DEV Community: Maggie‌ Wang@AnyPCBA

The Limits of Copper: What Happens to Manufacturing Physics When a PCB Carries 1,000 Amps?

Maggie‌ Wang@AnyPCBA — Thu, 05 Nov 2026 03:43:10 +0000

As a software engineer or application developer, you might think of "current" as just a parameter on a power adapter. But if you follow hardware trends in 2026, you'll see that current in the physical world is reshaping the entire electronics manufacturing industry at an alarming rate.

NVIDIA's latest Blackwell architecture GPU draws nearly 1,000W per chip. Because the core voltage is extremely low (under 1V), it must instantaneously draw over 1,000 Amps. At the same time, 800V high-voltage platforms in electric vehicles are forcing on-board chargers (OBCs) and motor controllers to handle massive power.

Under such extreme conditions, conventional PCB traces (standard 1oz copper, about 35μm thick) would vaporize instantly like a blown fuse.

To keep PCBs alive under these extreme currents, electronics manufacturing has deployed its ultimate weapon: Heavy Copper PCBs.

Today, from the manufacturing perspective at AnyPCBA, let's dive into what physical miracles—and process disasters—occur in the factory when copper thickness surges to 10oz or 20oz (around 700μm).

1. Rebuilding Physical Form: From "Traces" to "Physical Busbars"

On standard PCBs, we use 1oz thin copper traces to transmit a few hundred milliamps of signal. But at 1,000 Amps, traces are no longer traces—they become physical busbars embedded in resin.

A Quantum Leap in Current Capacity: A 20oz copper busbar has a cross-sectional area 20 times that of standard copper foil. It can withstand extreme current surges like a thick cable, and its extremely low resistance drastically reduces Joule heating (I²R).
A Structural Revolution in Thermal Management: High current inevitably brings high heat. Heavy copper isn't just for conduction—it's a massive heatsink. Large areas of thick copper quickly spread heat laterally from beneath the chip and conduct it to the liquid cooling cold plate on the back. In AI servers, heavy copper's thermal mass can even "shave peaks and fill valleys," delaying the onset of thermal throttling.

2. The Factory's "Hell Mode": The Process Limits of Heavy Copper

Drawing a heavy copper board takes a few mouse clicks. But in the factory, producing 10oz–20oz heavy copper boards challenges physical limits every second. At AnyPCBA, we've identified three fatal process bottlenecks:

① The "Side-Etch Disaster"
When a factory uses chemical etchants to dissolve 20oz copper, the chemicals don't just eat downward—they aggressively attack the sides. This is called undercut.

The Consequence: If your designed trace width is 10 mil, severe undercutting could leave 6 mil in production, causing impedance to go completely out of control. Worse, the side erosion thins the trace edges, making them highly prone to open-circuit failure under current surges.
The Solution: The factory must use special differential etching techniques or ultra-thick dry film to shield the sides, along with micron-level control of etchant concentration, temperature, and spray pressure.

② The Prepreg "Resin Starvation" Crisis
20oz copper is so thick it resembles raised steel plates. During multilayer lamination, the prepreg resin struggles to completely fill the massive gaps between heavy copper features. Incomplete filling creates internal voids (delamination), leading directly to high-voltage breakdown or interlayer shorts.

The Solution: High-resin-content, high-flow specialty prepregs are mandatory. For extreme copper designs like 20oz, the factory may even require multiple lamination cycles to guarantee absolute resin fill and interlayer adhesion.

③ Drilling and Plating "Blowout" Risks
Drilling vias through 20oz copper wears out drill bits rapidly, leaving rough hole walls. More critically, achieving uniform plating inside the hole is extremely difficult. If plating is uneven, the via becomes a resistive heater under high current, eventually leading to barrel cracks.

The Solution: High-precision CNC drilling machines with specially coated drill bits; and in plating, reverse-pulse plating technology to ensure deep-hole and heavy-copper plating thicknesses reach or exceed 25μm for current-carrying safety.

Practical "Mine-Sweeping" Advice for Hardware Engineers
If your hardware project involves high current (AI compute boards, automotive power, solar inverters), your design mindset must fundamentally shift:
Abandon "Rule of Thumb" Thinking: Never estimate using the conventional "1mm carries 2A" rule. Under high current, use professional tools (like Saturn PCB Toolkit) to calculate trace width based on ambient temperature, copper weight, and allowable temperature rise.
Avoid Long, Thin Traces—Use Teardrops and Copper Pours: At high-current nodes, avoid thin traces. Use large copper pours to reduce current density. Add teardrops at trace-to-pad connections to prevent tip discharge or thermal runaway at corners under high current.
Align with DFM Early: The minimum trace width/spacing limits for heavy copper boards are far more restrictive than standard boards. Before designing, confirm process windows with AnyPCBA's engineering team—otherwise, a perfect design on paper may be impossible to etch in the factory.

Conclusion: At the End of Compute Lies the Foundation of Physical Manufacturing
The explosive growth of AI and electric vehicles is pushing PCBs to the limits of physical law. Software engineers pursue maximum throughput; hardware engineers must carry a thousand amps across millimeters of copper.

At AnyPCBA, we have the extreme process capability to handle 20oz heavy copper boards, providing a rock-solid physical foundation for AI compute, new energy, and industrial power. Our DFM team doesn't just catch process pitfalls—we understand the application of materials science and thermodynamics in manufacturing.

If you're designing high-power, high-current hardware and facing bottlenecks in thermal management, current-carrying capacity, or high-voltage insulation, contact us through our website.

Have you encountered copper heating, blistering, or open-circuit failures caused by high current in your projects? Share your war stories in the comments.

The Physical Challenges of the Chiplet Era: The End or Rebirth of PCB Manufacturing?

Maggie‌ Wang@AnyPCBA — Mon, 02 Nov 2026 02:19:12 +0000

If you follow semiconductor news, you've surely heard the term: Chiplet.

In the past, chip design was a monolithic architecture. To integrate the CPU, GPU, and memory controller onto a single chip, you either had to tolerate extremely low yields or make the chip area massive.

But now, the industry is fully pivoting to Chiplet architecture—breaking a large chip into multiple smaller dies (e.g., a compute die, an I/O die) and assembling them together using advanced packaging technologies (like 2.5D/3D packaging). It's like building with Legos.

As a software developer, this sounds wonderful: decoupling, modularity, and on-demand composition.

But as someone working in PCB manufacturing at AnyPCBA, I see a brutally harsh physical reality behind this trend: When the internal interconnects of a chip become incredibly dense, the role of the PCB (Printed Circuit Board) is being completely disrupted.

1. The "Motherboard" at the Chip Level: Advanced Packaging Is Eating the PCB's Lunch

On traditional computer motherboards, the CPU and memory are separate, communicating through traces on the PCB.

But in Chiplet architecture, data transfer rates are so high that signals can't travel far on the PCB. They must be integrated onto a "silicon interposer" or "RDL" (Redistribution Layer)—which is essentially an extremely miniaturized "micro-motherboard" (advanced packaging substrate).

