Microwave Backhaul PCB for E-Band and mmWave Links: Material, Lamination and Manufacturer Selection

The Board Is the Hardest Part of mmWave Backhaul

In projects over the last two years that have pushed into the millimeter-wave backhaul stage, I have fully understood one thing: in this field, the most difficult thing to serve is not the RF specifications at all but the board itself. Think about how many processes a Microwave Backhaul PCB goes through from drawing to physical object that can distort phase and make insertion loss run out of control. Many people talk about channel simulation and link budget at the drop of a hat, but ignore that if the substrate carrying these precise designs flows a few microns of extra glue during lamination, all the earlier work is wasted.

The Label Problem: Many Factories Call Themselves Multilayer Manufacturers

There is a rather bad trend in the industry now, where any factory dares to attach the multilayer pcb manufacturer label to itself, and those that originally started with consumer HDI pull a few production lines and rush in to take microwave backhaul orders. I have no prejudice against hdi pcb suppliers, but millimeter-wave’s sensitivity to copper foil profile and glass cloth weave is simply outside the accumulated experience of this type of factory. Last year, on the core board of an E-band point-to-point microwave link, I first went to a quick-responding hdi pcb manufacturer to save trouble. When the sample came back, the microstrip edges had been bitten up like gnawed by a dog, and the VSWR deviated directly by two orders of magnitude. Later, in communication with their engineers, I found they did not even know to what order of magnitude the copper foil surface roughness should be controlled for millimeter-wave skin depth, and were still processing it as an ordinary digital board. After that batch of boards was entirely scrapped, I gave up: for backhaul work, the supply chain must be divided clearly.

For the selection of a multilayer pcb supplier, I now have a very stubborn set of standards. First, I look at whether he has made real high-frequency mixed lamination, not just saying he can press RO4000 series with FR4, but asking whether he understands the difference in stress release between ceramic-filled substrates and PTFE-type board materials in multilayer structures. Because microwave backhaul equipment hangs outdoors for a long time, with temperature jumping from minus thirty degrees to seventy degrees, and if the interlayer CTE difference is slightly large, via reliability collapses directly. High Frequency PCB manufacturers who can explain these details to me clearly can be counted on two hands in China. Some large factories have very advanced-looking equipment, but their technical team’s understanding of microwave links is still stuck in the traditional microwave concepts below 6GHz of a few years ago, and when told they need to run E-band backhaul from 71GHz to 86GHz, they back off, saying it is a material problem. It is not a material problem. It is that they have never established a closed loop of simulation and measurement feedback under high-frequency conditions.

I even started going around those high Frequency PCB suppliers who like to boast of how many first- and second-order blind and buried vias they have done. A microwave backhaul board does not necessarily reject HDI, but once the blind via structure is stacked with an antenna feed network, the stubs left by back drilling become parasitic resonance points for millimeter-wave signals. Blindly stacking layers to be compatible with dense inner layer routing instead shreds the RF ground return, and I suffered a big loss on this in an emergency communication prototype. At that time, during a rapid deployment of backhaul in a mountainous area, a multilayer board’s ground plane was cut up too finely by via clusters, raising the overall noise floor after downconversion, and in the end I had to temporarily swap in a separate waveguide transition module to barely establish the link. So now I stress to procurement that when looking for a multilayer pcb supplier, do not evaluate by layer count and hole type, but see whether he can provide specific process data for insertion loss and phase consistency control under specific environments, even if it seems troublesome at the test coupon stage.

The reality in this circle now is that many who market themselves as a High Frequency PCB manufacturer are in fact only good at double-sided high-frequency boards, and once a customer requires pressing the digital control layer, power layer and microwave layer into one complete backhaul module, all sorts of problems can no longer be hidden. Instead, the factories with long experience in volume production of downstream antennas or RF modules, although their quotes are high and delivery is firm, can discuss with you real questions such as copper foil type selection and the effect of surface plating on passive intermodulation. However the hardware architecture of millimeter-wave backhaul changes, in the end it has to land on a rigid substrate that works stably. If the manufacturing tolerance of this substrate has already eaten the link margin you painstakingly carved out, then even the best system design is empty talk. So in this matter I am willing to spend more time verifying rather than making do.

Lamination Registration and the Phase Consistency Problem

The deepest feeling I have from these years of microwave backhaul is: do not dive straight into board material parameters and be unable to pull out. Whether the board can finally be used is seventy percent manufacturing and thirty percent design. This sentence may sound a bit extreme, but you will understand after running to PCB factories a few times. Many times, in the office facing simulation software, we tune with excitement, pressing insertion loss to 0.2dB and return loss to minus 25dB, but when the first sample comes back, phase consistency is a mess, and several links are entirely off. After checking and checking, it is not at all a design problem.

