共创 | 汽车线束及电子电气架构行业 专家专访（七）

Q1

Could you briefly introduce your presentation topic today?

曹书峰 :

我今天主要想讲AI时代架构开发的赋能与变化。当下AI对各行各业都带来颠覆性影响，在汽车架构开发领域也不例外。一方面，AI让架构形态和设计思想发生改变，从功能定义、软件定义走向AI定义。另一方面，AI本身也在赋能我们开发工作的流程和效率。这两点是我分享的核心。

Q2

What do you think is the biggest change AI brings to automotive E/E architecture?

曹书峰 :

最大改变在于“内核”变了。以前我们围绕功能或软件来设计架构，现在要围绕AI能力来设计,包括数据流转、算力分布、算法部署方式，甚至硬件选型。这意味着架构不再只是承载功能的骨架，而要成为支撑AI持续进化的有机体，从静态集成转向动态可成长。

Mr. Cao Shufeng :

The biggest change is that the "core" has shifted. Previously, we designed architectures around functions or software. Now, we have to design around AI capabilities, including data flow, compute distribution, algorithm deployment methods, and even hardware selection. This means architecture is no longer just a "skeleton" that carries functions, but must become an organism that continuously supports AI evolution, moving from static integration to dynamic growth.

Q3

Under the vehicle-cloud integrated architecture, data is both an asset and a burden. How do you decide which data is worth uploading? Is your collection strategy rule-triggered or full-record-then-filter?

曹书峰 :

首先要坚守法规底线。像GBT32960要求的安全数据必须上传，而隐私数据不能离车。在此基础上，我们会做数据清洗，目前云端清洗效率更高，车端清洗会带来能耗和任务阻塞问题。至于采集策略，我们更倾向于按规则触发+云端筛选结合，而不是全量记录，因为并非所有数据都有高价值，要平衡成本与效用。

Mr. Cao Shufeng :

First, we strictly comply with regulations. Safety data are required by standards like GBT32960 must be uploaded, while privacy data cannot leave the vehicle. On that basis, we clean the data, and currently cloud-based cleaning is more efficient than on-vehicle cleaning which consumes energy and blocks tasks. As for collection strategy, we prefer a combination of rule-triggered collection and cloud-side filtering, rather than full recording, because not all data is high-value and we need to balance cost and utility.

Q4

现有量产车型如果升级到新架构，兼容性怎么处理？是硬件预埋等待OTA，还是必须等下一代车型？

For mass-produced models upgrading to a new architecture, how do you handle compatibility? Do you pre-install hardware and wait for OTA, or must it wait for the next-generation model?

曹书峰 :

两种思路并存。比如鸿蒙智行最近发布的新车，就做了L3级架构预埋，等法规落地后通过OTA开通。但预埋带来成本和冗余，消费者是否买单需要考量。所以大部分车企还是选择等到真正需要时再上全新架构。这没有标准答案，取决于各家对市场节奏和成本压力的判断。

Mr. Cao Shufeng :

Both approaches exist. For example, Harmony Intelligent Mobility recently announced pre-installed L3-level architecture on their new models, to be activated via OTA once regulations permit next year. But pre-installation adds cost and redundancy, and it's uncertain whether consumers will pay for it. So most OEMs choose to wait until L3 is truly needed before rolling out the new architecture. There's no one-size-fits-all answer. It depends on each company's assessment of market timing and cost pressures.

Q5

In developing AI models, like core algorithms, training, and on-vehicle deployment, how are these responsibilities divided between OEMs and software suppliers?

曹书峰 :

传统分层模式已经变成融合模式，边界越来越模糊。以前供应商可能包揽整个控制器，现在主机厂会自研一部分，供应商负责其中某个模块，甚至底软、应用、中间件由不同供应商分担。AI让主机厂有了更强的软件能力，也更愿意深入参与，分工正从黑盒交付走向联合开发。

Mr. Cao Shufeng :

The traditional layered model has given way to an integrated one, with increasingly blurred boundaries. Previously, a single supplier might handle the entire controller. Now OEMs develop some parts in-house, while suppliers take on specific modules. Low-level software, applications, middleware may come from different vendors. AI has given OEMs stronger software capabilities and a greater willingness to dive deeper. The division is shifting from black-box delivery to co-development.

Q6

As architecture moves from "function integration" to "AI-defined," have you adjusted your internal organizational structure? How do different teams collaborate now?

曹书峰 :

我们除了成立正式的AI部门，还在各业务单元设立AI虚拟组织，大家跨部门协作探索AI赋能场景。目前行业没有成熟范式，每家都在摸索，这种“虚实结合”的组织方式能灵活调动资源，快速试错，同时避免过度重构带来的管理成本。

Mr. Cao Shufeng :

In addition to establishing a formal AI department, we've set up virtual AI teams across various business units to explore AI empowerment scenarios through cross-functional collaboration. Since there's no mature paradigm in the industry yet, every company is experimenting. This "virtual-plus-formal" structure allows us to flexibly allocate resources, iterate quickly, and avoid the overhead of excessive organizational restructuring.

Q7

From a developer's perspective, what are the most noteworthy technical breakthroughs in E/E architecture evolution over the next 2–3 years?

曹书峰 :

我重点关注两个方向：一是光模块上车，光纤替代铜缆能大幅提升带宽和传输效率；二是芯片突破，尤其是能否实现真正高端的舱驾一体，目前高通方案还主要支持中低端，更高性能的融合芯片一旦落地，架构物理形态会随之改变

Mr. Cao Shufeng :

I'm focusing on two directions: first, optical modules going into vehicles. Fiber optics replacing copper cables can greatly increase bandwidth and transmission efficiency. Second, chip breakthroughs, especially whether we can achieve true high-end cockpit-driving integration. Currently, Qualcomm's solutions mainly support low-to-mid-end integration. Once more advanced fusion chips become available, the physical form of architecture will change accordingly.

Q8

How do you find today's conference?

曹书峰 :

非常好。各行各业尤其是主机厂和上下游供应商能聚在一起交流，非常难得。俗话说“独行快、众行远”，这样的平台能让大家思想碰撞、达成共识，对整个行业发展都很有价值。

Mr. Cao Shufeng :

It's excellent. It's rare to have such a platform where OEMs and upstream/downstream suppliers from all sectors can come together and exchange ideas. As the saying goes, "If you want to go fast, go alone; if you want to go far, go together." This kind of platform fosters thought exchange and consensus-building, which is very valuable for the entire industry's development.