In-depth analysis of Avita 11 electronic architecture, smart cockpit and smart driving-EEWORLD

Collect

Avita can be said to be born in a "rich family". It is jointly built by Changan Automobile, CATL, and Huawei. Huawei provides a full set of intelligent auto solutions, including mechanical parts, high-voltage electrical systems, smart cars and electronic architecture. Let's take a look at some of the technologies in Avita 11.

01.

Electronic architecture

The electronic architecture mentioned above is provided by Huawei. As early as the 2021 Shanghai Auto Show, Huawei exhibited the computing and communication architecture of the distributed Ethernet network gateway + domain controller (smart cockpit CDC , vehicle control VDC, smart driving MDC) , as shown in the figure below.

▲Figure 1 Domain control architecture proposed by Huawei in 2021

How is it actually implemented? The following figure is a simplified diagram of the electronic architecture of Avita 11. From the figure, we can see the cockpit domain control CDC, intelligent driving MDC and vehicle control unit VCU, which are basically the same as in Figure 1.

▲Figure 2 Electronic architecture of Avita 11

The gateway is connected to PTC ANFD, PTS CAN FD, CHDAND, TCANFD, RCANFD, IC ANFD, and BCAN. It also has 100base_T1, 100base_Tx, and 1000base_T1. The gateway and MDC, T-Box, and CDC are Gigabit networks, and the gateway and body domain control, VCU, and OBD port are 100M networks.

02.

Smart cockpit domain

The cockpit of Avita is undoubtedly powered by Hongmeng system , and the chip is Kirin 990A series, which is built on 7nm process. The CPU adopts 4+4 8-core design, with the highest main frequency of 2.86GHz and AI computing power of 3.5TOPS. It is equipped with 8GB running memory and 128GB storage memory. The figure below is a schematic diagram of the cockpit domain control and its accessories.

▲Figure 3 Cockpit and its connecting accessories

03.

Intelligent driving domain

Another highlight of Avita 11 is its far-leading intelligent assisted driving. First of all, from the hardware point of view, it can be said to be luxuriously configured, MDC810 + 3 laser radars + 6 millimeter waves + 13 cameras + 12 ultrasonic probes. The overall block diagram is shown in the figure below.

▲Figure 4 Intelligent driving hardware configuration

▲Figure 5 Intelligent driving hardware topology diagram

The core here is the MDC platform. Regarding Avita, the official did not give the specific MDC610 or MDC810, but it has little impact on our analysis, because from the introduction of Huawei's intelligent driving platform, the MDC platform unifies a set of software architecture, supports rapid development and serialized sharing of applications, has a unified size for the entire series, and supports smooth upgrades and replacements. In other words, the main control chip should be different, and the peripheral hardware and interfaces are similar.

Taking MDC 610 as an example, its main control chip combination uses an Ascend 610 + an Infineon TC397, and its hardware architecture is shown in Figure 2.

Figure 6 MDC 610 hardware architecture

Its various indicators are as follows:

1. AI computing power is 200 Tops (int8), and ARM CORE's integer computing power is 220K DMIPs;

2. The power consumption of the liquid-cooled version is about 120W;

3. Sensors In terms of interfaces, there are 14 LVDS camera interfaces, 8 Ethernet interfaces, 12 CAN/CANFD interfaces, 6 vehicle Ethernet interfaces, and 1 PPS interface. The definitions of each interface are shown in the figure below;

4. Waterproof grade is IP67;

5. The EMC level is Class 3.

▲Figure 7 Meaning of external interface

In terms of acceleration capability, first, it supports a wide range of AI operators, including mainstream AI frameworks in the industry, such as Caffe, Tensorflow , Pyt or ch, and ONNX, and supports operator libraries of more than 400 mainstream operators; second, it has configurable hardware accelerators, as shown in Figure 4; third, it has Vector Core, which provides tight coupling between vector acceleration and the scalar operation unit of the CPU , accelerates the program segment of mixed control and data parallel operation, and is conducive to the frequently recursive CV algorithm .

▲Figure 8 Hardware acceleration solution

The MDC platform software architecture is shown in the figure below. Its core is the AP+CP+OS independently developed by Huawei.

▲Figure 9 MDC software architecture

Huawei's self-developed Adaptive Autosar complies with R19-11 specifications (and above), and has communication management, execution management, state management, upgrade management, health management, persistence, time synchronization, access management, encryption capabilities, diagnostic services, and network management, as shown in the figure below, and has configuration tools that comply with the Adaptive AUTOSAR specification.

Figure 10 MDC Adaptive Autosar features

AOS is a real-time operating system developed by Huawei. It is compatible with Linux interfaces and has deterministic scheduling, low latency, functional safety and security features. It is also compatible with Linux driver frameworks and third-party libraries.

In terms of openness, it is seamlessly compatible with AUTOSAR. In terms of security, it supports security isolation, safety-critical and non-critical application security isolation, software and hardware co-design, decentralized architecture , shielding of single-point software and hardware failures, and deterministic latency is one-tenth of that of open source Linux, with a kernel latency of less than 10us and support for CC EAL4++.

▲Figure 11 AOS software framework

Finally, in terms of CP, there is VOS based on the Classic AUTOSAR standard, which meets the AUTOSAR CP4.4 specification, provides a complete CAN/ETH protocol stack, diagnosis, NM, calibration, storage and other functions and services, provides a high functional safety operating environment, and supports customers to develop/deploy ASIL-D level services. And like mainstream AUTOSAR tools, graphical modeling & development tools support customers to develop SWC, configure BSW, etc.