Performance and Functions of Microwave PCBs: Four Budgets That Decide the Board

The performance and functions of microwave PCBs come down to four budgets rather than a list of virtues: insertion loss, phase stability, power handling and thermal path. Each is set by a different property of the same 0.2 mm of dielectric, and each fails in a different way — loss as range, phase as beam pointing, power as a hot spot, heat as a cracked barrel or a drifted Dk. This article works through the four with numbers, then sorts the usual application list into the boards that really carry microwave energy and the ones that are simply called microwave because the product is modern.

Microwave conventionally starts at 300 MHz and runs to 300 GHz, and the useful part of that range for board design begins where a trace is long compared with the guide wavelength. At 1 GHz a 50 Ω line on Dk 3.38 has a guide wavelength of 184 mm, so a 20 mm feed is a ninth of a wavelength and behaves like a small inductor; at 10 GHz the same line is 18.4 mm per wavelength and the same feed is a half-wavelength resonator.

The table explains why the same physical board is a lumped circuit at one frequency and a distributed one at another, and why the transition happened at different times in different products: a base-station filter at 2.6 GHz has 71 mm of wavelength to work with, while a 77 GHz radar has 2.4 mm: a component footprint, a via and a solder fillet all become discontinuities. The definition of an RF or microwave board by its thresholds rather than by its frequency band is in what an RF PCB is.

Every microwave board specification resolves into four budgets. They are independent, they are set by different material properties, and a board that satisfies three of them will still fail on the fourth.

The order of importance changes with the application, and that is the first design decision. A point-to-point link is a loss budget; a phased array is a phase budget; a radar transmitter is a power and thermal budget; a satellite receiver is a phase and outgassing budget. Treating all four as equally weighted is why data sheets end up as lists rather than specifications.

Loss on a microwave board has four terms, and on a low-Df laminate the dielectric is the smallest of them.

The right-hand column is the argument for specifying the rest of the stack: on 5880, 84 % of the loss is copper, roughness, finish and radiation, so a project that pays 6× for PTFE and then specifies standard-profile copper and ENIG has bought a 0.14 dB dielectric and added 0.5 dB of conductor. At 10 GHz the skin depth is 0.65 µm in copper and 1.33 µm in nickel: a 3–6 µm nickel layer under 0.05 µm of gold puts the current in a metal with twice the resistivity of copper, which is measurable at 10 GHz and dominant at 28 GHz.

The same board also changes behaviour across the band, which is worth tabulating once because it decides where a design has to stop. On RO4003C at 0.508 mm, a 50 Ω microstrip loses 0.48 dB per 100 mm at 2.5 GHz, 0.59 dB at 3.5 GHz, 0.85 at 6 GHz, 1.24 at 10 GHz, 2.42 at 24 GHz, 2.72 at 28 GHz and 3.71 at 39 GHz. A 20 dB link budget therefore covers 4.2 m at 2.5 GHz, 1.6 m at 10 GHz and 0.54 m at 39 GHz — the frequency, not the material, decides how long a microwave board can be.

Loss is visible in a link budget and phase is not, which is why phase is the property that surprises projects. It moves with temperature, with Dk tolerance and with thickness, and on an array all three add.

Two things about this table are counter-intuitive enough to be worth stating. The first is that 5880, the lowest-loss laminate in the family, is one of the worst for phase: 13.7° per 100 mm per 100 K at 10 GHz, three times 4003C. A design that chose it for loss and then discovers a phase requirement has chosen wrong, and the replacement is usually 5880LZ at +22 ppm, not a lower-loss material. The second is the sign: a negative TCDk means Dk falls as the board warms, so phase advances rather than lags — useful if the rest of the chain drifts the other way, and a nuisance if it does not.

Phase also moves with the two tolerances that are not thermal: Dk spread across a panel and thickness variation. On a 28 GHz array, ±0.05 of Dk is about ±5° per 100 mm, and ±10 % of dielectric thickness is about ±0.6 % of impedance with a comparable effect on phase velocity. This is why the phase budget is normally written on the drawing as a TDR/VNA measurement with a limit, not as a material property.

Power handling is usually quoted as a dielectric strength in kV/mm, which is the least useful number on the datasheet. In practice a microwave board fails in one of two ways: a voltage breakdown at altitude or at a connector, and a thermal failure where the loss is concentrated.

The thermal failure is the one that can be calculated. Heat is generated in the conductor and in the dielectric, and it leaves through the dielectric to the ground plane, so the dielectric’s conductivity is in the path no matter what the stack looks like.

The via rows explain the design rule: on a PTFE board the dielectric is an obstacle and the via array is the only good path, so a power amplifier is placed over a filled array rather than over a large pad. Where the application needs conduction through the dielectric itself, the choice is a filled laminate such as 6035HTC at 1.44 W/m·K, whose price is a Df of 0.0013 rather than 0.0009 and a TCDk of −66 ppm — the trade is set out in RT/duroid 6035HTC.

Microwave boards are specified at 50 Ω with a window, and on a 28 GHz design the window is usually ±5 % rather than ±10 %, because the return loss at a launch and the ripple in a filter both scale with it. Fitting into ±5 % requires knowing where the error comes from.