The Impact on PCB Manufacturing: Traditional organic resin PCBs (like FR-4) cannot meet such high-density routing demands. The substrate underneath the chip is evolving from a "PCB board" into "semiconductor-like" processes (such as ABF substrates). For traditional PCB factories, this is a life-or-death cross-industry challenge.

2. High-Density Fan-Out: Where Is the Limit of the PCB?

Although a massive amount of signal routing is resolved inside the package, the Chiplet chip must ultimately be soldered onto a System PCB.

Because the Chiplet integrates a massive number of pins internally, the BGA (Ball Grid Array) pitch on the PCB becomes extremely fine (even smaller than 0.3mm). This means:

Traditional mechanical drill bits are completely obsolete; laser drilling (microvias) is mandatory.
Traces must be incredibly fine (mSAP process).
PCB layer counts are skyrocketing (high-end AI server motherboards reach 30+ layers).

The Impact on PCB Manufacturing: Manufacturing such motherboards requires extreme **layer-to-layer alignment precision and impedance control. **If PCB warpage is not well-controlled, the Chiplet chip soldered onto the motherboard will crack due to stress, causing the entire system to fail.

3. Power Integrity: The "Vampiric" Effect of Chiplets

Although Chiplet architecture is flexible, it concentrates multiple high-power compute units into an extremely small area.

When these chiplets run at full load instantaneously, they draw terrifying amounts of current from the PCB motherboard. If the board's Power Delivery Network (PDN) is poorly designed, the voltage will droop instantly, causing a system crash. This is the hardware equivalent of "high concurrency crushing the server."

The Impact on PCB Manufacturing: Ultra-heavy copper must be used to carry high currents, with massive decoupling capacitors placed beneath the Chiplet. This imposes extremely stringent requirements on PCB stack-up and material selection.

Conclusion: "Decoupling" in the Physical World Is Harder Than in Software

In software, decoupling only requires refactoring code. In the physical world, chip decoupling (Chiplet) is forcing the entire electronics manufacturing supply chain—including advanced packaging, high-end PCB, and PCBA—to make a technological leap.

Although advanced packaging is taking on more and more interconnect tasks, this absolutely does not mean the PCB is becoming less important. On the contrary, the PCB is becoming the "hardcore skeleton" that carries these high-density compute modules.

At AnyPCBA, we are closely monitoring the impact of Chiplets and AI computing on PCB manufacturing processes. We currently support PCB manufacturing up to 64 layers, with HDI and high-frequency hybrid capabilities, and have provided prototyping and small-to-medium volume production services for multiple AI hardware startups.

If you are working on hardware design related to advanced packaging, or need manufacturing support for high-density, high-layer-count PCBs, we welcome you to reach out.

Do you think Chiplet architecture will ultimately make PCBA simpler or more complex? Share your thoughts in the comments.

The Physics of AI: Why Your Next Server Needs Liquid Cooling and Heavy Copper

Maggie‌ Wang@AnyPCBA — Thu, 29 Oct 2026 06:21:42 +0000

If you follow the tech news, you know that the AI arms race is currently a race for power.

NVIDIA’s latest Blackwell architecture GPUs are pushing power consumption to unprecedented levels—approaching 1,000W per chip, with entire server racks now drawing over 100kW.

As software developers, we usually think of scaling in terms of cloud instances and Kubernetes clusters. But at the physical layer, scaling AI means dealing with extreme thermodynamics. The hardware running your models is hitting the absolute limits of physics.

Here is why AI is forcing a massive shift in server PCB design, and why traditional air cooling and standard copper are no longer enough.

1. The End of Air Cooling: Liquid is the New Normal

For decades, servers were cooled by massive fans pushing air through the chassis. But air has a very low specific heat capacity. It simply cannot carry away the massive heat generated by a 100kW+ AI rack.

This is why tech giants and data centers are rapidly transitioning to Direct-to-Chip (D2C) liquid cooling.

Instead of blowing air, cold plates are mounted directly on top of the hottest components (GPUs and CPUs), and a dielectric or water-based coolant is pumped through them.

The Hardware Challenge: Liquid cooling introduces immense mechanical and chemical stress. The PCB must be highly resistant to moisture and chemical degradation. More importantly, the CTE (Coefficient of Thermal Expansion) of the PCB materials must perfectly match the heavy metal cold plates. If they expand and contract at different rates, the board will warp, and the BGA solder joints under the GPU will crack.

2. Power Delivery: The Heavy Copper Revolution

A GPU drawing 1,000W at sub-1V voltages requires immense current—often exceeding 1,000 Amps.

On standard PCBs, delivering this much current would instantly vaporize the traces. To handle this, AI server boards are shifting to Heavy Copper PCBs (up to 10oz or even 20oz). These aren't just traces; they are physical copper busbars embedded in the board.

The Manufacturing Challenge: Etching 20oz copper is a nightmare. The undercut effect (where etching chemicals eat sideways into the trace) is severe. If a designer routes a 10 mil trace on a heavy copper board without accounting for etching tolerances, it could short out or break during production. This requires highly specialized DFM (Design for Manufacturing) reviews and advanced etching techniques.

3. The Signal Integrity Nightmare: 224G PAM4

To feed data to these monstrous GPUs, networking speeds are jumping from 112G to 224G per lane.

At 224G, signal attenuation through standard FR-4 materials is catastrophic. The data literally dissolves into noise before it reaches the chip.

The Hardware Challenge: This forces the industry to use ultra-low-loss materials (such as M8 or M9 grade laminates) and ultra-smooth copper foils. It also demands extreme precision in impedance control. Even a 5% deviation in trace width or dielectric thickness can destroy signal integrity.

4. The Shift to High-Layer Count HDI

Because the physical distance between the GPU and memory (HBM) must be minimal to reduce latency and power loss, AI server boards are incredibly dense.

Standard boards won't cut it. AI accelerators now use 20 to 30-layer HDI boards (High-Density Interconnect). The BGA pitch is becoming so small that conventional mechanical drills cannot be used. Laser-drilled microvias are mandatory.

The Bottom Line

The AI revolution is not just a software revolution. It is a brutal test of materials science, thermodynamics, and precision manufacturing.

At AnyPCBA, we are seeing this transition firsthand. We support the manufacturing of high-layer HDI, heavy copper (up to 20oz), and high-frequency PCBs for demanding applications. We know that when you are dealing with 100kW racks and 224G signals, a "working" design on a breadboard means nothing without rigorous DFM reviews and precise impedance control.

If you are building hardware for the AI era, or if you need a manufacturing partner that understands the physics behind heavy copper and HDI, let's talk.

At AnyPCBA, we specialize in small-to-medium volume manufacturing of advanced PCBs and PCBA. We hold ISO 13485 and IATF 16949 certifications and are equipped to handle heavy copper, HDI, and rigid-flex boards. Contact us through our website to discuss your high-reliability projects.

What's the biggest physical constraint you've faced when designing hardware for high-performance computing? Share your experience in the comments!

