Automotive Rain Sensor PCB: How Optical Paths, Potting, and Fabrication Quality Decide Field Performance
Do you think a rain sensor is just a simple circuit board?
Power Distribution Unit PCB Design: Heavy Copper, HDI and High-Speed Supplier Choices for Smart Rack PDUs
- sprintpcb-grupp
A PDU Board Is a Miniature Switchgear Cabinet
I once worked on a smart PDU project early in my career, and the power distribution board in that rack was, in the end, a delicate power circuit board. Many people think it is just a power strip with some chips, but when you take it apart, you see copper foil so thick it is startling and a carefully considered layer stackup, almost a miniature version of a switchgear cabinet. In my view, what decides how long a Power Distribution Unit PCB can run and whether it will burn out is not the metering chip on it but how you talk to the multilayer board supplier about copper thickness and withstand voltage from the very beginning.
PDUs for different scenarios demand completely different treatment of the PCB. One placed in an ordinary office building machine room may get by with a four-layer board and 1 ounce of copper, but in a data center like a GPU cluster, once current density goes up, you have no choice but to find a Heavy copper PCB manufacturer that specializes in heavy copper boards. Once we designed a power distribution board for a liquid-cooled cabinet where a single output branch had to carry over 30 amps, ordinary traces simply could not hold, and the inner layers directly used 4 ounce copper, with copper blocks locally reinforced. This kind of thick copper board is especially grinding when you go back and forth with suppliers on prototypes, since lamination, drilling and plating each easily produce surprises, and heavy copper PCB suppliers who can do this well are in fact only a handful, most willing to take small batches.
Edge Computing Nodes and the Case for HDI
Then there are edge computing nodes, which follow another logic. In this scenario space is compressed severely, and a PDU PCB has to be stuffed into a box narrower than a rack while still keeping intelligent switching and remote monitoring. You cannot solve the problem by adding copper thickness or layers because height is limited. The scheme of an edge site that I led turned to HDI, with laser-drilled stacked vias squeezing out more routing space and separating the control circuit from the power part. The hdi pcb supplier that I cooperated with at the time helped us design a second-order stacked structure, relying on fine lines and microvias to fit the temperature and humidity sensor, relay drivers and Ethernet physical layer all into one small card, saving a lot of room. This experience made me understand that not all high current should go the rough and bulky route: once space shrinks, the capability of the hdi pcb manufacturer becomes the key to whether the board can land.
My choice of multilayer board partners is very picky now, especially when high-voltage isolation and a high-current path are demanded at the same time. You say there are so many multilayer pcb manufacturers, so why are there so few that can do a heavy copper PDU well? The main reason is that ordinary factories are not used to handling the thermal stress and layer registration deviation brought by thick inner copper. After a good Pdu Pcb goes through thermal cycling tests, you should see no board burst or hole copper fracture, but the samples delivered by many factories are simply scrapped at this point. So I first ask the multilayer pcb supplier to produce cross-section reports of power-class thick copper boards they have made before, and only then decide whether to cut material. Do not mind the trouble: it is much better than being called up in the middle of the night in a data center to replace a power distribution board.
Another case that is easily underestimated is high-speed communication and power running together on the same board. Many smart PDUs require support for gigabit or even higher-speed management networks, and then heavy copper alone is not enough: once the impedance control of the differential traces collapses, monitoring data jumps around. I once ran into it: remote switching command delay was unacceptably large, and tracing it to the end, the board factory’s high speed PCB manufacturer process had not kept up and the disturbance the power layer caused to the signal reference layer could not be pressed down. Later I separated the power control module and used another four-layer high-speed board to handle SNMP and BMC communication, and only then was I at ease. In other words, not every scenario needs all functions piled onto one PCB, and sometimes separating is actually more reliable.
So you see, from the data center to the edge node, even though both are power distribution, the PCB approaches are almost never the same. Others discussing scenarios may value communication protocols and metering accuracy, while I care more about the direction of copper and the placement of heat-generating devices: the terminal temperature at connectors, the inductance of the high-current loop, and whether the supplier can give you an impedance test report. These trivial matters do not appear on glossy product brochures, but they truly decide how long a Power Distribution Unit PCB can be lit. I write this not to give some checklist, but to record my own pits so that people who come later do not repeat the detours.
Metering, Heavy Copper and Why Experience Counts
At first I thought a PDU PCB was about the same as an ordinary industrial power board, and only when I got my hands on it did I find this thing is too special. Space is stuck dead, high current and high-precision metering signals both run on a palm-sized board, and noise interference can eat the two digits after the decimal point of accuracy directly. For the earliest prototyping, I found a multilayer pcb manufacturer with a considerable name, but on the boards they made, once temperature rise came up, the metering chip readings began to drift, and checking showed that copper thickness and thermal vias were both a bit short. Later I switched to a heavy copper pcb supplier, pulled copper thickness directly to 6oz, plugged all vias with copper and attached a heat sink, and the temperature drift was controlled. Interestingly, many PDUs now also integrate gigabit Ethernet for remote management, and these high-speed differential signals force you toward hdi pcb. For work below 3 mil line width, an ordinary multilayer pcb supplier simply cannot take it, and impedance control easily flips over. At this time you have to look for a partner who is both familiar with heavy copper and doubles as an hdi pcb supplier, or at least has a reliable hdi pcb manufacturer behind them who can sort out microvias and stacked vias for you, and also needs the experience of a high speed PCB manufacturer to check signal integrity for you, otherwise a prototype cycle drags to a month. My habit now is that whenever I touch a Power Distribution Unit PCB, I first ask the factory whether they have done PDU cases with metering, and those without real experience I pass straight away, to avoid using my own order to let others practice. For metering, if the hardware foundation is not solid, however the software compensates it is hollow. Thinking back, the most underestimated thing in PDU hardware development is the engineering foundation of the PCB itself, and a reliable Heavy copper PCB manufacturer can help move half of the metering calibration forward into the board prototyping stage, digesting it directly.
