A conceptual illustration of automotive liquid cooling and heat sink approaches discussed in Yao-Nian Zhong’s engineering research that can help tackle heat and manufacturing challenges in the AI era. [Photo provided to chinadaily.com.cn]
As artificial intelligence moves rapidly into cars, robots and industrial equipment, with computing power becoming a core part of the physical systems around it, a Chinese engineer is tackling the heat and manufacturing challenges in the AI era — linking autonomous driving hardware experience with AI-assisted heat sink and liquid cooling research.
Yao-Nian Zhong, a mechatronic engineer with nearly 14 years of experience, thinks devices are being asked to perceive their surroundings, process large volumes of data and make decisions in real time, often while operating continuously in confined spaces. Yet stronger algorithms and faster processors do not automatically translate into products that can be deployed at scale.
“Higher power density brings greater heat loads, while compact packaging, energy use, environmental protection and long-term reliability place additional demands on hardware design and manufacturing,” Zhong said.
For Zhong, these constraints are not secondary details. They form the hardware foundation that determines whether advanced computing systems can move from laboratory performance to dependable operation in the real world.
The AI era will require smarter algorithms, but it will also require a more reliable hardware foundation, he said, noting that connecting the two is becoming an increasingly important part of turning advanced computing into scalable applications.
“I focused on how components can be arranged within limited space, how heat can be transferred efficiently, and how design decisions can account for validation, manufacturing and long-term use from the beginning of product development,” he said of his road to find answers in autonomous driving hardware.
While working on onboard unit and roadside unit products associated with Baidu’s Apollo autonomous driving ecosystem, Zhong was responsible for structural design work involving printed circuit board placement, heat dissipation, interface configuration, production cost and assembly feasibility.
“Such equipment must continuously process vehicle, road and communications data while operating under space, environmental and reliability constraints,” he said. “As processors, power supplies and communications modules are concentrated into smaller enclosures, heat accumulation, signal interfaces, assembly tolerances and component reliability become increasingly interconnected.”
His project records showed that the redesigned products reduced cost by about 20 percent and chassis volume by nearly 50 percent compared with the preceding generation models. The products were also used in autonomous driving demonstration areas.
“The reduction in volume was not simply a matter of making an enclosure smaller,” Zhong said, noting that compressing more components into a limited space can create new thermal, mechanical and assembly risks if the system is not considered as a whole.
Zhong’s career has spanned industrial equipment, computing products and automotive electronics, helping him to come to a conclusion: manufacturability cannot be postponed until after a design is completed.
“For high-performance computing equipment, thermal management should not be treated as a component added at the end of development,” Zhong said. “It needs to be considered from the start together with PCB placement, power consumption, reliability and manufacturing conditions.”
So he used AI to widen the design space. In a 2025 IEEE-published study on heat sink design, he combined a physics-informed generative AI model with multi-objective optimization to explore structures beyond conventional geometries and experience-based parameters. The study evaluated not only temperature reduction, but also airflow resistance, structural dimensions and manufacturability.
“My aim is not to replace engineers with AI, but to use algorithms to expand the design space, shorten the screening process and bring manufacturing constraints into research at an earlier stage,” Zhong said.
However, as the power density of chips and computing devices continues to increase, conventional air-cooling faces growing pressure.
In another IEEE-published study, Zhong turned to liquid cold plates and used topology optimization and AI-based methods to examine internal flow-channel designs.
The study evaluated thermal resistance, pressure loss and temperature uniformity together, while using AI to accelerate comparisons among different channel configurations.
From structural optimization for autonomous driving computing products to AI-assisted heat sink design and liquid cooling research, Zhong’s work has followed a consistent question — how can increasingly powerful computing systems remain stable in real operating environments and still be practical to manufacture and deploy?
“As AI becomes more deeply integrated with vehicles, industrial manufacturing and other physical devices, competition will depend on more than algorithms and chip performance,” Zhong said. “Structure, thermal management, energy efficiency, reliability and production capability will together determine whether a technology can become a durable product.”
“Engineering innovation is not only about proposing a new design. It is also about making sure the design can pass testing, enter production and operate reliably in real-world conditions,” he said.
Please contact the writer at [email protected]
Tanks to chinadaily.com.cn
Please visit:
Our Sponsor