The Open-Source Hardware "Mass Production Curse": Why Your GitHub Project Never Leaves the Breadboard

Maggie‌ Wang@AnyPCBA — Mon, 26 Oct 2026 03:28:47 +0000

If you spend any time on GitHub or Hackaday, you'll notice a fascinating pattern: many brilliant open-source hardware projects—like custom mechanical keyboards, smart home hubs, or desktop robots—look stunning in the creator's video. But if you try to replicate them, or if the creator tries to turn them into a marketable product, the project often fizzles out.

In the software world, you write good code, git push it, and developers worldwide can git clone and run it. The marginal cost of replicating software is zero.

But in the hardware world, replicating a project is never as simple as downloading a few files.

As someone working on the manufacturing side at AnyPCBA, I frequently receive prototype orders from individual developers and small maker teams. Today, let's look at why open-source hardware faces the "mass production curse" from a manufacturing perspective, and how developers can bridge that gap.

1. The Lethal BOM: When Your "Dependencies" Vanish

When writing code, if a package you rely on becomes deprecated, you can fork it and maintain it yourself.

But in hardware, your package.json is the BOM (Bill of Materials). Many makers populate their BOM with a specific chip they found on Taobao, LCSC, or Mouser. But when you open-source the project, and someone tries to build it months later, they discover the chip is End-of-Life (EOL) or has a 52-week lead time.

You cannot "fork" a silicon chip. You have to find an alternative, redesign the board, and rewrite the firmware to adapt.

Developer Mindset Shift: Always prepare a "fallback" for your BOM. Note Pin-to-Pin compatible alternatives and regularly check lifecycle statuses.

2. Passing DRC ≠ Manufacturable

Running the DRC (Design Rule Check) in KiCad or Altium and seeing zero errors makes many developers think they're done.

But factory machines don't care about DRC; they care about the physical process window. For example:

Your minimum trace width is 4 mil, but this is right at the factory's etching limit—yield cannot be guaranteed.
You placed vias directly on a BGA pad (via-in-pad) for thermal dissipation but didn't specify "resin plugging." During reflow, solder paste wicks through the hole, causing a cold joint.
Your panelization lacks edge rails, so the pick-and-place machine can't transport it, or depanelization rips off edge capacitors.

Developer Mindset Shift: Before sending your Gerber files to a factory, check their "Process Capabilities" table. In software, the compiler tells you if it works; in hardware, you need the factory's DFM engineer to tell you if it works.

3. The "Hand-Soldering" Trap

On your workbench, you can hand-solder 0402 or even 0201 resistors with a fine-tipped iron, and check for shorts with a multimeter.

But when 100 people replicate your project, or you decide to build 50 units to sell, you'll eventually need an SMT assembly factory.

To an assembly factory, your project is a "non-standard" job. They need Pick-and-Place files, a complete BOM, and most importantly—Test Points. If you didn't reserve ICT (In-Circuit Test) points during design, the factory can't detect defects after assembly. You'll have to manually plug in and test every board, leading to high labor costs and scrap rates.

Developer Mindset Shift: Design test points like you write unit tests. Expose key power rails and signals—this is the lifeline of production yield.

4. Certification: The Final Boss of Open-Source Projects

If you're just building a toy for your desk, it doesn't matter. But if you want to bring it to market, you'll face FCC (US), CE (Europe), RoHS, and other compliance certifications.

Exceeding EMC radiation limits or using non-eco-friendly solder... these issues won't show up in the lab, but in a certification chamber's anechoic room, they can kill your product before it launches.

The Leap from Maker to Founder

The spirit of open-source hardware is sharing, but to truly bring it to the world, you need to introduce manufacturing thinking.

At AnyPCBA, we love serving developers and makers. We know that every world-changing hardware product starts as a breadboard. Our DFM team frequently provides free design reviews, pointing out fatal details that would ruin mass production. We support small-to-medium volume production from 5 to 5,000 pieces—we don't mind prototypes; we care about your ideas.

If you have a finished circuit board or an incredibly cool open-source project but no idea how to turn it into a real product, contact us through our website.

What "physical traps" have you encountered when replicating open-source hardware or building your own PCBs? Share your horror stories in the comments!

The Open-Source Hardware "Mass Production Curse": Why Your DIY Project Never Leaves the Breadboard

Maggie‌ Wang@AnyPCBA — Thu, 22 Oct 2026 03:04:48 +0000

If you've spent any time on GitHub or Hackaday, you've probably noticed a pattern: many brilliant open-source hardware projects never make it past the "prototype" stage.

You've seen them—the smart home hubs built with a Raspberry Pi and a spaghetti of jumper wires, or the desktop robots running on breadboards. They work flawlessly in the creator's YouTube video. But if you try to replicate one, or if the creator wants to turn it into a product that people can actually buy, you hit a brutal reality:

The gap between "it works on my desk" and "it's a manufacturable product" is a massive chasm.

In the software world, when you write cool code, you just git push. Developers anywhere can git clone it and run it locally in seconds. But in the hardware world, replicating a project isn't just about downloading a few files.

I work at AnyPCBA, where we frequently receive prototype orders from makers and independent developers. Today, let's talk about why open-source hardware struggles to scale into production, and how to bridge that gap.

1. The Lethal BOM (Bill of Materials)

Open-source software relies on npm, pip, or Maven. Open-source hardware relies on the BOM (Bill of Materials).

Many makers write their BOMs by simply pasting links from Taobao, LCSC, or Mouser. But when you open-source your project, and someone tries to build it three months later, they discover that the specific capacitor or IC you used is now End-of-Life (EOL), or has a 52-week lead time.

In software, if a dependency is abandoned, you can fork it and maintain it yourself. But in hardware, if a power management IC goes EOL, you can't "fork" a silicon chip. You have to re-select a new component, redesign the board, and re-validate the reliability.

How to bridge the gap: Always provide a "pin-to-pin compatible" alternative in your open-source BOM, and run lifecycle checks on critical components.

2. DRC Passed ≠ Manufacturable

In KiCad or Altium, hitting "Run DRC" and seeing zero red errors makes many makers think they're done.

But factory machines don't care about your DRC. They care about the process window. For example:

Your minimum via is 0.2mm, but the factory's mechanical drill limit is 0.3mm.
You placed thermal vias directly on a BGA pad (via-in-pad) without specifying "resin plugging and plating over." During reflow, the solder paste will wick right through the hole.
Your panelization has no edge rails (breakaway rails), so the pick-and-place machine can't transport it, or depanelization rips the MLCC capacitors off the edge.

How to bridge the gap: Before sending your Gerber to the factory, don't just rely on software DRC. Check the factory's "Process Capabilities" table to ensure your trace width, hole spacing, and annular rings are within their safe range.

3. The "Hand-Soldering" Trap in Assembly and Testing

On a personal workbench, you can hand-solder 0201 resistors and check for shorts with a multimeter one by one. But when you need 50 or 100 units, you'll have to find an SMT assembly house.

To an assembly factory, your project is a "non-standard" job. You need to provide the correct Pick-and-Place file, a complete BOM, clear silkscreen orientation markings, and most importantly—Test Points.

If you didn't reserve test points for ICT (In-Circuit Testing) or flying probe testing during layout, the factory can't help you detect cold joints and shorts after assembly. This means you have to plug in every board and test it manually, leading to high defect rates and labor costs.