There is one more term that does not appear in any calculator and shows up on every measured board: the launch. An SMA connector to a microstrip transition is typically 0.1–0.3 dB and can produce 5 dB of return loss on its own; at 28 GHz the same geometry is worse because the ground-via spacing is now comparable with a fraction of a wavelength. If the measured impedance is worse than the budget predicts, the connector and the launch should be checked before the laminate is blamed. The measurement and acceptance side is in testing in high-frequency PCB production, and the geometry side in impedance control in high-frequency PCB manufacturing.

Mechanical performance on a microwave board matters for one reason: it moves the electrical geometry. Articles describe it as compressive strength and toughness, which is the wrong frame for laminates and the wrong direction for PTFE.

The cold-flow row is the one that catches designers: PTFE is soft, so a clamp, a connector flange or a housing rib changes the dielectric height where it presses, and the impedance changes with it. A 4 % reduction is enough to move a 50 Ω line to 48.8 Ω, and because the material is thermoplastic in practice, the change persists after the load is removed. On a board with a bolted block over the RF section, that is a specification item, not a handling detail.

Microwave PCB articles list applications, and the list is usually a mixture of boards that carry microwave energy and products that are simply electronic and modern. The distinction matters because only the first group needs a microwave laminate.

Two corrections in the table are worth spelling out because they recur. An MRI coil is not a microwave board: at 128 MHz the dielectric loss of even a mid-range laminate is a few thousandths of a decibel per 100 mm, so the material question is repeatability and mechanical stability, not Df — the material reasoning is in RO4003C in medical devices. And wireless charging is a magnetic, kilohertz problem where the PCB’s job is thick copper, not a controlled dielectric.

The medical field does have genuine microwave work: ablation at 915 MHz and 2.45 GHz, where the applicator is a small, loss-critical structure that must be stable and sterilizable. That is a different specification from an imaging coil, and the two should not share a material argument — 5880 in the medical industry covers the microwave end.

Five process steps decide whether a microwave design reaches the board intact. Each has a window, and a fabricator that cannot quote the window is not qualified to quote the board.

The lamination row is where microwave schedules usually break: a hybrid board with a PTFE outer layer and FR-4 core cannot be pressed in one cycle, so it is pressed twice and the second press sees an already-consolidated PTFE layer. That is a planning constraint from the first quote, not a detail discovered in production. The full process map is in high-frequency PCB manufacturing process, and the stack geometry in high-frequency PCB stack-up design.

Four measurements, each answering a different question, and a fifth for anything that has to survive a field life.

Impedance on a coupon from the same panel, per layer, by TDR or VNA — did the stack come out as designed. Insertion loss (S21) on a 50 Ω line coupon across the band, compared with the calculated figure — is the laminate what was ordered, and is the copper what was specified. Phase on a meander or a known electrical length, measured at two temperatures — is the TCDk claim real. Cross-section per panel — are the holes plated to thickness and the dielectric at height. The fifth, thermal cycling with Z-axis strain or a resistance monitor, is what distinguishes a board that passes today from one that survives a hundred excursions.

The discipline that makes those numbers meaningful is that each is measured against a coupon from the same panel, at the same time, with the method named. A VNA sweep of a finished assembly with connectors includes the launches and tells you very little about the laminate.

The original version of this article was a description of virtues. Six of its claims do not survive being checked, and each one is a common way to specify the wrong board.

Nine items cover a microwave board, and each one has a numeric answer that can be checked after the fact.

1. Band and electrical length — the longest run in wavelengths at the top of the band. 2. Loss budget in dB for that run, split into dielectric, copper, finish and launch. 3. Phase budget in degrees over the temperature range, with the Dk tolerance contribution stated. 4. Power level and duty cycle, with the thermal path drawn. 5. Stack with laminate grade, core thickness and bondply, all named. 6. Impedance per layer with tolerance, plus the coupon geometry. 7. Finish, with an explicit statement on nickel. 8. Via policy: back-drill depth, stub allowance, filled or plugged. 9. Acceptance: which coupon, which instrument, which limit.

A specification with those nine can be quoted, built and inspected. One without them becomes a conversation about what “microwave grade” was supposed to mean, usually after the first panel.

The performance and functions of microwave PCBs resolve into four budgets. Loss is set by Df only where the dielectric is the dominant term — 84 % on FR-4 and 16 % on 5880 — which is why low-loss laminates demand low-profile copper and a nickel-free finish to deliver their advantage. Phase is set by TCDk and by Dk and thickness tolerance, and the lowest-loss laminate is not the most stable one: 5880 at −125 ppm/°C drifts 13.7° per 100 mm per 100 K at 10 GHz, against 4.9° for 4003C. Power and heat are the same problem seen twice, and both are solved by copper and filled vias rather than by the dielectric, which on PTFE is a better insulator than FR-4.

The function side is narrower than the application lists suggest: base-station radios, satellite terminals, radar at 24 to 77 GHz, and microwave ablation are microwave boards; MRI coils, ultrasound probes, glucose sensors and wireless charging are not, and specifying a microwave laminate for them buys nothing and costs a multiple.

XCE PCB builds microwave boards from 1 GHz to 77 GHz on Rogers, Taconic and F4B material: 2 to 72 layers, panels to 600×1200 mm, impedance held to ±10 %, samples in as fast as 24 hours, 5,000 m² of plant with 80+ engineers, 10,000 m² monthly output, inspected to IPC-A-600G and IPC-6015. PTFE and mixed stacks are pressed on their own cycles, and the coupon set is measured on the panel it came from. The performance side of RF boards generally is in RF PCB performance and functions.