On this page
Hardware Interfaces & PCB/Flex Design
Part I — Interface map
1. What you should know at a glance
| Interface | Physical style | Main use | Hardware topics to know |
|---|---|---|---|
| PCIe | high-speed differential serial lanes | CPU/SoC ↔ GPU, SSD, NIC, accelerator | lanes, LTSSM, equalization, refclk, PERST#, channel loss, retimers |
| USB | USB2 differential + SuperSpeed pairs; often USB-C | peripherals, storage, charging, tunneling | USB2 vs USB3/USB4, Type-C, CC, PD, orientation mux, ESD, retimers |
| Thunderbolt | high-speed USB-C cable interface | PCIe/DisplayPort/USB tunneling | controllers, USB4 relation, bandwidth, security, PD, cable/retimer design |
| I²C | 2-wire open-drain bus | sensors, PMICs, EEPROM, control | pull-ups, capacitance, addressing, ACK/NACK, arbitration, clock stretching |
| SPI | synchronous single-ended point-to-point/multi-target | flash, ADC/DAC, RF ICs, displays | CPOL/CPHA, chip-select, timing, edge rate, fanout, termination |
| UART | asynchronous TX/RX | debug, GPS, modules, MCU links | baud rate, framing, level standards, clock mismatch |
| HDMI | high-speed differential video/audio link | display interface | TMDS/FRL, DDC, HPD, eARC, ESD, connector/launch quality |
Part II — PCI Express
2. PCIe architecture
A PCIe link is built from one or more bidirectional serial lanes.
Common widths are
\[x1,quad x2,quad x4,quad x8,quad x16.\]Each lane contains a transmit and receive differential pair.
Conceptually,
\[\boxed{ \text{Transaction Layer} \rightarrow \text{Data Link Layer} \rightarrow \text{Physical Layer} \rightarrow \text{PCB / connector channel} }\]The PHY handles serialization, equalization, clock recovery and lane behavior; the higher layers handle packets, flow control and configuration.
3. PCIe generations
| Generation | Raw rate per lane | Signaling |
|---|---|---|
| PCIe 3.0 | 8 GT/s | NRZ |
| PCIe 4.0 | 16 GT/s | NRZ |
| PCIe 5.0 | 32 GT/s | NRZ |
| PCIe 6.x | 64 GT/s | PAM4 + FLIT/FEC |
| PCIe 7.x | 128 GT/s | PAM4 |
PCI-SIG lists PCI Express Base Specification Revision 7.1 as the current approved base specification as of September 2026.
Official reference: PCI-SIG PCI Express Specifications.
4. Signals you should recognize
Beyond the differential data lanes, a platform may include
- REFCLK;
- PERST#;
- CLKREQ#;
- WAKE#;
- sideband management;
- SMBus/I²C-related management depending on form factor.
For M.2 or add-in cards, the connector specification adds mechanical and power constraints.
5. Link training
PCIe does not simply begin sending application data after power-up.
The LTSSM moves through states such as
- Detect;
- Polling;
- Configuration;
- Recovery;
- L0;
- low-power states.
A board that enumerates at Gen3 but not Gen5 may have a channel/equalization problem rather than a software problem.
6. PCIe equalization
At high speed, the channel attenuates high-frequency components.
The link uses combinations of
- transmitter de-emphasis / FFE;
- receiver CTLE;
- DFE;
- link-training coefficient adaptation.
For PCIe 6.0 and later, PAM4 makes vertical noise margin smaller and FEC/FLIT operation becomes part of the architecture.
PCI-SIG’s PCIe 6.0 overview documents 64 GT/s, PAM4, FLIT mode and lightweight FEC. PCIe 6.0 overview.
7. PCIe PCB layout
Key rules:
- maintain controlled differential impedance;
- keep pair skew small;
- minimize via count;
- avoid long via stubs;
- use backdrilling where needed;
- control connector/BGA breakout discontinuities;
- maintain continuous return paths;
- keep aggressive signals away from the pair;
- budget package + PCB + connector + cable loss together.