An AI Cluster PDU: Splitting the Job Across Three Suppliers
I recently had a job allocating PDUs for an AI training cluster, the kind of power distribution unit used in cabinets. Some older models used before could display at most total current, but this batch of PDU PCBs needs to watch real-time power consumption of each outlet and also withstand the transient loads of GPU servers that crash at the drop of a hat. When actually pushing down, I found that the difficulty was not circuit design but finding the right factories to make the boards.
In this AI scenario, cabinet power easily reaches thirty or forty kilowatts, and the current through the PDU main copper bar can reach several hundred amps. The thin copper foil of an ordinary PCB can burn itself into a hand warmer in an instant, so you must find a heavy copper PCB manufacturer, and copper thickness must reach 6 ounces or even higher before it feels solid. But the problem came again: many of these factories dedicated to thick copper are used to making simple double-layer boards and have no interest in high-density signal handling and are not good at it. And the PDU circuit board, besides carrying large current, is also crowded with metering chips, RS-485 communication and even isolated SPI bus routing, with space extremely narrow, and it simply cannot be laid out without HDI. This forces you to look separately: give the power base board to a heavy copper pcb supplier, and the control board and acquisition module to an hdi pcb supplier.
One acquisition board was especially troublesome, needing to run gigabit Ethernet and CAN FD with fairly strict impedance control requirements. At first, to save trouble, we sent the board to a multilayer pcb manufacturer that claimed it could make anything, and the signal integrity of the prototypes that came back was a mess, with the eye diagram barely able to open. Only later did we figure out that although they could make multilayer boards, their high-speed experience was almost zero. We had no choice but to transfer the order for the communication backplane separately to a high speed PCB manufacturer that had long made communication equipment, and only then was the problem solved.
The biggest lesson this project gave me about finding suppliers is: multilayer pcb suppliers are everywhere, but those that can handle thick copper, HDI and high-speed directions at the same time are extremely few. Maybe I was unlucky, but anyway after running around for two or three months, I did not meet a single all-rounder. In the end I bit the bullet and split it into three vendors: the main power board to a heavy copper pcb manufacturer, the outlet metering module to an hdi pcb manufacturer, and the communication daughter card to that high speed PCB manufacturer. Managing three suppliers indeed added some communication cost, but after the PDU came out and was tested for a full two weeks, current sampling accuracy held steady and no communication node ever dropped the line.
I used to always think a PCB is just drawing the picture and throwing it to the factory. Now I understand there are deep process barriers between different factories. Especially after being squeezed by this kind of perverse load as in AI, on a Power Distribution Unit PCB that needs to carry hundreds of amps of large current and also squeeze dense metering and communication links into a board a few centimeters square, any blur in your understanding of the supply chain costs you several rounds of revision time.
High-Density Data Centers: Heat, Signal Acquisition and Layout
I stepped in quite a few pits when I made PDUs before, especially in the PCB part, where things that look simple are all details in actual practice. Many people feel a PDU just distributes current, that a copper bar plus a few sockets will do, but in a high-density data center, with dozens of outputs on one board and each independently monitoring voltage, current and power, without a reliable Power Distribution Unit PCB you simply cannot hold up.
Once current is large, heating is a matter as big as the sky. The most outrageous design I have seen simply calculated the current-carrying cross-section of traces on an ordinary FR4 board and put it into service, and in less than half a year the copper foil began to peel, with temperature rise jumping up and down. Later we gritted our teeth and changed to Heavy copper PCB, starting at 4 ounce copper thickness, with key loops pushed to 6 ounces or even thicker, and temperature control was completely different. A Heavy copper PCB supplier once mentioned that thick copper boards are not simply a matter of thickening copper. The lamination process, etching uniformity and solder mask adhesion all have to be adjusted along with it, and a board from a factory without experience has line-edge burrs as big as saw teeth and gap control is a mess.
This again involves the question of choosing suppliers. There are many multilayer pcb suppliers on the market, but those that can truly achieve high current and high reliability are few. Many multilayer pcb manufacturers mark that they can make 4 layers and 6 layers, but once the inner copper thickness goes up they are out of tricks, and poor control of layer alignment and prepreg filling degree directly affects insulation and local overheating. I tend to look for factories with a long shipping record in power class boards, to see how many large-current PDU projects they have done, rather than looking only at how many layers they boast they can make.
There is another key point: the precision of signal acquisition. A PDU today is not just conducting electricity, and the measurement of every branch current relies on a precision ADC. How the analog ground next to the chip is routed, how the sampling resistor is placed and whether the reference voltage is stable involve layout details even more than the power loop. As for the high-speed acquisition signal lines, if put into an ordinary multilayer board running parallel to high-current traces, the interference is too large to look at. So we started having an HDI pcb manufacturer step in, making the analog acquisition part a local high-density interconnect separately, using buried and blind vias to minimize the loops among the amplifier, filter and chip, and noise went down by several grades. Some HDI pcb suppliers can provide very compact stacking schemes, and the area saved is exactly left for copper bars and heat dissipation channels.