How to bridge the gap: Proactively reserve test points for key signals and power rails during design. If it's an open-source project, try to use standard footprints and mark polarity clearly on the silkscreen.

4. Certification: The Final Boss of Open-Source Projects

If you're just making a toy for your desk, it doesn't matter. But if you want to turn it into a product to sell to others (especially in Europe or the US), you'll face EMC, FCC, CE, and RoHS certifications.

High-frequency signals not handled well? EMC radiation exceeds limits, and the test fails.
Using cheap, non-eco-friendly solder? RoHS fails, and customs seizes your shipment.

Many brilliant maker projects get killed immediately when sent for EMC pre-testing, requiring a complete redesign.

Conclusion: Open-Source Hardware Needs "Manufacturing Thinking"

Open-source hardware isn't just about uploading code and schematics to GitHub. To truly bring it to the world, you need to introduce "manufacturing thinking."

At AnyPCBA, we love serving makers and open-source projects. Because we know that every great product starts as a breadboard prototype. Our DFM (Design for Manufacturing) team frequently provides free DFM checks for customers, pointing out the fatal details that would ruin mass production.

If you have a finished circuit board design but no idea how to turn it into a real product, contact us through our website. We support small-to-medium batch production from 5 to 5,000 pieces. We don't mind prototypes; we care about your ideas.

What "manufacturing traps" have you encountered in your open-source hardware or DIY projects? Share your horror stories in the comments.

The "Extreme Tests" of a PCB: From Thermal Shock to Drop Tests, What Is the Factory Actually Testing?

Maggie‌ Wang@AnyPCBA — Mon, 19 Oct 2026 02:09:34 +0000

At AnyPCBA, when a board is destined for automotive, medical, or industrial equipment, it must undergo a series of "extreme tests" before it leaves the factory.

Many engineers assume testing just means "power it on and see if it lights up." But in reality, reliability testing is a comprehensive interrogation at the physical, chemical, and electrical levels. Its purpose isn't to prove the board "works"—it's to find out under what conditions it will fail.

This article breaks down the four "extreme tests" that every high-reliability PCB must endure, from the manufacturing perspective.

1. Thermal Shock and Thermal Cycling: Simulating "Ice and Fire"

Electronic devices experience drastic temperature changes in real-world use. A car starting on a winter morning is at -40°C, and the engine bay can quickly reach +125°C. Medical devices also experience rapid temperature changes during sterilization and cooling.

Test methods:

Thermal Cycling: The board is placed in a thermal chamber and cycled between -40°C and +125°C (e.g., 500 or 1,000 cycles).
Thermal Shock: More severe than thermal cycling, usually using a two-chamber method (liquid-to-liquid), switching between high and low temperatures in a very short time (e.g., a few seconds).

Purpose:
To verify the compatibility of different materials (copper, FR-4, solder mask) under thermal expansion and contraction. If the coefficients of thermal expansion (CTE) differ too much, via walls can be pulled apart (microcracks), or solder joints can fatigue and crack.

Manufacturing perspective:
For high-layer-count boards and HDI, we pay special attention to interlayer adhesion. If the stack-up is asymmetric or copper distribution is uneven, delamination is highly likely during thermal shock. This is why we require balanced copper in the DFM stage.

2. Damp Heat Testing (THB / HAST): Accelerated "Aging"

Moisture is a silent killer of electronics. When water vapor penetrates the substrate, it not only reduces insulation performance but also triggers Conductive Anodic Filament (CAF) growth, causing short circuits.

Test methods:

Temperature-Humidity Bias (THB): The board is subjected to a 85°C, 85% RH damp heat environment with a bias voltage applied for hundreds of hours.
Highly Accelerated Stress Test (HAST): Higher temperature (e.g., 130°C), higher humidity, and higher pressure, simulating years of natural aging in just tens of hours.

Purpose:
To evaluate insulation reliability of materials and processes in harsh environments. Especially for medical and automotive boards, it must be ensured that ionic migration between conductors stays within safe limits under long-term high-humidity conditions.

Manufacturing perspective:
This depends on the substrate's CAF resistance rating and the cleanliness of the via walls. If there is flux residue or ionic contamination during manufacturing, HAST will immediately expose it. That's why, for high-voltage boards like electrosurgical generators, we require extremely high ionic cleanliness (<1.56μg/cm²).

3. Drop and Mechanical Shock: Testing Durability

Boards must not only work but also withstand the impacts and drops during transport and installation.

Test methods:

Drop Test: The assembled PCBA or device is dropped from a specified height (e.g., 1 meter) to check if the board deforms, components detach, or solder joints crack.
Vibration Test: Simulating continuous vibration during vehicle driving or aircraft takeoff/landing.

Purpose:
To verify the mechanical strength of the board. Large or heavy components (such as large capacitors, inductors, connectors) can easily cause solder joint tearing during drop or vibration if they lack sufficient fixation or stress relief in packaging or layout.

Manufacturing perspective:
During DFM review, we focus on whether "large components are too close to the board edge" or "fixed in flexible areas." If there are heavily suspended components (especially long-lead through-hole parts), we recommend adding glue fixation or bottom support.

4. Solderability Test: The "Foundation" of Soldering

No matter how good the design or how expensive the materials, if the pads or pins are not solderable, everything is moot.

Test methods:

Solderability Test: Pads or pins are dipped into molten solder to observe the wetting area.
Steam Aging: Simulates the degree of surface oxidation after components have been stored in a warehouse for a period of time.

Purpose:
To confirm whether the surface finish process (such as ENIG, OSP) is effective. If the surface finish is too thin or severely oxidized, solder cannot spread on the pad, directly causing cold joints or weak soldering.

Manufacturing perspective:
This depends directly on the factory's surface finish process control. For example, if ENIG is not well controlled, "black pad" (nickel layer corrosion) can occur, leading to soldering failure. Before shipment, we perform regular sampling tests on surface finishes.

5. Why Perform These "Destructive" Tests?

Because these tests can expose risks before mass production, rather than causing disasters on the customer side.

A board that passes all tests cannot be guaranteed 100% problem-free, but it can be guaranteed that it will not fail due to basic physical and chemical laws.

For medical devices, automotive electronics, and industrial control equipment, "zero field failures" is never a matter of luck—it is forced out by these rigorous physical tests.

At AnyPCBA, our high-reliability PCBA manufacturing process covers full support from thermal shock and damp heat testing to cleanliness verification. We serve small-to-medium volume projects from 5 to 5,000 pieces, holding ISO 13485 (medical) and IATF 16949 (automotive) certifications. Our engineers not only help you build the board but also help you avoid fatal design flaws that will be exposed in extreme testing during the DFM stage. If you are working on high-reliability electronics, contact us through our website.

What is the most challenging reliability test in your projects? Share your experience in the comments.

Why Your "Working" Prototype Fails in Production: A Manufacturing Perspective

Maggie‌ Wang@AnyPCBA — Thu, 15 Oct 2026 06:45:20 +0000

I work at AnyPCBA, a PCB and PCBA manufacturer. Every week, files arrive from customers who've built a prototype that works perfectly on their bench—and then fails the moment they try to scale it.

