Why do servers need air tightness sealing tests?

Server sealing tests mainly address two scenarios: 1) Outdoor edge/base station servers need to achieve IP55/IP65 ratings to prevent dust and moisture intrusion that can cause motherboard corrosion or short circuits; 2) If the cold plates and pipe joints of liquid-cooled servers fail to seal, coolant leakage can occur, burning out hardware and causing serious accidents. Air tightness testing can quantify leak rates and is a mandatory inspection step before high-end servers leave the factory, providing equivalent verification of IPX7/IP67 ratings.

What are the sealing process requirements for liquid-cooled server cold plates?

High-end cold plates use CNC precision-milled microchannels with a surface roughness of Ra≤0.8μm, combined with O-ring sealing structures to achieve zero leakage. During operation, the coolant flow rate is 5~15L/min, and the pressure drop must be <2bar. Cold plates and pipe joints must pass air tightness leak testing to ensure micron-level leak points are intercepted before leaving the factory, preventing coolant leakage after deployment that could burn out high-value components such as GPUs/CPUs.

What level of precision and efficiency can server sealing inspection achieve?

The PLUS MT100 multi-station leak detector achieves Pa-level test precision and can provide equivalent verification of IPX7/IP67 and higher ratings. It supports multi-station parallel testing, with a single-station test cycle compressible to under 10 seconds, meeting the 100% full inspection requirements of large-scale server mass production. Test data is automatically stored to support quality traceability.

In-depth analysis of server process structure and panorama of industrial ecology-

Q: What level of precision and efficiency can server sealing inspection achieve?

The PLUS MT100 multi-station leak detector achieves Pa-level test precision and can provide equivalent verification of IPX7/IP67 and higher ratings. It supports multi-station parallel testing, with a single-station test cycle compressible to under 10 seconds, meeting the 100% full inspection requirements of large-scale server mass production. Test data is automatically stored to support quality traceability.

I. Detailed Explanation of the Server's Core Manufacturing Process Structure

1.1Motherboard-level core manufacturing process

The server motherboard is the core of the overall process precision. A single board integrates tens of thousands of components, and process control directly determines signal integrity and long-term operational stability.

PCB Substrate Process

High-end servers commonly use14-20 layer low-loss high-speed PCBThe substrate uses M6-grade low dielectric constant fiberglass cloth to control transmission loss in PCIe 5.0 and DDR5 high-speed signals.

Sheet thickness tolerance controlled within±5%Within
Copper foil thickness uniformity deviation ≤0.5 μm
Adopt buried via and blind via processes to increase wiring density and reduce signal crosstalk.

SMT placement and soldering process

Incoming material dehumidification and baking → 3D SPI solder paste inspection → High-precision placement → Nitrogen-protected reflow soldering → 3D AOI visual inspection → X-ray (AXI) solder joint inspection

BGA component void rate ≤5%
CPU socket placement offset ≤25 μm
Temperature difference control in reflow soldering for large-size board surfaces5℃within, to avoid board warping and deformation that could lead to soldering failure

Final Test Verification Process

After welding is completed, execute the following in sequence:ICT(circuit continuity/component parameter testing) andFCT(Functional power-on test), verify the timing of all voltage rails, signal integrity of high-speed interfaces, and stability of memory channels. Nonconforming products must not flow to the next process.

1.2Cooling System Process System

As single-GPU power consumption exceeds 700W, thermal management has been elevated from an auxiliary module to a core system, formingAir cooling, cold plate liquid cooling, immersion liquid coolingThree-tier technology roadmap.

Liquid cooling process

Cold plate liquid cooling supports 600~1200W/U heat dissipation density and is the mainstream solution for AI training servers.

Cooling path	Core Process Composition	Heat dissipation density	Typical PUE	Applicable Scenarios
High-efficiency air cooling	Counter-rotating axial fans (N+1 redundancy) + skived-fin aluminum heatsink + heat pipe vapor chamber, PWM smart speed control	≤ 600 W/U	1.5 ~ 1.6	General-purpose rack servers, low-to-medium computing power scenarios
Cold plate liquid cooling	Copper microchannel cold plate + coolant distribution unit (CDU), flow rate 5–15 L/min, pressure drop < 2 bar	600 ~ 1200 W/U	1.15 ~ 1.25	AI training servers, high-density computing clusters
Immersion liquid cooling	The entire unit is immersed in fluorinated fluid/mineral oil, where phase-change boiling carries away heat, supporting online maintenance.	> 1500 W/U	1.05 ~ 1.1	Supercomputing centers, extreme PUE data centers

Process highlights:High-end cold plates use CNC precision-milled microchannels, with a surface roughness ofRa ≤ 0.8 μm, combined with an O-ring sealing structure to achieve zero liquid leakage; immersion systems require full sealing of all interfaces on the entire machine to prevent coolant evaporation and moisture ingress.