Honestly, for a product like a PDU, chip selection is only half, and the other half depends entirely on board-level implementation. However good a metering chip you pick, such as those supporting six- or eight-channel synchronous sampling, if the PCB impedance does not match and ground bounce is not handled cleanly, the current values read out jump until your heart aches. One of our projects was debugged for three weeks, and in the end we found that high-speed digital interface lines were not routed as differential pairs and coupled in a pile of switching noise, and even the chip’s own filtering could not save it. This part has to rely on a team that has done high speed PCB manufacturer work to review the layout and sort out length matching, impedance and return paths, because even if it is not a signal of tens of GHz, a digital interface at the hundred-MHz level can still cause trouble on a power board.
On the current measurement side, I increasingly feel there is no absolute winner between discrete schemes and integrated chips, it only depends on how you serve the PCB. A design using discrete current sensors has large layout freedom and can place the sensor tight against the output terminal to reduce lead inductance and copper loss error, but board area and solder joint reliability are the cost. An integrated chip saves space, but once the routing goes around far, the error brought by line resistance temperature drift has to be tough-handled by calibration, especially since the copper resistance of the inner layers of a multilayer board does not change linearly with temperature, and the calibration curve has to be split into several segments. My habit now is to have the multilayer pcb manufacturer provide precise copper thickness tolerance data and substrate thermal resistance parameters at the design stage, and for loops I cannot be sure about, I add Kelvin contacts in advance, since hardware correction is easier than algorithmic compensation later.
People always ask me how to evaluate whether a Heavy copper PCB manufacturer is reliable, and I personally look at three points: first, whether they have industrial-grade thermal shock aging process records; second, their control capability over copper surface roughening and lamination cycle; third, whether they can give actual measured values for the withstand voltage parameters of solder mask and legend. A PDU runs in a machine room for five to ten years, and the environment is not harsh to say but is absolutely not ideal, so a board that cannot withstand long-term heating and humidity has a failure rate that rises exponentially. So the lesson of a sample board is perfect, mass production is garbage is enough once you have paid for it.
TMR Current Sensors and Metering Accuracy
In all these years of hardware work, I have increasingly felt that PCB design inside a PDU is a seriously underestimated link. When many people discuss power distribution units they open with circuit breakers, relays and displays, but what truly decides how long a smart PDU can be stable and whether its measurement is accurate is precisely the board that carries the current path and metering circuits. I have seen far too many initial schemes directly using an ordinary FR-4 double-sided board to carry dozens of amps, and once temperature rise came, the IR drop dragged down metering accuracy directly, not to mention long-term reliability.
Data centers now demand power density to a near-crazy degree, with a single rack easily reaching dozens of kilowatts and PDU input current heading for 60A or even higher, and at this point if you still apply the design logic of general-purpose PCBs, you will definitely flip over. My own habit is that as long as current exceeds 30A, I start to seriously consider using a thick copper board. It is not simply thickening the copper foil but requires finding a Heavy copper PCB manufacturer that truly understands, because they know where the boundary of combining thick copper with FR-4 lies, and with small differences in these details such as line width compensation after etching and resin fill during lamination, the thermal uniformity of the whole board is entirely different. Moreover, PDU boards of this type are usually not simple double-sided routing and often need four or even six layers, strictly separating the current-carrying power layer, the low-voltage metering layer and the high-speed signal layer used for communication. So a reliable multilayer pcb supplier is especially important, since they must have the ability to handle 4oz inner copper and 0.5oz outer signal copper on the same board while also controlling layer alignment well, otherwise once the high-voltage isolation spacing shifts, everything afterward is a hidden danger.
For the current sensor part, in the last two years I have clearly leaned toward TMR schemes to replace traditional Hall or current transformers. It is not because of the parameters publicized by manufacturers, but because during actual layout I found that the integration of a TMR sensor makes PDU routing much simpler. It is essentially a magnetic-field-sensitive element and does not need to open an air gap to gather flux like a Hall sensor, and its tolerance to stray magnetic fields is naturally stronger. In the output branches of several three-phase PDUs I used packaged TMR current sensors, and each channel can be independently mounted on the PCB, not occupying a huge space like a transformer and not requiring a copper bar to be cut and a shunt resistor inserted like a shunt. Most critically it naturally achieves electrical isolation, which saved me a lot of trouble when handling the reference ground of the metering chip and the high-voltage ground. Of course TMR is not a panacea. It is sensitive to temperature gradients, and if there is a high-current copper bar continuously heating on the board, you must place the sensor in an isothermal area as much as possible in the PCB layout, or, through the process capability of the multilayer pcb manufacturer, lay thermal copper under the sensor and drill dense vias, forcing the upper and lower layers to equalize temperature.
On metering accuracy, many people fight over the 0.1% reference voltage source on the chip manual, but the actual bottleneck I meet is often the parasitic effects of PCB routing. If you have used HDI you know that those high-density laser blind vias can shorten the signal path to an outrageously short length, and on a PDU motherboard with tight space, using the any-layer interconnect process of an hdi pcb manufacturer, you can pull the differential sampling terminals of the metering chip directly through buried holes to near the Kelvin connection point of the copper bar, making the loop area almost negligible. This largely suppresses common-mode interference brought by switching power ripple and relay bounce. I even, on some extremely compact 48V DC PDUs, to fit metering of 18 outputs in a 1U height, had to use a second-order HDI stackup, and then choosing the right hdi pcb supplier who can control impedance continuity matters more than anything. Those board factories that claim they can do high speed show timidity once they meet a process combining thick copper and HDI, and in the end you still have to find the few that both understand high speed PCB manufacturer technology and can master thick copper lamination parameters.