The pattern is always the same. The design is electrically correct. The prototype functions. But something breaks between "it works in the lab" and "it works in production."

Here's what I've learned about why this happens—and what engineers can do about it.

The Prototype Bias

Prototypes are built in a controlled environment. You pick the components yourself. You solder by hand or use a small-batch service. You test under ideal conditions.

Production is different. It involves:

Automated pick-and-place machines with placement tolerances
Reflow ovens whose thermal profiles affect every joint
Panelization that introduces mechanical stress during depanelization
Test coverage that must catch defects across thousands of boards

A design that ignores these realities is a design that will fail.

The Five Most Common Gaps

After reviewing thousands of designs, I've seen five gaps appear again and again.

1. Via treatment isn't specified.

A via can be tented (covered with solder mask), plugged (filled with resin or paste), or open. Each has a different process path.

Most designs don't specify. The factory defaults to tented. But if the via sits under a BGA and needs to vent during reflow, tenting causes voids. The board comes back wrong, and you lose a week to rework.

2. Slot plating isn't specified.

Non-circular holes require extra milling time. Whether they're plated determines the process route. A plated slot carries current; a non-plated slot is purely mechanical.

Designs often say "slot" without specifying which. The factory guesses—and it's wrong about half the time.

3. Panelization isn't considered.

The design delivers a single-board Gerber. The factory produces a panel containing multiple boards. Panelization affects depanelization stress, component safety, and production efficiency.

If the design has no panelization notes, the factory plans it during production prep—adding a day or two, and potentially mismatching the design intent.

4. Stack-up is incomplete.

The design says "6-layer board" but doesn't specify dielectric thickness, copper weight, or material. The factory has to guess a reasonable stack-up and wait for confirmation.

This confirmation can take one or two email rounds. If the stack-up needs adjustment, it may affect impedance and require re-simulation.

5. Special processes aren't confirmed upfront.

Heavy copper (2oz+), depth-controlled slots, half-holes, and impedance control all require special processes. If the design is sent without confirming feasibility, the factory discovers the issue during review—and lead time extends. Worse, the factory may not have the capability, forcing a redesign.

Why This Matters More in 2026

AI tools are making PCB design faster. They can auto-route, check design rules, and generate layouts in minutes.

But AI doesn't know your factory. It doesn't know the minimum trace width, the available surface treatments, or which via placement will cause problems during depanelization.

AI generates designs that are "electrically correct." But "electrically correct" is not the same as "manufacturable."

This means the engineer's value is shifting. It's no longer about who can draw boards fastest. It's about who can judge whether a design can actually be built—reliably, at cost, on schedule.

What You Can Do

1. Document your intent.

Don't assume the factory will guess correctly. Write down via treatment, slot plating, panelization preference, stack-up details, and impedance targets.

2. Talk to your manufacturer early.

If you have a doubt about a process, ask before sending the file. A question asked upfront costs less than a redesign downstream.

3. Treat DFM review as a conversation, not a checkpoint.

A good manufacturing partner doesn't just flag issues—they ask about intent. When they ask "do you want this via tented or plugged?", they're not being difficult. They're trying to build what you actually meant.

4. Understand the limits of your design.

Your design isn't finished when it works on the bench. It's finished when it can be reliably and affordably manufactured.

The Takeaway

A working prototype is an achievement. A product that ships reliably, at cost, at scale—is a different achievement entirely.

The gap between those two things is where projects succeed or fail. And closing that gap starts with recognizing that design and manufacturing aren't separate steps. They're the same conversation.

At AnyPCBA, our DFM review is built around this idea. We don't just flag issues—we ask about intent. If you want to see what that looks like for your project, contact us through our website.

What's the biggest manufacturing surprise you've encountered after a "working" prototype? Share your story in the comments.

The "Health Check" of a PCB: From AOI to X-Ray to Functional Test

Maggie‌ Wang@AnyPCBA — Mon, 12 Oct 2026 03:55:52 +0000

At AnyPCBA, every board goes through a "health check" before it leaves the factory.

This isn't a single step—it's a chain of inspections: AOI, X-Ray, ICT, and FCT. Each one answers the same question: is this board safe to ship?

But many engineers have a fuzzy understanding of this chain. They know the factory "tests it," but not exactly what's tested, what can be caught, and what can't.

This article breaks down the inspection chain, layer by layer: what each test does, what it can catch, and what it can't.

Layer 1: AOI—Visible Defects Don't Stand a Chance

AOI (Automated Optical Inspection) is the first layer of the chain.

The principle is simple: high-resolution cameras capture the PCB surface and compare it against a "golden reference" image to find differences. It inspects visible solder joints and components.

What AOI can catch:

Missing or misaligned components: a component that wasn't placed, or placed incorrectly
Solder bridges: two pins that shouldn't be connected, connected by solder
Insufficient solder or cold joints: weak or insufficient solder joints
Tombstoning: a component standing on one end
Polarity errors: diodes, capacitors, or ICs placed in the wrong orientation

AOI is fast and high-coverage, well-suited for high-volume inspection. But its limitation is obvious: it can only see the surface.

For BGA and QFN packages, where the solder joints are hidden beneath the package body, AOI is helpless. The joints are under the chip—the camera can't see them.

Layer 2: X-Ray—Seeing Through the Joints
X-Ray inspection exists to solve AOI's blind spot.

X-Ray uses the penetrating power of X-rays to "see through" BGA bottoms, QFN bottoms, and through-hole interiors. It inspects solder joint quality hidden beneath packages.

What X-Ray can catch:

BGA voiding: bubbles inside solder balls, affecting mechanical strength and thermal conductivity
Solder ball bridging: adjacent balls under a BGA connected together
Missing or shifted solder balls: balls that should be there but aren't, or are misaligned
Poor via filling: incomplete fill inside vias, affecting reliability
QFN bottom joint defects: poor soldering on the thermal pad under the chip

X-Ray's limitation: it can see the "shape" of a solder joint, but it's hard to directly judge "strength" and "electrical performance." A joint that looks normal may actually be a cold joint.

Layer 3: ICT—Checking Electrical Connections

ICT (In-Circuit Test) uses probes to contact test points on the board and measure electrical parameters point by point.

It checks circuit connection integrity and component electrical performance.

What ICT can catch:

Opens: connections that should exist but don't
Shorts: connections that shouldn't exist but do
Component value deviation: actual resistance or capacitance outside spec
Component orientation errors: polarized components placed backwards
Missing components: components that should be installed but aren't

ICT's core metric is "coverage." Ideally, every electrical net should have a test point. But in reality, high-density boards have limited test point space, and coverage often falls short of 100%. Insufficient test point coverage means some defects may slip through.

Layer 4: FCT—Verifying Function

FCT (Functional Circuit Test) is the last layer—and the closest to "real-world use."

It powers up the board, simulates actual operating conditions, and verifies whether the board functions correctly.

What FCT can catch:

Functional abnormalities: a functional module that doesn't work
Performance below spec: excessive output noise, insufficient response speed
Timing issues: power-up sequencing or signal timing that doesn't meet requirements
Boundary condition failures: abnormal operation under extreme temperature or voltage

FCT's key lies in test case design. The more comprehensive the test cases, the more potential problems are caught. If test cases only cover "normal operating conditions," boundary condition issues will be missed.

