NOMI voice text, etc., through thousands of atomic information and calculations related to people, vehicles, environment, and remember users' exclusive names, preferences, habits, and more. ![](https://mmbiz-qpic.wallstcn.com/sz_mmbiz_jpg/UnA9a9gzqTBub9XZdRt19JAjcGyaKVrT28gjrTkEGu1O6p4BoRgYxrAFcoOqCp4ibiasFOTgWHdmvT4fEgJbXpMQ/640? NOMI itself can also have personality, with intonations and natural fluency in voice output, providing emotional value. More importantly, NOMI has the ability to continuously learn and evolve. Intelligent Driving Since September 2023, NIO's total mileage for full-domain navigation assistance verification has exceeded 3.898 million kilometers. Based on these available roads, users have driven over 1.112 billion kilometers using navigation assistance. Ren Shaoqing stated, "Our NIO users are a group of users who love self-driving and driving very much." The actual computational power of NIO's NAD group intelligent system is 287,100 POPS, with actual road verification mileage of 420 million kilometers, and a monthly sample analysis and exploration quantity of 220 million. End-to-end architecture enhancement AEB NIO's AEB based on end-to-end architecture has increased scene coverage by 6.7 times. In addition to covering more scenarios, NIO's models have also seen an improvement in maximum braking speed. NIO World Model NWM The two major advantages of end-to-end are the ability to empower artificial logic code with data-driven capabilities, while also automatically extracting information to reduce information loss. NIO has publicly disclosed its world model for the first time. NIO hopes that its model not only has the ability to reconstruct imagination, but also the ability to imagine changes, which is the true understanding of time, and even space. In simple terms, the landing of this model is to make vehicles easier to learn how to drive. Based on the NADArch 2.0 architecture, NIO's point-to-point full-domain navigation assistance 2.0 will be on board in the second half of 2024, and the enhanced version of AEB function has been on board since version Banyan 2.6.5. ![](https://mmbiz-qpic.wallstcn.com/sz_mmbiz_jpg/UnA9a9gzqTBub9XZdRt19JAjcGyaKVrTkUHUFzSYmTBcMGFHbzzU9ncu891yibPLofdTnniamKfxtJqM7YKBlC1Q/640? The Second Generation NIO Phone Li Bin described the second generation NIO Phone as "more stylish, more user-friendly, cleaner, and smarter". The price range of the second generation NIO Phone is between 6499-7499 yuan, with the new phone featuring a Nebula Red color scheme and introducing the EPedition version. The second generation NIO Phone continues to adhere to the zero-advertising concept, being able to identify and intercept some software's built-in "shake to jump" advertisement. The new phone adopts a nationally certified independent security chip to ensure privacy and financial security. In terms of vehicle connectivity, it can replicate destination addresses on the phone for use in car navigation, expand NIO Link application to 21 functions, and more On November 29th, VESYNC spent HKD 5.6445 million to repurchase 1 million shares SkyOS TianShu System released, ShenJi NC9031 successfully taped out, we seem to see NIO building a more solid foundation continuously. In the AI era, what kind of tall building will NIO build? Authors: SiWei, YuFei, Source: Electric Planet News, Original Title: "NIO Technology Day Heavyweight Full Record: Self-developed Chips, Self-developed Full-stack OS, Next-generation Mobile Phones, and More..."