Speaking of this, an unavoidable reality is that many PDU projects, when choosing PCB suppliers, look only at unit price and ignore the hidden costs brought by whole-board loss and long-term drift. A PDU main board using high-quality multilayer pcb, with copper thickness and stackup carefully optimized, saves in Joule heat loss each year enough to cover that small board material price difference. Not to mention that if the thermal stability of the PCB in the current sensor area is good, the zero drift of the TMR output is much smaller, and the frequency of later calibration can also drop. This is a point I have chewed through after paying quite a bit of tuition over the years.
Financial Data Centers: Hot-Swap Modules and Isolation
In all the years I did power distribution for financial machine rooms, the biggest pit I stepped in was underestimating the process gap of a power distribution unit board (Power Distribution Unit PCB). Many people think a PDU is just a power strip and that any factory can make the PCB inside, which is completely wrong. The power supply requirements of a financial-grade machine room are almost harsh, and especially for a smart PDU with hot-swap modules, the PCB inside must both withstand sustained large current of dozens of amps and squeeze metering, communication and drive into an extremely small space, which an ordinary double-layer board simply cannot hold. Later I would rather spend more cost and insisted on finding a Heavy copper PCB manufacturer to prototype, with copper thickness at least starting at 3oz, because if the trace cross-section is insufficient, long-term live temperature rise will discount the power supply safety of the whole cabinet.
Another overlooked point is the signal layer. High-end PDUs now require every output socket to be independently monitored and remotely switchable, which forces you to run large current and high-density signal lines on the same board. I tried finding a general multilayer pcb supplier to make a four-layer board, and the result was that crosstalk between the communication chip and the transformer sampling circuit was serious and monitoring data jumped around. In the end I turned to an experienced hdi pcb supplier, who used laser blind and buried vias to press the signal layers tighter, and only then did isolation stabilize. Do not look at it as only a few more processes: when the whole module was connected to the back-end system, the bit error rate dropped directly by an order of magnitude. Hot-swap modules are especially sensitive, and the surge and ground bounce at the instant of insertion, if the PCB’s stackup design and ground plane treatment are not in place, burning the communication port is common.
So now, when I make PDU schemes for securities or bank machine rooms, I emphasize just one thing: the board’s own design and process capability decide whether your entire redundant power architecture is real. It is not that using a modular structure and supporting hot swap makes everything fine. If the constitution of the underlying PDU PCB is not good, the best management system is also in vain. When choosing a multilayer pcb manufacturer, I generally first look at whether they have experience with high-frequency and high-current mixed designs, especially cases that have done high-speed signals plus thick copper. A high-speed PCB maker (high speed PCB manufacturer) cannot necessarily handle large current-carrying, and the reverse is also true, with only a few able to blend the two into place. What this industry fears most is applying consumer-electronics thinking to industrial-grade things, and once an accident occurs, the loss is not the money of a few boards.
Hot-Swap Controllers, Breakers and Rack PDU Boards
Anyone who has done data center operations knows in their heart that what gives the most headaches in a cabinet is not a server going down but a PDU suddenly dropping the ball. I have seen too many times the case where, because a control module was broken, the whole cabinet had to be powered off for repair, and the cost of business interruption is simply not comparable to that little hardware money. Later I particularly valued the hot-swap function, which is not just a convenience design but real backstop for operations. A good rack PDU’s controller must be able to be replaced live, otherwise so-called high availability is an empty phrase.
Many people choosing a PDU stare only at power and number of sockets while ignoring the foundation of the PCB. A good Power Distribution Unit PCB must consider current carrying and heat dissipation uniformity from the design stage, and it cannot be fobbed off by finding just any ordinary multilayer pcb manufacturer. I have disassembled the PDUs of several brands, and the boards used inside differ enormously. Some look like four-layer boards, but copper thickness is insufficient, and at the large-current trace connecting the circuit breaker, after a year or half a year of use the board begins to yellow and discolor, a hidden danger that sooner or later causes an accident.
Here I have to talk about circuit breakers. For the breaker on a PDU, my habit is to always choose the hydraulic-magnetic type, and a compact package that does not take up too much space in a 1U or 2U shell. I never trust software overcurrent protection, as it is only a supplement, and when something really goes wrong, millisecond-level response still depends on a hardware breaker. When laying out breakers on a laminate board, the surrounding copper pour must be widened, vias must be drilled more, and the grounding point of the heat sink must be considered separately. I once cooperated with a Heavy copper PCB manufacturer, and the board they made for a PDU had 4 ounce copper thickness, temperature rise in the high-current area was pressed down particularly well, and structural strength was sufficient too, with no warping at the edges even after repeated installation and removal of module latches.
Speaking of this pluggable module, the board used inside is another subject. The controller circuit is highly integrated with dense routing, and an ordinary FR4 board simply cannot route it. The mainstream approach now is to use HDI boards, with laser blind holes plus stepped holes, so that a pile of metering chips and communication components can be stuffed into that palm-sized panel shell. Not every hdi pcb manufacturer can take this kind of order. A board for a PDU control module requires precise impedance control and crosstalk suppression, especially when your communication runs gigabit Ethernet or out-of-band management signals, where even a slight deviation in differential routing makes packet loss severe. I have stepped in pits in my own projects, and later I only looked for hdi pcb suppliers with high-speed board experience, and when prototypes come back I always measure TDR and eye diagrams, rather than finishing after one glance at a flying probe test report.