That time we were making an E-band backhaul RF front-end board, using high-order HDI with several pairs of blind and buried vias and finely precise line width and spacing. The earlier microwave backhaul link budget was calculated beautifully, but when the sample was measured on the VNA, the phase difference of several receive channels was completely out of control. Finally the board was sent for cross-section analysis, and only then did we find that the multilayer pcb supplier’s lamination registration had shifted by nearly 40 microns, which directly pulled apart the electrical lengths of two groups of differential pairs. You must know that in this frequency band, never mind 40 microns, even a 10 micron offset can be clearly seen in RF performance. What is more headache-inducing is that this factory had made Sub-6GHz RF boards for us before, always performing steadily, and no problem could be seen at all.

This incident completely changed my habit of choosing factories. To make a Microwave Backhaul PCB, you cannot look only at whether the other party is marked as processing high-frequency boards, which is not so simple. Later I simply flew several trips and stationed myself on the production line to watch. I found that a truly capable high Frequency PCB manufacturer for this kind of board has an obsession with lamination process far beyond imagination. For example, they measure thickness separately for every core and every prepreg and strictly control the heating rate during pressing, because slightly uneven resin flow introduces tiny dielectric constant gradients in a high-layer-count board, and for a phased-array feed network, the problem this gradient causes is more fatal than simple loss.

In addition, what peers who transitioned from digital boards easily overlook is that multilayer boards in backhaul equipment are often not just a stack of high-frequency material. They have to mix digital, power, RF and even waveguide transition structures, and then an hdi pcb manufacturer that truly understands hybrid lamination matters more than anything. The hdi pcb supplier I later cooperated with permanently had a dedicated team to interface with RF engineers, getting involved from the stackup arrangement stage, able to tell you clearly which two layers of dielectric using the same batch is safer, and where the back drill residual stub of buried vias will fall near a resonance point of the working band. If this communication is missing, however beautiful the design, it is only a drawing.

By now, when I do projects, the time spent in early screening of multilayer pcb manufacturers is no less than the design itself. Especially for outdoor backhaul equipment, the board must withstand temperature differences and vibration, and also guarantee that solder joints do not crack and impedance does not drift within ten years. At this point every parameter is empty, and real manufacturing data, production line control video, even the model of drill bits that supplier uses, convince me more than simulation curves. RF, in the end, is not designed but made.

Waveguide Transitions, Copper Roughness and Why Process Capability Matters

Some time ago I was tinkering with a millimeter-wave backhaul project, yes, the kind that builds a wireless bridge on E-band. The requirements sounded not complicated, but when the board really ran, I found I had thought of PCBs too simply in the past. Especially with a Microwave Backhaul PCB of this kind, from board material selection to finding the right manufacturer, every step may bury a mine.

Many people think high frequency is nothing more than choosing a better laminate and calculating impedance accurately. My lesson is that impedance matching is only the passing line, and what really torments in the millimeter-wave band is processing consistency and the insertion loss brought by all kinds of transition structures. For example, if you let an ordinary high-frequency PCB manufacturer handle a waveguide-to-microstrip transition, with just a bit of deviation, the return loss doubles on you. This kind of structure is extremely sensitive to copper foil thickness, sidewall roughness and even surface treatment, and it is simply not something an ordinary multilayer board factory can handle. My last version fell exactly here: the simulation ran beautifully but the measured data was a mess, and when cross-sectioned later, copper thickness tolerance and pattern accuracy both deviated from design requirements.

So now when I look for a high frequency PCB manufacturer, the first thing I do is confirm whether they have made products of a similar frequency band and whether they can provide impedance coupons and TDR reports. Do not believe those factories that say they can do it but whose actual experience only goes to below 6GHz. For millimeter-wave backhaul links, every micron of copper roughness as the signal travels along the PCB trace becomes loss, and this accumulated over a 50 centimeter trace can directly collapse the link budget. On this point, those multilayer pcb suppliers who only do general multilayer boards often cannot give any commitment, because they themselves lack quantitative control over this kind of high-frequency characteristic.

HDI, Hybrid Stackups and Thermal Management

There is also an unavoidable problem, which is high density and heat dissipation. Today’s backhaul equipment, to reduce size, would like to stuff everything into a palm-sized module, forcing you to go HDI. At this point, if you still hold the traditional through-hole board mindset and rely on a multilayer pcb manufacturer to stack a dozen-plus layers for you, routing will be a mess, and power integrity and crosstalk will be hard to control. Later I forced the use of any-layer interconnect HDI and specified that the hdi pcb supplier must have a stable production line for laser drilling and copper filling, and only then did I knead the RF front end, digital baseband and power management into one board. After using HDI, ground vias can be drilled directly on pads, the RF ground return became shorter, and incidentally the heat dissipation path was optimized, at least hot spots were not concentrated just in that small area of the power amplifier.

