Posted in

How to handle PCB layout for high frequency signals?

Alright, let’s cut to the chase—if you’re deep in the PCB layout game and working with high-frequency signals, you know that one tiny misstep can turn a perfectly good design into a ghost that loses data, glitches out, or just plain refuses to work how you need it. As a PCB design and layout supplier that’s spent the last 12 years troubleshooting and cranking out designs for everything from 5G small cells to industrial sensor nodes, I’ve seen so many designers (even the experienced ones) mess these up because they think high-frequency PCB rules are just “more of the same” as regular boards. Spoiler: they’re not. Let’s break this down like I’m walking you through a recent client fix that saved their launch. PCB Design and Layout

First, let’s get a quick vibe check: high-frequency signals here aren’t your grandma’s 1MHz clock lines. We’re talking 300MHz and up, usually up to like 10GHz, sometimes higher for mmWave stuff. The big thing you gotta remember is that at these frequencies, signals don’t just travel—they act like waves. So reflection, impedance mismatch, crosstalk, and even signal loss from copper roughness don’t just matter; they’re the whole game. I’ve had a client come to me with a 2.4GHz IoT board that was dropping connections every time someone walked near it—turns out their 2.4GHz trace was running parallel to a 12V power line, no ground shield, and the impedance was off by 20 ohms. That’s the kind of small thing that kills products. Let’s start with the foundation: impedance control. This isn’t optional. If your high-speed signal’s impedance doesn’t match the source, the load, and every component in between, you get signal reflection. That’s when part of the signal bounces back like a rubber ball off a wall, messing with the rising/falling edges and making your data look like static. For 2.4GHz, we’re usually targeting 50 ohms for single-ended traces, and 100 ohms for differential pairs (like HDMI or USB 3.x). How do you nail that? Not with a guess. Use your PCB design tool’s impedance calculator—Altium, OrCAD, KiCad even has a basic one now. You need to account for trace width, dielectric thickness, the type of core/prepreg you’re using, and even the copper weight. For a 2oz copper board, a 50-ohm trace is way skinnier than for 1oz, so don’t just use a default trace width. Pro tip from our team: always mark impedance-controlled traces in your layer stackup and add a note in the schematic so the manufacturer doesn’t accidentally change widths during fabrication (I’ve seen that happen too).

Next, layer stackup. This is non-negotiable for high-frequency boards. You can’t just cram all signals and power on one layer anymore. The ideal stackup is symmetric first—this keeps the board from warping after fabrication, which is huge because a warped board can shift trace dimensions and throw off impedance. Then, you want solid ground planes adjacent to signal layers. Signals love a nearby ground—they travel on the path of least impedance, so a ground plane acts as a return path that’s short, which cuts down on both inductance and crosstalk. Let’s use that 2.4GHz IoT client again: their original stackup was top layer (signals), inner 1 (power), inner 2 (ground), bottom layer (power). So their high-frequency signal’s return path had to jump through two inner layers to get back to the source, creating a long return path. We flipped it to top (signal 1), inner 1 (ground 1), inner 2 (signal 2), inner 3 (ground 2), bottom (power)—symmetric, every signal has an adjacent ground, and the power is on the outer layers where it can use wide traces without interfering with high-speed stuff. That alone cut their signal loss by 15% immediately. Another rule: never split ground planes for high-frequency signals. If you split a ground to route a power line, that creates a gap in the return path, forcing the signal to take a longer route around the gap. That’s like driving 10 miles out of your way to pick up a coffee—unnecessary and slow, at high frequencies that adds up to signal distortion.

Now, routing traces—this is where most designers mess up. First, keep high-frequency traces as short as possible. I know, that sounds obvious, but I’ve seen a client route a 5GHz clock trace 2 inches long because they “didn’t have space.” The wavelength of a 5GHz signal is about 0.6 inches in FR-4, so a trace longer than 1/10th the wavelength (0.06 inches) starts causing issues. 2 inches is 30x that—total disaster. If you can’t shorten it, add a series termination resistor at the source or load to absorb reflected signals. Second, route high-frequency traces in straight lines, no 90-degree angles. A 90-degree bend creates a discontinuity in the trace width and impedance, which causes reflection—think of it as a sharp corner on a water pipe, where flow gets disrupted. Instead, use 45-degree angles or gentle curves. Even a 135-degree bend is better than 90. Third, keep differential pairs tightly coupled. If you’re working with differential signals (like PCIe, Ethernet, or high-speed LVDS), the two traces need to be the same length, same width, and spaced evenly apart the whole way. Any mismatch in length (called skew) will mess with the differential signal balance, leading to crosstalk and data errors. For a 10Gbps PCIe link, skew can’t be more than 5 mils—so you have to match them trace for trace, no cutting corners. Also, never route differential pairs parallel to single-ended high-frequency traces, and keep them away from power lines or clock lines by at least 3x their spacing to avoid crosstalk.

Speaking of crosstalk—let’s talk about that. Crosstalk is when one signal bleeds into another through the electric and magnetic fields around the trace. At low frequencies, it’s barely noticeable, but at high frequencies, it can flip a bit in your data line and cause a checksum error. How to fight it? First, spacing: as I just said, separate high-speed traces from other signals by enough distance—for single-ended, usually 2-3x the trace width, more if they’re parallel for a long distance. Second, ground guards: run a ground trace between two high-risk signals, connected to the ground plane at both ends (and every 1/10 wavelength along the trace) to shield them. Third, don’t run parallel high-speed traces for more than 1/10th the wavelength. If you have to, widen the spacing. I once fixed a 5G router that was dropping packets because two adjacent 10GHz traces were parallel for 1.5 inches—we spaced them 5x wider and the issue was gone.