Going further, in some special scenarios, such as on ships or in outdoor cabinets, the PDU has to withstand salt spray and high temperature. Then the board needs not only heavy copper but also high-Tg substrate and even ceramic-filled board materials. Not every multilayer pcb supplier is willing to fuss with this kind of small-batch, high-difficulty order, but once you find a reliable one, do not change easily. I have worked with a factory that is a high speed PCB manufacturer and also makes heavy copper boards as a sideline, with smooth process switching, and their engineers would remind me that after thick copper board lamination, resin shrinkage is uneven, so the compensation coefficient must be calculated at the engineering design end, otherwise the finished board’s warpage is so large that it cannot fit into the module shell. These details are not written in documents and are purely accumulated by experience.
Back to the PDU itself. Doing hot swap well comes down to the selection of the board-to-board connector and the gold plating thickness at the PCB edge. I have seen some brands where, after the module was plugged in and out a few dozen times, the gold fingers began to show nickel, contact resistance shot up, and when the temperature was high it oxidized directly and later power dropped intermittently, and troubleshooting drove people crazy. So when we make module boards, we specifically tell the multilayer board manufacturer that the board edge finger gold plating must follow 30 microinch hard gold, with wear cycles measured above 500 before release. Spending a little more on this is much more cost-effective than replacing an entire PDU on site later.
There is also a trend now of data centers going underwater, into sealed cabins. The challenge that scenario poses to PDUs is not just heat dissipation and circuit protection but also remote management and moisture-proof sealing. But that is another topic. Once I chatted with a friend who said that the PDU they configured for an underwater cabin used a full potting process, with the board placed in a vacuum cavity and thermally conductive compound poured, and they also had to ensure the mechanical transmission of the breaker handle did not jam, and the PCB design coordination in between involved extreme control of board thickness tolerance and slot accuracy. So a PDU hdi pcb looks unremarkable, but every layer has too many things to take care of. Once you understand these, you will not feel supplier quotes are expensive, and there is a reason expensive is expensive.
Subsea PDU Boards: Corrosion, Pressure and Clear Requirements
You know, for choosing a PCB for a PDU power distribution board for an undersea data center, what gives the biggest headache is not technical parameters but how to make clear to those multilayer board manufacturers your real needs. Many multilayer pcb manufacturers habitually quote with the thinking of ordinary server cabinets, and when you stress to them that this is to be sunk into an undersea cabin, they reply, in that case we just thicken the copper foil, right? I have experienced this kind of conversation too many times.
Thick copper is indeed part of the answer, but it is not that simple at all. The undersea environment is not a single thermal management problem but a compound trouble where corrosion, pressure and electromagnetic interference are mixed together. Some time ago I went back and forth with a heavy copper pcb supplier, who insisted that ordinary FR4 thickened to 4 ounces could handle it, but what you need is something that can still run stably for over ten years in a salt spray environment. The final compromise was to have them do full-board conformal coating, but the uniformity of their coating flipped over on the sample board, with one edge clearly thinner. What does this show? Many manufacturers’ understanding of the word undersea is still stuck at the IP68 waterproof rating level and have no idea that micro condensation and salt penetration can appear inside the cavity, which is harder to deal with than being soaked in water.
Later I went to find a manufacturer that could do both hdi pcb and heavy copper, because a PDU board has both high-current paths and dense signal control pins, and this mixed structure tests the process heavily. The HDI part has to bear the creepage distance constraint of high voltage and also fit enough vias to connect the signal layers, otherwise your communication lines drop packets in that strong electromagnetic field at any moment. I remember very clearly one manufacturer calling itself an hdi pcb supplier whose stackup scheme hollowed out the ground layer under the pads of the isolated transceiver, and when I asked why, they said it was following ordinary digital board experience. But the isolation withstand of an undersea PDU must be considered at DC 1500 volts, and hollowing out the ground is equivalent to cutting the creepage path yourself, and a board like this dropped into the cabin is a time bomb.
So now when I screen a multilayer pcb supplier, the first thing is to see whether they have done projects that have truly gone underwater for over two years, not tank soaking tests but environments that have gone through simulated deep sea pressure cycling. Ask them how they handle the board material’s Z-axis expansion coefficient, how they balance the etching uniformity of large copper areas and dense circuits, and whether they can provide reports in which impedance control and high-voltage testing both meet standard. These three questions can basically screen out half of them. Another thing: if a high speed pcb manufacturer comes and tells you he can swallow the whole PDU board in one bite, you must be careful, because with high-speed digital and power traces of dozens of amps crowded onto the same board, ground bounce and thermal gradient make the board warp unrecognizably. Those who have truly solved this problem will certainly talk to you about structural symmetry and copper balance, rather than just pushing how advanced the laser drilling machines they use are.
There is another easily overlooked pit in undersea PDUs, which is the reference layer design for the remote communication path. Some people like to fill the area around the RS-485 transceiver with copper pour for shielding, and the result is that heat insulation is not handled properly, all the heat is stuffed in the sealed cavity, the transceiver drifts in secret, and you do not even know whom to cry to when data frames drop. My habit now is to first have the heavy copper pcb manufacturer run the current density distribution map of the power section for me, then combine the signal integrity simulation of the hdi pcb manufacturer and cross-verify them, and neither can pass alone. Meeting a supplier willing to tinker with you like this, you can basically place a small batch order.