Speaking of heat dissipation, the heat flux density of a millimeter-wave power amplifier is really scary. If you still depend on the old method of thermal vias plus a metal substrate, it may fight with RF performance. My idea now is to let the PCB itself take part in thermal management, for example customizing a hybrid structure at a high frequency PCB supplier that uses high-frequency material only on the key RF layers and uses FR-4-type material with higher thermal conductivity on the other layers, with cost controllable and the thermal path smoother. This approach requires the factory to have precise registration capability and lamination process, and an ordinary multilayer board factory cannot take it, so you can only go chew on factories that truly do mixed-lamination high-frequency boards.

Finding the right hdi pcb manufacturer and high-frequency PCB supplier is in essence finding whether their process boundary and frequency band experience can cover your design. Do not listen only to sales brag about yield, and directly have them take out the cross-sections and test data of millimeter-wave boards they have made. A manufacturer I cooperated with recently gave me measured E-band antenna board data they had done for a certain satellite communication equipment, and only because of that did I dare to hand over the power-splitting network and beamforming network that are the most critical in the backhaul link. The feeling was that he finally was not treating me as an ordinary customer who only cares about layer count and copper thickness but was seriously taking the demands of millimeter-wave signals on materials. The final measured insertion loss and our simulation deviation was controlled within 0.2dB, which I would never have dared to think of before.

After saying so much, the core is really one point: for making millimeter-wave backhaul PCBs, you must treat the manufacturing end as part of the design, not as a contract manufacturing link found only at the end. Sitting down early with those multilayer pcb manufacturers or even HDI pcb suppliers who truly understand high frequency to talk through process limits is more useful than simulating ten versions by yourself. Because a PCB is not finished by drawing a picture. It is the thing in the physical world that decides whether your signal lives or dies.

Waveguide Transitions and Registration Tolerances in 71-76GHz Boards

In many projects, by the time they reach microwave backhaul, the trouble is usually not in the schematic or simulation but in how to find a manufacturer who really knows the field. In recent years, Multilayer PCB cases in E-band and higher bands have followed one after another, and players are more and more crowded, but qualified Multilayer PCB manufacturers have not increased by many. Especially when the transition between waveguide and PCB is involved, however beautiful the design idea, once it reaches the board factory and the bottom is exposed, it all changes flavor. I handled a 71 to 76GHz backhaul prototype, in which the transition structure loss was pressed to the extreme in simulation, but the prototype measured about 1dB more than expected. Tracing it, it was not a topology problem but the board factory taking the reference layer thickness tolerance too wide, and the backshort position was off by a full 60 microns.

This kind of precision is something an ordinary multilayer board supplier simply cannot support. The so-called high-frequency experience of many Multilayer PCB suppliers on the market is only having made a few antenna boards or power splitter boards, without ever touching chip-carrier-level microwave backhaul links. They cannot understand that waveguide sidewall via arrays are not only vias but also the boundary of current, and if the ground vias are slightly sparse, the mode leaks and the return loss curve of the whole transition region breaks apart directly. Once I probed a High Frequency PCB manufacturer indirectly, and their technical director asked me back: are vias not just connections? Would it not be fine to drill a few more? At that moment I knew the board could not be handed to them.

Later I turned toward HDI to look, and several HDI pcb suppliers were indeed used to precise alignment and laser blind vias, but another layer of trouble came out. The waveguide transition section needs mixed dielectric layers with copper surface treatment, and what HDI manufacturers are good at is thin boards, small holes and fine lines, but when it comes to lamination uniformity of RF material, such as mixing Rogers 3003 with FR-4, their layer registration control is not as good as factories that specialize in high-frequency mixed lamination. I went through a half-year running-in period with a fairly large HDI pcb manufacturer and modified the stackup four times before the impedance of the transition section measured within the simulation window. The apprentice cost in that period was much higher than the design cost.

So now when I evaluate suppliers for a Microwave Backhaul PCB, I do not look at how many layers they can do or what the minimum line width is at all, but flip through whether they have really mass-produced whole boards with waveguide ports in the past. The sticking point of this process is that it must be able to play with material anisotropy and copper foil roughness like a High Frequency PCB supplier, while also having the interlayer alignment capability and back drill accuracy of HDI. Missing one, and the transition structure becomes a marvelous simulation story. When the board comes back and is tested, the waveguide plugged in will not cover for you.