The exact target impedance/tolerance comes from the applicable PCIe/CEM/platform specification and stackup.
8. PCIe debug
Useful tools include
- high-bandwidth oscilloscope;
- BERT;
- TDR/TDT;
- VNA;
- PCIe protocol analyzer;
- link-training/status registers.
Useful questions:
- Does the device enumerate?
- At what negotiated generation and width?
- Is the link down-training?
- Does Recovery occur repeatedly?
- Are errors lane-specific?
- Is the problem channel loss, REFCLK/jitter, power, reset, firmware, or protocol?
Part III — USB and USB Type-C
9. USB is not the same as USB-C
USB-C is a connector system.
A USB-C port can carry different combinations of
- USB 2.0;
- USB 3.x;
- USB4;
- DisplayPort Alt Mode;
- Thunderbolt;
- USB Power Delivery.
Do not infer speed merely from connector shape.
10. USB speed families
| Technology | Nominal data rate |
|---|---|
| USB 2.0 High-Speed | 480 Mb/s |
| USB 3.2 Gen 1 | 5 Gb/s |
| USB 3.2 Gen 2 | 10 Gb/s |
| USB 3.2 Gen 2x2 | 20 Gb/s |
| USB4 | up to 40 Gb/s class |
| USB4 Version 2.0 | 80 Gb/s class with higher asymmetric capability |
The current USB4 Version 2.0 specification was published by USB-IF in April 2026.
Official: USB4 Specification v2.0.
11. USB-C signals
Important USB-C pins/signals include
- D+/D− for USB 2.0;
- high-speed TX/RX differential pairs;
- CC1 / CC2;
- SBU1 / SBU2;
- VBUS;
- GND.
The CC pins establish
- cable orientation;
- source/sink role;
- current advertisement;
- Power Delivery communication context.
Current official reference: USB Type-C Cable and Connector Specification Release 2.5.
12. USB Power Delivery
USB PD negotiates voltage/current over the Type-C ecosystem.
Board-level considerations include
- PD controller;
- VBUS FETs;
- current sense;
- over-voltage/current protection;
- thermal behavior;
- ESD;
- connector current rating.
High-speed data and high-power delivery coexist physically, so SI, PI and thermal design interact.
13. USB PCB layout
For SuperSpeed/USB4 paths:
- treat the connector launch as part of the channel;
- maintain differential impedance;
- minimize discontinuities;
- route pairs over continuous references;
- minimize stubs;
- place ESD components so their parasitic capacitance does not destroy the channel;
- use muxes/retimers/redrivers only with a complete loss budget;
- consider cable + connector + board together.
USB2 D+/D− routing is less demanding than USB4 but still benefits from controlled geometry and clean return paths.
Part IV — Thunderbolt
14. What Thunderbolt is
Thunderbolt uses the USB-C connector ecosystem and transports high-performance data/display traffic.
Thunderbolt 5 supports USB4 compatibility and can carry
- PCIe traffic;
- DisplayPort;
- USB;
- power through USB-C infrastructure.
Intel specifies Thunderbolt 5 links at up to 80 Gb/s bidirectional, with bandwidth boost modes reaching up to 120 Gb/s in one direction for display-intensive use cases.
Official overview: Intel Thunderbolt Technology.
15. What matters in a Thunderbolt design
Know the system around the controller:
- USB-C connector;
- Thunderbolt/USB4 controller;
- retimers;
- high-speed muxes;
- DisplayPort inputs;
- PCIe connection to the host;
- PD controller;
- cable identification;
- security / DMA protection;
- certification.
Thunderbolt is therefore a system integration problem, not merely four high-speed traces.
Part V — I²C
16. Electrical principle
I²C uses two lines:
\[\boxed{ SDA\quad \text{and}\quad SCL }\]Standard I²C uses open-drain/open-collector signaling.