In the end, a PCB for an undersea PDU is not a single-point technical breakthrough but a whole logic from substrate selection, circuit compensation and coating process to connector coordination. Without finding the right supplier, however much you spend on industrial solid-state capacitors and stainless steel housing, it is useless, because the signal may start rotting from inside the board layers. In this line, there are not many manufacturers who can truly settle down and ponder underwater power reliability, but when you meet one, you know what they produce is not circuit boards but some kind of stubborn confidence.
Outdoor Cabinet PDUs: Three Mistakes I Made
Two years ago, when I replaced smart PDUs in several outdoor cabinets, I truly appreciated how many pits one Power Distribution Unit PCB can hide. That batch of cabinets was stuffed into base station corners, the iron sheet scorched to the touch in summer, with only 1U of space left for the PDU inside while powering a pile of network equipment and edge gateways. At that time I presumptuously found a multilayer pcb manufacturer I had worked with before, thinking it was just stacking relays, a metering chip and Ethernet PHY on a four-layer board. The prototype came back and ran for two weeks after power-up, and the copper foil simply blistered, RS-485 communication kept dropping, and checking showed that ordinary multilayer copper thickness simply could not carry 16A at long-term full load, and the area running high-speed signals did not have a decent ground reference layer. Later I looked everywhere for a factory that could handle both thick copper and signal integrity, which was hard. Many that claimed to be a Heavy copper PCB supplier could in fact only deliver above 6 ounces with tight lead time, and waved their hands when they heard a need to stack an HDI structure. Conversely, the hdi pcb suppliers used to light and thin consumer boards began raising prices when they saw high-current copper bars and dense vias on the drawing. After going around a whole circle, I met an hdi pcb manufacturer with a real power electronics background. They did not rush to look at the volume but first had me fill in a heat dissipation parameter sheet, and then suggested that I press the heavy copper area separately in the middle layer, run the high-speed signals on the surface layer and do impedance simulation, and also changed the pads of the sampling resistor from thermal pads to solid copper connection. The modified board ran in the cabinet for a month, and after a thermal imaging scan, the hottest MOSFET area was only thirty-some degrees above ambient, and those broken things of disconnect-reconnect and cache handling in the communication stack never appeared again. This matter made me reflect for a long time. I used to always think PCB manufacturing was a matter for the supplier after the Gerber was out, but now that PDUs in cabinets are made more and more compact, with high power and high-frequency signals mixed together, you must choose a multilayer pcb supplier from the very beginning with the thinking of heavy power electronics, asking clearly whether he uses plating or lamination to make thick copper, whether the high-speed layer has an independent reference ground, and how batch consistency is. Do not be fooled by glamorous factory videos and quotations: for a 10-layer 2 ounce PDU PCB, if the manufacturer has not done high-voltage insulation tests and thermal cycling aging, once it is racked it is a time bomb.
Heat Paths, Return Paths and a Holistic View of PDU PCBs
In all these years in the trade, I have seen too many people think too simply about PCB design on a power distribution unit. They start by staring at copper thickness, as if stacking up to 3 ounces solves everything. Two years ago there was a project where we used a 4-layer board, and the power part indeed followed the conventional 2 ounce copper, but the whole board’s temperature still could not come down when running. Later, checking and checking, we found that the return path of the ground plane had not been straightened out and all the heat was crowded in the middle layer and could not dissipate. Since then I have particularly accepted one principle: a PDU PCB cannot be judged by whether some local area is thick enough. It must be treated as a whole, balancing the heat path and current return.
Now routing in many smart PDUs is getting denser. On the signal side there are not only several RS485 and I2C metering lines but also Ethernet PHY connections, and if you still force wires on an old-style double-sided board, you are basically burying a mine for yourself. I would rather spend a bit more money up front to find a multilayer board factory that can do HDI, compressing the control part into two or three layers, with much cleaner routing. When choosing an HDI PCB supplier do not just compare price. Some factories claim they can do laser blind vias, but layer alignment deviation is large, and mass production hands you a pile of micro shorts that are hard to even trace. Later I switched to a multilayer PCB manufacturer that does both high-speed boards and heavy copper boards, handling signal layers and power layers separately, and finished product consistency was clearly steadier.
On switching control, I am also not very superstitious about the routine of combining solid-state relays with electromagnetic relays. It sounds beautiful in principle, but when actually deployed in a cabinet, if the switching timing is slightly mismatched, the surge at that instant can send the MOSFET away. A batch of our PDUs simply used high-current magnetic latching relays, with a simple drive circuit and small heating, the only drawback being slow switching speed, but in a data center scenario where actions are almost never frequent, it is quite enough. For relay selection on the power loop, you really have to see whether your application end is normally open, normally closed or cyclic switching, and do not chase concepts from the start.
There are also many easily ignored details in thermal design. I have seen many people line up high-power devices in a row, saying this is neat, and the result is that hot air cannot get out and temperature inside the cabinet soars above 60 degrees directly. Later we staggered the main relays and current sensors slightly and used copper and filled holes on the back of the PCB to pull heat into the entire ground layer, and relying on natural convection we could control it without derating at 55 degrees ambient. In fact, as long as you treat the Power Distribution Unit PCB as a three-dimensional thermal system rather than simply drawing lines on it, many so-called high-temperature problems do not need a fan at all.