One Chain, Different Roles, None Optional

Looking at all four inspections together, their roles are clear:

No single inspection is universal. AOI can't see BGA bottoms. X-Ray can't see electrical performance. ICT can't see functional behavior. FCT can't see long-term reliability.

That's why the chain must be complete. Each inspection covers the blind spot of the one before it.

What This Means for Hardware Engineers

1. Test coverage determines recall rate. If you only do AOI, BGA-related defects will slip through. If you only do ICT, functional abnormalities will slip through.

2. Test point design must happen at the design stage. ICT coverage depends on test point accessibility. If test points aren't reachable, even the best ICT equipment is useless.

3. FCT test cases must cover boundary conditions. Testing only "normal operating conditions" isn't enough. Temperature boundaries, voltage boundaries, and timing boundaries are all potential failure points.

4. Understand the boundaries of each inspection. Don't expect any single inspection to "handle everything." The value of the chain lies precisely in each inspection knowing what it can't see.

At AnyPCBA, our inspection chain covers AOI, X-Ray, ICT, and FCT, serving small-to-medium volume PCB and PCBA projects from 5 to 5000 pieces.

Specifically, what we can support includes:

Full AOI inspection: every board goes through automated optical inspection before shipment, checking component placement, solder joint quality, and polarity
X-Ray inspection: for BGA, QFN, and other hidden-joint packages, providing void rate, bridging, and offset detection
ICT testing: based on the customer's test point design, providing in-circuit test fixtures and test programs
FCT functional testing: working with customers to define test cases covering normal operating conditions and boundary conditions
Test point accessibility review: checking test point layout during DFM review, catching coverage gaps early
Test fixture design and fabrication: for small-to-medium volume projects, providing fast, low-cost test fixture solutions

Our manufacturing capabilities cover 2–64 layers, including HDI, rigid-flex, and high-frequency hybrid. We hold ISO 13485 (medical) and IATF 16949 (automotive) certifications. If you're working on a PCB or PCBA project that needs a complete inspection chain, contact us through our website.

What defects have slipped through inspection in your projects? Share your experience in the comments.

When a PCB Is No Longer Just a Circuit Board: How AI Computing Is Reshaping Hardware Engineering

Maggie‌ Wang@AnyPCBA — Thu, 08 Oct 2026 02:12:23 +0000

If you've been working on AI-related hardware projects lately, you may have noticed a signal: PCBs are starting to look less like PCBs.

One change in NVIDIA's next-generation Rubin platform is particularly worth paying attention to: the PCB has evolved from a "connection carrier" into a rack-level core interconnect layer. A 78-layer M9-grade orthogonal backplane directly replaces tens of thousands of copper cables, carrying high-speed signals between GPUs and switches.

What does this mean? For hardware engineers, the design constraints and performance requirements of PCBs are evolving into an entirely new dimension.

Signal Integrity: From "Good Enough" to "Push the Limit"

In the past, designing high-speed digital circuits meant that 50Ω impedance control, differential pair routing, and basic stack-up design could cover most scenarios.

But in AI server environments, signal rates are moving from 112Gbps toward 224Gbps and beyond. At this level, the signal attenuation of traditional FR-4 materials already exceeds 30%, with bit error rates above 5%. In other words, using traditional materials for AI hardware means the signal simply won't get through.

NVIDIA's Rubin platform fully adopts the M9-grade material system, requiring dielectric loss to be pushed below 0.0015. What does that number mean? M8 materials still sit around 0.002 dielectric loss, while M9 uses a combination of quartz cloth and HVLP4/5 copper foil to bring the metric down.

For hardware engineers, this means: when working on AI hardware projects, material selection is no longer a matter of "cost optimization"—it's a matter of "whether it can be built at all."

Thermal Management: PCBs Are Becoming "Thermal Structures"

Another change worth noting is power consumption.

The power consumption of AI server racks has jumped from roughly 130kW in the GB series to around 600kW in the Rubin platform. At this level of power density, the PCB is no longer just a channel for signal transmission—it must simultaneously handle power distribution and thermal management.

In terms of process, this means several changes:

Power layers require thicker copper to carry high currents
High-power areas require embedded heat pipes or high-thermal-conductivity substrates
Board-level power design must consider new architectures such as VPD (Vertical Power Delivery)

The "thermal property" of PCBs is being redefined. It is no longer a passive structural component, but part of the thermal management system.

Impact on Hardware Development Projects

What do these changes mean for R&D teams working on AI hardware, automotive electronics, and medical devices?

First, PCB design complexity is rising. 78-layer orthogonal backplanes, 15–25μm line width/spacing, and the introduction of mSAP processes—precision requirements that once belonged to the "semiconductor level" are becoming the new normal for high-end PCBs.

Second, material selection space is narrowing. High-end materials (M9, quartz cloth, HVLP4 copper foil) have high supply concentration, and both lead times and capacity carry uncertainty. High-end sub-micron spherical silica is monopolized by a few overseas manufacturers, who hold more than 60% of the market share.

Third, supply chain strategies for small-to-medium batch projects need adjustment. The explosion of AI server orders is squeezing high-end PCB capacity. Industry analysis indicates that effective high-end capacity capable of matching AI server orders accounts for less than 20%. This means R&D teams need to pay more attention to a manufacturer's process ceiling and capacity priority when selecting partners—not just price.

A Role Being Rediscovered

Against this backdrop, PCB factories focused on small-to-medium batch, high-mix manufacturing are being rediscovered as a key role in the hardware R&D supply chain.

The reason: the R&D model for AI hardware has changed. Technology routes have not yet converged, multi-scheme parallel validation has become the norm, and demand for prototypes and small batch orders continues to grow. The characteristics of this demand are: high variety, small batch, high process complexity, fast delivery.

This is a different scheduling logic from the large-batch consistency that AI server customers pursue.

For engineers working on AI hardware R&D, supply chain resilience is shifting from a "procurement problem" to a "design problem." Your stack-up design, material selection, and collaboration approach with manufacturers all determine whether your project can close the loop quickly.

AnyPCBA's positioning in this shift is clear: focused on 5–5000 pcs small-to-medium batch PCB & PCBA manufacturing, covering 2–64 layers, including HDI, rigid-flex, and high-frequency hybrid, with ISO 13485 and IATF 16949 certifications. Our engineers check stack-up and material feasibility during DFM review.

If you're working on AI hardware, automotive electronics, or medical device R&D projects, let's talk about supply chain strategy.

What PCB-related challenges have you encountered in AI hardware R&D? Share your experience in the comments.

Why Is PCB Prototyping Getting More Expensive? The Pricing Logic of Small-Batch Manufacturing in 2026

Maggie‌ Wang@AnyPCBA — Wed, 23 Sep 2026 02:44:32 +0000

If you've placed a PCB prototyping order recently, you've probably noticed something unpleasant: the same board costs more than it did a year ago, and lead times are longer too.