Another big one: component placement. This is half the battle before you even start routing. High-frequency components (like RF transceivers, oscillators, high-speed connectors) need to be placed close together, and close to the edge of the board if they’re connecting to an external antenna or cable. Don’t place a 2.4GHz RF chip on the opposite side of the board from its antenna, because that means a long trace with extra signal loss. Also, separate analog and digital high-frequency sections. Digital circuits have lots of fast switching noise, which can leak into analog RF sections and mess with your signal. Put them on different sides of a ground plane, with a solid ground between them, not a split. And always place the oscillator (clock source) as close to the high-speed chip as possible—its trace is super sensitive to noise, so a short, direct route is non-negotiable. Oh, and when placing capacitors for decoupling (you knew this was coming), place them as close to the chip’s power pins as possible—like within 5 mils. The decoupling cap’s job is to supply instant current to the high-speed chip, and a long trace makes that cap useless. I’ve seen designers place a 0.1uF cap 0.5 inches away from a 5GHz chip—total waste, the noise still gets in.

Wait, let’s not forget about copper roughness and material choice. A lot of new designers use standard FR-4 PCB material because it’s cheap, but for high frequencies above 5GHz, FR-4’s dielectric loss gets really bad. Dielectric loss is when some of the signal energy is absorbed by the board material, turning it into heat. For 10GHz and up, you need a material like Rogers or Taconic, which have much lower dielectric loss. Also, copper roughness: if you use standard rolled copper, it’s smoother than electrodeposited (ED) copper. Rough copper adds to signal loss at high frequencies, so if you’re working with 10GHz+, ask your fabricator for smooth copper. That 2.4GHz client I mentioned earlier was using standard FR-4 and ED copper—swapping to a low-loss material and smooth copper cut their signal loss by 10% more, on top of the stackup fix.

Now, let’s talk about common mistakes I see every week, because even the pros slip up. First, ignoring return paths. I can’t stress this enough: every high-speed signal’s return current follows the path of least inductance, which is under the trace in the ground plane. If that path is broken (like a split in the ground, or a trace that jumps over a gap), the return current has to take a longer route, creating a big loop that acts like an antenna and picks up noise. Second, not doing a DFM (design for manufacturing) check for high-frequency specs. A trace that looks right in your CAD might be impossible to fabricate because it’s too narrow (for 50 ohms on a thin board) or too short, or has a bend that’s too sharp. Third, forgetting about test points. If you’re prototyping, test points for high-speed signals need to be small, short, and not placed right on the trace or in the middle of a differential pair. A big test point on a 5GHz trace is an impedance discontinuity, causing reflection. We usually place test points off to the side, with a small via that doesn’t extend into the trace.

Let me wrap this up with a real quick success story to prove this works. A client came to us last year with a mmWave 28GHz board for a self-driving car sensor—they had 20% packet loss over the 1cm trace between the RF chip and the antenna. We fixed it by: reworking the stackup to have the RF layer adjacent to a solid ground, shortening that trace to 0.5cm, using a Rogers material (low loss at 28GHz), routing the bend at 135 degrees, and placing a ground guard between that trace and a nearby power line. The packet loss dropped to 0.1%, and they launched on time. That’s the kind of result you get when you follow these rules, not guess.

Here’s the thing: high-frequency PCB layout isn’t rocket science, but it is detail-oriented. It’s not about fancy tools or magic tricks—it’s about getting the basics right: impedance control, proper stackup, short traces, no 90s, tight differential pairs, clean ground planes, smart component placement, and using the right materials. Skip even one of these, and you’ll be dealing with issues for weeks, maybe months, while your competitor launches before you.

If you’re working on a high-frequency design and hitting snags—whether it’s signal loss, crosstalk, reflection, or just not getting the impedance right—hit us up. We’ve worked on everything from sub-1GHz industrial sensors to 40GHz mmWave radar, and we can help you nail your layout so your product works the first time, no re-spins, no delays. We’re here to turn your high-frequency design into a reliable, high-performing product, not a headache.

Matter Wireless Module References:

  1. Johnson, H., & Graham, M. (2003). High-Speed Digital Design: A Handbook of Black Magic. Prentice Hall.
  2. Rogers Corporation. (2022). High-Frequency Materials Design Guide.
  3. Saxon, B. (2019). PCB Layout for High-Speed and RF Circuits. Newnes.
  4. IPC-2251. (2012). Design Standard for High-Speed Interconnects.
  5. Microwaves & RF. (2021). “Key Layout Rules for High-Frequency PCB Design.”

Shenzhen STHL Technology Co., Ltd.
Shenzhen STHL Technology Co., Ltd. is one of the leading pcb design and layout manufacturers and suppliers in China, also supports custom service and OEM&ODM service. We warmly welcome you to buy cost-efficient pcb design and layout from our factory. Contact us for more details.
Address: B4, Anle Ind. Park, 172 Hangcheng Blvd, Bao’an, Shenzhen 518126, Guangdong, China
E-mail: info@pcba-china.com
WebSite: https://www.pcba-china.com/