So in new projects I now cooperate with, I generally designate a heavy copper PCB supplier for the power board and outsource the signal board to the same factory with HDI capacity, so that production is managed under one system and quality is better aligned. After trying this combination, overall cost did not rise much, but the debugging time saved later is real.
Ethernet, Power Traces and Choosing the Right Mix of Processes
A couple of years ago I helped someone with a smart PDU project in a cabinet, and when the prototype boards came back and were measured, it nearly made me collapse. It was not that functions were wrong. When several boards were put together, the power part was as hot as a radiator, and Ethernet also dropped occasionally. Later I slowly worked out that for a PDU PCB, a board that both carries high current and runs high-speed signals, many faults are not caused by a single factor.
You may not believe it, but the Ethernet communication part really does not need any top-tier chip. Many people start by wanting to use an industrial-grade controller with integrated MAC and PHY and even dual redundant network ports, and the result is that the board routing is a mess and the differential pair impedance falls short by one notch however you adjust. Later I used an ordinary ARM-core MCU with an external PHY, honestly drew lines at 100 ohm differential impedance, chose a better isolation transformer under the RJ45, and it was actually very stable. Factories that make high-speed PCBs all know that the thing this kind of Ethernet part fears most is an unclean reference layer, and if digital ground and power ground are cut randomly, packet loss is sooner or later. The high speed PCB manufacturer I found did help me adjust the stackup twice before the Ethernet link was stabilized.
What really drives people crazy on a PDU board is actually the power traces. You say thick copper, and indeed 2oz to 4oz copper foil is common in multilayer boards now, but the hottest place is often not the copper itself but the section coming out of the terminal pad. I have seen too many layouts where a wide copper pour coming from the transformer, near the MOSFET or wiring terminal, gets cut into seven bends by a few signal vias or fixing screw holes, and when current goes up, flipping through the infrared thermal imager, it is lit white-hot right there. Later I imposed a hard requirement that the via arrays on all power paths must not be judged by count alone but by the spreading angle of the copper pour when changing layers, and I had the heavy copper PCB supplier specially reinforce and thicken the few centimeters of terminal exits for me. Do not be too superstitious about automatic copper pour either: once a local neck appears on a large-current path, no number of vias can dissipate the heat.
There is one more point that is quite interesting: when you want complex control logic on a PDU board, or even want to stuff in a small touch screen, an ordinary multilayer board may not have enough space. At that time you may have to find a multilayer pcb manufacturer to make a high-density board or simply go to HDI. But I found that unless you must make the board palm-sized, finding an hdi pcb supplier who understands your needs, changing vias to blind and buried vias and handling the fine dense routing well, costs will not double, but board reliability goes up a grade. For this kind of Power Distribution Unit PCB, what is feared most is a factory that understands only one process. You clearly need thick copper, high speed and high density at the same time, but some multilayer pcb supplier tells you that thick copper plus HDI cannot be done, and some heavy copper PCB manufacturer is not familiar with high-speed signals, and then you can only spend time screening slowly.
There is no silver bullet in the industrial line, just constant trial and error and finding the right suppliers. Now when I make a PDU board again, I first talk with the factory clearly about whether their production line can handle 3oz copper, 100 ohm differential and blind and buried vias at the same time, and if they can, we talk about the next step.
Two Suppliers, Two Philosophies: Aligning Heavy Copper and High-Speed Thinking
A few years ago I was working on a power distribution unit for an edge computing node, with board size stuck very tight, and the full-power loop and a pile of metering circuits had to be stuffed into a 1U chassis. At that time my confidence in thick copper traces was a bit excessive, and I always felt that as long as the copper foil was wide enough and the vias were piled dense enough, dozens of amps would surely be carried. Facts proved this kind of idea was too much like a student.
Later I directly contacted two factories of different styles, one a multilayer pcb manufacturer used to complex stackups and precision impedance control of high-speed digital boards, and the other a heavy copper pcb supplier specializing in chewing hard bones, who deals daily with copper bars, thick copper and aluminum substrates. Throwing the same design scheme to both, the feedback was completely different. The hdi pcb manufacturer that made high-speed boards adjusted the via layout to be extremely even, and when the prototype came back it was indeed visually pleasing, but at full-load temperature rise, the hot spot at the neck of the solder pad exit was like a soldering iron tip. On the other side, the engineer of the heavy copper pcb manufacturer told me directly: in your scenario current is not carried evenly, and increasing copper thickness alone cannot solve local current crowding, and you need to add a section of copper sheet behind the terminal pad so the current first spreads out smoothly and then enters the inner layer. That sentence convinced me.
After that, I slowly formed a habit: when designing any power distribution unit pcb, I think through the real flow scenario of current in advance, rather than only staring at the colors in simulation software. Take the metering part: many schemes lay out the sampling resistor neatly and wind a large pile of equal-length lines, but in fact when dynamic load switches quickly, if the snapshot position of the metering is not on the true main path of the current, the sampled power fluctuation can differ by several percentage points. Later, when I made a pdu pcb for a data center, I insisted on treating the metering circuit as a living body, whose accuracy does not depend on the plus or minus 1% nominally marked on the chip itself but on how you embed this plus or minus 1% in the real thermal and electromagnetic scenario.
On multilayer boards too, I gradually stopped worshipping the pure layer count. Some designs squeeze sixteen layers, packed full of routing, but if isolation between the power loop and signal loop is off by a bit, ground bounce can make the metering data jump in a mess. I would rather find a multilayer pcb supplier willing to communicate, even if not large in scale, who will use the copper foil of each layer to the extreme, and the homemade methods such as opening windows on upper and lower layers, local thick copper and copper paste via filling are often more useful than stacked vias. So now when looking for an hdi pcb supplier, I do not start by asking how many orders of blind and buried vias they can do but first ask one sentence: have you made power boards continuously above 40 amps? If the other side hems and haws, I basically pass.