This isn't your imagination. In 2026, the pricing logic of PCB prototyping is undergoing a structural shift. From raw material price increases to capacity crowding, from small-batch setup costs to engineering fee adjustments — every link in the chain is pushing up the cost of small-batch manufacturing.

This article breaks down the underlying reasons behind rising PCB prototyping costs and the strategies hardware teams can adopt to respond.

1. Why Is Prototyping Getting More Expensive?

1.1 Upstream Raw Material Price Increases
Copper-clad laminate (CCL) accounts for the largest share of PCB prototyping costs. In 2026, CCL prices have undergone multiple rounds of increases:

Kingboard Laminates has issued its seventh price increase notice this year, with cumulative FR-4 CCL price hikes exceeding 100%
Glass fabric has completed its fifth round of price increases this year, with commonly used specifications rising about 100% from their Q3 2025 low
Copper foil prices are up approximately 30% year-to-date

These raw material increases are being passed directly through to prototyping prices.

1.2 High-End Capacity Crowding
Demand for AI servers and high-speed switches is crowding out capacity that was previously used for prototyping.

When factory production lines are filled with AI orders, prototyping orders naturally drop in priority. More critically, tight supply of high-end materials (such as M7 and M8-grade CCL) reduces the material options available for prototyping.

1.3 Small-Batch Setup Cost Allocation
PCB manufacturing has fixed setup costs — whether you're making 5 boards or 5,000, the equipment calibration and chemical consumption for cutting, drilling, plating, and etching still occur.

The smaller the prototyping order, the higher the setup cost allocated per board. This is the fundamental reason why prototyping prices are naturally higher than volume production.

1.4 Engineering and Testing Fees
Beyond the board cost, prototyping orders also carry:

Engineering fee (NRE) : CAM processing, DFM review, test fixture fabrication
Testing fee: Flying probe testing or fixture testing
Expedite fee: If the lead time is short, expedite fees can exceed the board cost itself

2. Structural Changes in Prototyping Pricing in 2026

3. How Can Hardware Teams Respond?

3.1 Plan Ahead and Avoid Expedite Fees
The most expensive part of prototyping isn't the board cost — it's the expedite fee. If the project timeline allows, try to reserve 2–3 weeks for prototyping to avoid paying high expedite charges.

3.2 Panelize Prototypes to Share Costs
If you have multiple projects that need prototyping, consider *panelization *— combining boards from different projects on a single panel to share setup costs.

Note: Panelization requires the same layer count, laminate, and process requirements.

3.3 Optimize Design to Reduce Special Processes
Special processes (such as blind/buried vias, via-in-pad, heavy copper, and impedance control) significantly increase prototyping costs. Where design allows, use standard processes.

3.4 Choose the Right Manufacturer
Not all manufacturers excel at small-batch prototyping. Choosing a manufacturer that specializes in small-to-medium batches, supports no MOQ, and offers engineering collaboration is often more economical than choosing a large factory.

3.5 Build Long-Term Partnerships
Building long-term relationships with manufacturers can secure more stable pricing and lead times. Long-term customers typically receive priority scheduling and more favorable pricing.

4. Prototyping vs. Small Batch: When to Scale Up?

Many teams struggle with the question of "prototyping or small batch." Here's a simple decision framework:

Key point: Prototyping and small-batch production have different pricing logics. Prototyping prioritizes "speed" and "engineering support," while small batch prioritizes "quality stability" and "cost control."

5. Summary

PCB prototyping is getting more expensive due to a combination of raw material price increases, capacity crowding, and cost structure changes. This isn't a short-term fluctuation — it's part of a structural shift in the industry.

Core Response Strategies:

Plan ahead: Reserve adequate prototyping lead time to avoid expedite fees
Panelize prototypes: Share setup costs across multiple projects
Optimize design: Reduce special processes and manufacturing complexity
Choose the right partner: Select manufacturers that specialize in small-to-medium batches and support engineering collaboration
Build long-term partnerships: Stable relationships bring stable pricing and lead times

If You're Looking for Prototyping or Small-Batch PCB Manufacturing Support
AnyPCBA has over a decade of experience in small-to-medium batch PCB manufacturing, supporting no-MOQ prototypes, fast delivery, and engineering collaboration. Our manufacturing capabilities cover 2–64 layers, including HDI, rigid-flex, and high-frequency hybrid processes. Whether you're a hardware startup or a corporate R&D team, we provide end-to-end support from prototype to production.

👉 If you have PCB prototyping or small-batch manufacturing needs, reach out through our website.

The New Fundamental Skill for Hardware Engineers in 2026: From PCB Design to Supply Chain Management

Maggie‌ Wang@AnyPCBA — Sun, 20 Sep 2026 01:50:54 +0000

If you're a hardware engineer, you've probably noticed that knowing how to draw schematics, do layout, and debug boards is no longer enough.

The 2026 job market is sending a clear signal: hardware engineers who understand the supply chain are commanding a premium.

This isn't a "soft skill" — it's real engineering capability: understanding PCB manufacturing processes, material properties, stackup design, impedance control, DFM rules, and component availability, lead times, and alternatives. In the past, this knowledge belonged to the "procurement side," and the design side could afford to ignore it. Today, the boundary between design and supply chain is disappearing.

Why Supply Chain Knowledge Matters

1. AI Hardware Is Pushing Design to Manufacturing Limits
AI servers, high-speed switches, GPU accelerator cards — these products are pushing PCB and component requirements to the physical limits of manufacturing processes.

A single GPU consumes 700–800W, requiring 2oz or even 3oz heavy copper to carry over 200A of current. 112G PAM4 signals require insertion loss controlled within 0.5dB/inch, with back-drill stubs controlled to under 4 mil. Boards with 20–32 layers have 3–5 times more blind and buried vias than standard servers.

These aren't "design parameters" — they're manufacturing constraints. If your design doesn't account for a factory's actual capabilities — whether they can achieve a 4 mil stub, whether they can consistently control ±5% impedance — even the most perfect simulation can't be manufactured.

2. Component Supply Is Becoming a Design Constraint
The 2026 component market is not calm. Murata has discontinued selected MLCC part numbers, ABF substrate supply is tight, and lead times for some MCUs exceed 40 weeks.

If you don't confirm component availability during the design phase, you may discover "no materials" when you're ready to fabricate — throwing your project schedule into disarray. Worse, some components may already be End-of-Life (EOL) by the time you reach production, forcing a redesign.

Engineers who understand the supply chain consider component availability during the design phase. This isn't a procurement issue — it's part of engineering judgment.

3. DFM Is Becoming a "Front-End" Step in the Design Flow
In the past, DFM reviews were often done after design completion. The design team sent Gerber files to the factory, and the factory checked for obvious issues.

But in 2026, leading hardware teams are moving DFM to the front of the design process. Manufacturing capabilities are discussed with the manufacturer during the design phase, and manufacturing constraints become part of the design rules.

This means engineers need to understand the language of manufacturing — minimum trace width/spacing, minimum via diameter, solder mask dam width, layer-to-layer registration accuracy, back-drill depth tolerance. These aren't "factory issues" — they're "design issues."