1.3Power Supply and High-Speed Interconnect Processes

Power supply system

Architecture:CRPS redundant power supply, 1+1 or 2+2 hot-swap configuration
Energy efficiency: Mainstream power supplies achieve 80 PLUS Titanium certification, with conversion efficiency >96% at 50% load.
Control: Supports PMBus telemetry, seamless switching via ORing MOSFET, ensuring power supply continuity.
Board-level VRM: Multiphase digital architecture, AI server single CPU/GPU power supply can reach 20+ phases, paired with 90A SPS power stage devices

High-speed interconnect

SupportPCIe 5.0（32 GT/s）、DDR5（6400 MT/s）High-speed transmission
Equal-length routing, impedance control, and differential pair shielding processes ensure signal integrity.
The backplane connector uses gold-plated contacts, with a mating cycle life of >1000 cycles and a contact resistance of ≤ 10 mΩ.

1.4Chassis structure and protective sealing process

Enclosure Protection and Sealing Design

Outdoor edge servers must achieve IP55/IP65 ratings, and air tightness sealing test is a mandatory inspection process before factory delivery.

Main structure process

Rack-mounted server chassis1.0 ~ 1.5 mmThe main body is made of heavy-gauge galvanized steel sheet, formed through CNC punching, bending, and riveting, with the dimensional accuracy of the 1U chassis controlled within±0.2 mmInside, a motherboard tray, hard drive bracket, and air guide are provided to regulate the airflow path.

EMC Electromagnetic Shielding

Install conductive foam at the seams.
The vent adopts a waveguide window structure.
I/O interface is equipped with a filter circuit
Complies with CE and FCC Class A EMC standards

Protective sealing design

General server room servers: Mostly IP30-rated dust protection design
Outdoor edge/base station server:Needs to reachIP55 / IP65, core sealing processes include:
- Install a foamed silicone sealing ring on the box joint surface.
- The interface is sealed with dispensing adhesive.
- Buttons/indicator lights are treated with secondary overmolding.

Seal failure allows dust and moisture to intrude, causing motherboard corrosion, poor contact, and even short circuits; for liquid-cooled servers, seal failure in cold plates and piping can lead to coolant leaks and serious hardware burnout accidents. Therefore,Air tightness and sealing testing is a mandatory inspection process before high-end servers leave the factory., verify that the IP rating is met by quantifying the leak rate.

1.5Complete Unit Assembly and Reliability Verification

The entire machine is assembled using a modular pipeline process, in the following sequence:

Motherboard → Power supply → Cooling → Storage → Cables

Key screws useTorque screwdriverTightening, torque deviation ≤ 5%. Before delivery, the following must be completed:

48-hour full-load aging
High-low temperature cycling
Vibration and shock
Sealing leak testing

Ensure batch-to-batch product consistency.

II. Landscape of Mainstream Server ODM/OEM Manufacturers at Home and Abroad

Global server manufacturing is highly concentrated,Taiwanese ODM manufacturers account for more than 70% of global capacity., domestic local manufacturers are rapidly rising in the Xinchuang, liquid cooling, and edge computing tracks.

2.1Global leading ODM manufacturer

Manufacturer	Headquarters	Core Positioning	Key Customers	Process Advantages
Hon Hai Precision (Foxconn Industrial Internet)	Taiwan, China	World's largest server EMS/ODM	Google, AWS, NVIDIA, Dell, HPE	AI server full-rack manufacturing capability is the strongest, a core OEM for GB200
Quanta Computer (QCT)	Taiwan, China	Leading customized server provider for cloud service providers	AWS, Azure, Oracle	Rack-mounted design and manufacturing integration, with extensive hyperscale customization experience
Wiwynn	Taiwan, China	Leading OCP-standard server provider	Meta, Microsoft	Open-source architecture deeply optimized, liquid-cooling server manufacturing process matured
Inventec	Taiwan, China	HPC and AI Server ODM	AWS, Google Cloud	Outstanding achievements in high-performance computing R&D, leading in multi-GPU high-density integration
Jabil	The United States	High-end server EMS	Dell, HPE, Cisco	North America localized delivery, medical/industrial-grade quality control system
Flex	Singapore	Full-Spectrum Electronics Manufacturing Services	Several European and American enterprise-level manufacturers	Global production capacity layout with strong supply chain integration capabilities

2.2Domestic local server manufacturers

Manufacturer	Headquarters	Core Positioning	Core advantage
Inspur	Jinan, Shandong	Leading domestic server manufacturer, self-developed + self-produced	Leading domestically in liquid cooling technology, first in AI server market share, with a complete line of Xinchuang (domestic innovation) products
Zhongke Controllable	Kunshan, Jiangsu	Core vendors of Xinchuang servers	Strong adaptability to domestic CPU platforms, fully independent and controllable manufacturing system
Lenovo ISG	Beijing	The world's third-largest server manufacturer	Neptune liquid cooling technology, global manufacturing and delivery network, NVIDIA core partner
H3C	Hangzhou, Zhejiang	Core vendors in the government and enterprise market	Integrated network and computing design, with mature edge server manufacturing processes
Great Wall Chaoyun	Beijing	Xinchuang and Liquid-Cooled Servers	Domestic computing power platform integration, large-scale implementation experience with immersion liquid cooling

In-depth analysis of server process structure and panorama of industrial ecology