Another easily neglected point is the mapping of environment and heat dissipation mechanisms in the circuit board design stage. The same pcb placed in a machine room with smooth airflow and stuffed into a sealed edge cabinet is two different things. I handled a power distribution board accompanying an undersea data cabin, with demands on the heavy copper pcb supplier pushed to the maximum: not only did materials have to be selected according to the salt spray tolerance standard of IPC-6012 Class 3, but even the thickness of the solder mask ink and the design of solder mask bridges had to be confirmed one by one. At that time we even prototyped the flatness after via resin plugging three times just to ensure that the conformal coating could cover completely and that the edge would have no silver migration at all. In this scenario, even an excellent high speed pcb manufacturer, if it has not experienced this kind of industrial environment project, would find it hard to understand from the first moment that what you want is not a lab sample but an iron plate that can run for ten years under alternating torment of salt spray and condensation.
I started thinking deeper about the metering matter because of a lesson from a financial machine room project. That project used a modular hot-swap smart pdu, requiring each outlet to be metered independently with billing-grade accuracy. At first we built it following the manufacturer’s reference design, and steady-state tests had no problems at all, but under load dump and fast phase switching, the metering chip’s data would briefly drop to zero or jump large numbers. After checking around, it was not the chip at all but that when I designed the pcb I had split digital ground and power ground in the multilayer board too rigidly, so that the transient path of the current loop passed through the analog ground pin of the metering chip. Since then, for any power circuit involving metering scenarios, I deliberately bridge the analog ground with a 0 ohm resistor in layout, designing the reference plane with the thinking of actual current, rather than cutting the ground according to the doctrine of digital and analog boundaries.
In summary there is one feeling: finding the right board factory and using the board factory well are equally important.
Judging Suppliers by Real Deployments, Not Brochures
In these years I have dealt with data center infrastructure, having visited no fewer than twenty machine rooms, and I have a deep feeling: when everyone talks about PDUs, their eyes are all fixed on those flashy remote management functions, and what really decides whether this thing can survive three summers of stuffy full-load running is actually the Power Distribution Unit PCB inside. But precisely this board, most people do not even look at it, and during selection they glance over the parameter sheet given by the supplier and call it done.
I paid for it. In a project a few years ago, cabinet density was pulled to 40A per circuit, and in less than half a year the copper foil inside the PDU began to delaminate, the areas the large current passed through heated severely, and it nearly harmed the power busbar of the entire row of cabinets. Upon opening it up, that standard six-layer board used conventional copper thickness, and in a closed cold aisle at 24 hours full load, heat dissipation simply could not keep up and the board material had turned brittle. Later I looked at several multilayer pcb suppliers and found that many people’s understanding of heavy copper still stayed at the stage where 3oz counts as thick, but for a single-phase branch of continuous 60A and above, locally 6oz or even 8oz is needed to be safe. Heavy copper pcb manufacturers that can truly make this thickness stably in volume are not many, and for many factories that label can do, the yield in prototyping and volume production is two different things entirely.
Another thing that made me re-examine this whole set was the control board of the smart PDU. Any infrastructure that looks decent now requires branch metering, outlet-level switching and power-loss alarms, with the main control chip and a pile of sampling circuits squeezed on a palm-sized board and routing density not low. I met a factory that, to save cost, used an ordinary four-layer board to run a high-speed bus, and the result was that metering data jumped inexplicably when load changed suddenly, and later it was found that impedance had not been matched well and signal integrity became a joke. To truly guarantee the stability of this kind of module, you have to find an hdi pcb supplier that has done mixed-signal design, and preferably a team that also understands heavy copper process, because the main loop copper thickness and signal layers are squeezed on the same board, and the thermal expansion and contraction stress difference slowly pulls vias apart, which someone who has never mass-produced simply cannot realize. For the layout on the high-speed communication side, you also need the obsession with length matching and return paths of a high speed pcb manufacturer, otherwise even remotely restarting one outlet can cause communication packet loss and operations explode immediately.
The rule I set for the team now is very simple: do not look at how many certifications a manufacturer lists on its brochure, look only at which real scenarios they have supplied. For example, whether they have experience shipping continuously to 800A cabinet-level PDUs? Whether, on emergency stop signals and strong-current isolation, they can lock safety distance and creepage distance thoroughly from the PCB design, rather than perfunctorily adding insulating sheets later? Every time I follow up with a new supplier, I must go around their production line, look at the actual parameters in the lamination section, and even bring back a cut sample to look at copper thickness uniformity.
More Posts
Automotive Sunroof Control PCB Design: Why Heavy Copper and Hall Signal Integrity Matter More Than HDI or High-Frequency Hype
Sunroof control PCB design is easy to get wrong when teams are
PCB with Components: Why Passive Parts Decide More on a Chassis Management Module PCB Than the Main Chip Ever Does
As a veteran hardware engineer, I’ve found that a lot of people
Din pålitliga leverantör av PCB-tillverkning och PCB-montering
- Expert på produktion av små till medelstora serier
- Kretskortstillverkning med hög precision och automatiserad montering
- Pålitlig partner för OEM/ODM-projekt inom elektronik
Öppettider: (mån-lör) från 9:00 till 18:30