4. AI Tools Are Accelerating Design Iteration, But Supply Chain Validation Remains the Bottleneck
In 2026, AI-assisted EDA tools are changing how designs are created. Cadence's AuraStack and Xpeedic's collaboration with Lenovo on EDA Agent can dramatically shorten layout and simulation time.

But AI-generated designs still need to be validated by manufacturing, and supported by the supply chain. AI can cut layout time from three days to half a day — but if components can't be sourced, the time saved will be doubled while waiting for lead times.

Supply Chain Knowledge for Hardware Engineers

How to Build Supply Chain Capability

1. Establish Early Communication with Manufacturers
Don't wait until the design is complete to contact the factory. Involve the manufacturer early in the design process to confirm stackup, material availability, and process constraints.

2. Learn DFM Rules and Make Them a Design Habit
Every manufacturer has their own process capability list. Import these rules into your EDA tool and set them as part of your design rule checks. Make DFM a natural part of the design flow, not an after-the-fact review.

3. Annotate Critical Components in the BOM
Distinguish between "non-substitutable" (P1) and "substitutable" (P2) components. For critical components, prepare at least one alternative and confirm its availability.

4. Monitor Lead Times and Lifecycles
Confirm lead times and lifecycle status of critical components during the design phase. If a component has a long lead time or is approaching EOL, evaluate whether an electrically equivalent alternative exists.

5. Build an Alternative Parts Library
Record validated alternatives for future reference. Don't wait until a shortage occurs to start searching.

Conclusion

In 2026, hardware engineers can't live solely inside EDA software.

Supply chain knowledge is shifting from "nice-to-have" to "must-have." Engineers who understand the supply chain can mitigate manufacturing risks during the design phase, more accurately estimate project timelines and costs, and make smarter decisions amid supply volatility.

This isn't about becoming a supply chain expert — it's about understanding the language of the supply chain: knowing what's possible, what isn't, and what needs to be planned in advance.

In an era where AI is accelerating design iteration, supply chain judgment is becoming one of the scarcest skills for hardware engineers.

If You're Looking for a Manufacturing Partner That Understands the Supply Chain
AnyPCBA's engineering team focuses on stackup design, impedance control, material selection, and manufacturability during DFM reviews. We don't just "build to print" — we help you identify potential issues and optimize your design during the design phase.

Our manufacturing capabilities cover 2–64 layers, including HDI, rigid-flex, and high-frequency hybrid processes. Whether you're an independent developer or a corporate hardware team, we provide engineering support from a manufacturing perspective.

👉 If you have PCB design or manufacturing needs, reach out through our website.

The Real Bottleneck in AI Hardware Isn't the Chip — It's the Substrate

Maggie‌ Wang@AnyPCBA — Wed, 16 Sep 2026 02:12:33 +0000

If you've been following the AI hardware race, you've probably seen plenty of coverage about GPU shortages, HBM capacity, and advanced packaging. But there's a layer beneath all of that which gets far less attention — and it may be the most critical bottleneck of all.

The substrate.

And in most cases, the substrate is a printed circuit board.

Why the Substrate Matters More Than Ever

For decades, the PCB was treated as a passive carrier. You designed your circuit, sent the Gerber files to a fabricator, and got back a board. The PCB didn't determine performance — the chips did.

That assumption is breaking down, fast.

Consider the data:

PCB value per AI server rack has grown to nearly 10× that of a standard server rack.
AI server PCBs now require 20–40+ layers, compared to 8–12 layers for standard servers.
CoWoS packaging and similar 2.5D/3D technologies require substrates with line width and spacing below 15μm — approaching IC substrate precision.
Glass core substrates, the next big thing in advanced packaging, are being developed by Intel and Samsung to replace organic substrates, with a CTE of ~3.2 ppm/°C compared to 12–17 ppm/°C for organics.

The substrate is no longer just connecting chips. It's becoming the performance enabler — and in some cases, the limiting factor.

Three Forces Driving the Substrate Revolution

1. AI Chip Packaging Is Pushing Substrates to Their Limits
AI accelerators are getting bigger. Much bigger. NVIDIA's next-generation chips are pushing package sizes beyond 100mm × 100mm. At these dimensions, organic substrates exhibit warpage that degrades yield.

This is why the industry is moving toward glass core substrates. Glass offers:

CTE closer to silicon (~3.2 ppm/°C vs. 12–17 ppm/°C)
Better dimensional stability at large sizes
Lower dielectric loss for high-speed signals
Higher interconnect density (line/space down to 2–3μm) The challenge? Glass processing is difficult, and the supply chain for glass substrates is still in its early stages.

2. Signal Integrity Requirements Are Reaching New Extremes
112G PAM4 and 224G SerDes are pushing PCB materials to their limits. Standard FR-4 can't support these speeds. The industry has moved to M7, M8, and M9-grade laminates — with dissipation factors dropping from 0.02 to below 0.002.

But materials alone aren't enough. The substrate design — layer stackup, via structures, impedance control — must be optimized for these speeds. This is why back-drilling, via-in-pad, and mSAP processes are becoming standard in high-end substrates.

3. Supply Chain Constraints Are Reshaping the Industry
ABF substrate capacity has been tight for years, and AI demand is making it worse. Major substrate manufacturers are prioritizing AI customers, leaving other segments to compete for remaining capacity.

This is driving investment in new capacity, particularly in Southeast Asia and China. But new capacity takes time — and the gap between demand and supply isn't closing quickly.

What This Means for Hardware Engineers

1. The PCB Is No Longer a Commodity
If you're designing high-performance systems, the PCB can no longer be treated as an afterthought. It's a performance-critical component that must be designed with the same rigor as the chips it connects.

2. Substrate Design Is Becoming a Specialized Skill
Designing a 30-layer AI server board with M9-grade materials, 112G signal integrity requirements, and glass-core substrate technology is not the same as designing a standard 4-layer board. The skills required — high-speed signal integrity, power integrity, thermal management, material science — are increasingly specialized.

3. Supply Chain Knowledge Matters
When substrate materials have long lead times and limited supply, understanding the supply chain becomes part of the design process. Engineers who know which materials are available, which are constrained, and which have alternatives will design better products.

4. Testability and Reliability Are Non-Negotiable
AI servers run 24/7 at full load. Reliability isn't optional. Thermal cycling performance, via reliability, and material stability are all under scrutiny. Designers need to think about long-term reliability from day one.

The Bottom Line

The AI hardware race isn't just about chips. It's about everything that connects them — and the substrate is at the center of that.

For hardware engineers, this is both a challenge and an opportunity. The challenge: keeping up with rapidly evolving substrate technologies. The opportunity: positioning yourself as someone who understands the full stack — from chip to package to board.

The engineers who thrive in the next decade will be those who see the substrate not as a passive carrier, but as the connective tissue that makes advanced systems possible.

If You're Designing High-Performance PCB Projects

AnyPCBA has over a decade of experience in small-to-medium batch PCB manufacturing, supporting 2–64 layers with HDI, rigid-flex, and high-frequency hybrid capabilities. Our engineering team provides DFM/DFA design reviews to help you identify potential issues in stackup, impedance, material selection, and manufacturability.

👉 Contact our engineering team →