Devices pull the line LOW; external resistors pull it HIGH.
This allows wired-AND behavior and arbitration.
17. I²C speed modes
NXP’s UM10204 defines:
| Mode | Data rate |
|---|---|
| Standard-mode | 100 kbit/s |
| Fast-mode | 400 kbit/s |
| Fast-mode Plus | 1 Mbit/s |
| High-speed mode | 3.4 Mbit/s |
| Ultra Fast-mode | 5 Mbit/s, unidirectional |
Official reference: NXP UM10204 — I²C-bus specification.
18. Pull-up resistor tradeoff
The bus rise time is approximately related to
\[\tau\sim R_P C_B.\]A pull-up that is too large gives slow edges.
A pull-up that is too small requires excessive sink current.
Therefore choose $R_P$ from
- bus capacitance;
- required rise time;
- voltage;
- device sink capability.
19. I²C protocol concepts
Know
- START / STOP;
- 7-bit and 10-bit addressing;
- read/write direction;
- ACK/NACK;
- repeated START;
- arbitration;
- clock stretching;
- multi-controller behavior.
Common problems are incorrect address, missing pull-ups, wrong voltage level, excessive bus capacitance and a target holding SDA low.
Part VI — SPI
20. SPI signals
A typical SPI bus contains
- SCLK;
- MOSI / COPI;
- MISO / CIPO;
- CS#.
SPI is normally push-pull and synchronous.
Unlike I²C, there is no universal central SPI standard defining one maximum speed or one connector.
Always read the device datasheet.
21. CPOL and CPHA
The four common SPI modes are defined from
- clock polarity, CPOL;
- sampling phase, CPHA.
A controller and target using different modes can produce completely plausible-looking clocks but corrupted data.
22. SPI PCB issues
At low speed SPI behaves like ordinary digital logic.
At fast edge rates it becomes a transmission-line problem.
Watch
- clock ringing;
- long stubs;
- fanout;
- poor return paths;
- overshoot/undershoot;
- source termination;
- chip-select timing.
The edge rate, not merely the clock frequency, tells you when SI becomes important.
Part VII — UART
23. UART basics
UART is asynchronous.
A typical frame contains
\[\boxed{ \text{start} + \text{data} + \text{optional parity} + \text{stop}. }\]Both ends must agree on baud rate and framing.
24. UART is not RS-232
UART describes the logical serial framing/peripheral.
Electrical standards such as
- logic-level CMOS/TTL UART;
- RS-232;
- RS-485;
use different voltage levels and drivers.
Do not connect an MCU UART pin directly to an RS-232 voltage-level interface without a transceiver.
25. UART debug
Use
- oscilloscope;
- logic analyzer;
- USB-UART adapter.
Check
- idle level;
- voltage compatibility;
- baud;
- start/stop configuration;
- crossed TX/RX;
- ground reference.
Part VIII — HDMI
26. HDMI architecture
HDMI transports uncompressed/digitally encoded video and audio plus control/management channels.
Modern HDMI includes
- high-speed video lanes;
- DDC;
- HPD;
- CEC;
- eARC in appropriate products.
Older HDMI generations use TMDS; newer high-rate modes use Fixed Rate Link (FRL).
27. HDMI 2.2
HDMI Forum released HDMI 2.2 in June 2025.
Its headline maximum bandwidth is 96 Gb/s, with higher-resolution/high-refresh display modes and next-generation FRL.
Official overview: HDMI 2.2 Specification.
28. HDMI board layout
Important concerns include
- differential impedance;
- intra-pair skew;
- lane-to-lane timing;
- connector launch;
- ESD-device capacitance;
- common-mode conversion;
- return-path continuity;
- DDC/HPD pull-ups and level requirements.
For high-rate HDMI, connector region layout and ESD selection can determine whether a nominally correct schematic passes compliance.
Part IX — High-speed PCB design
29. Start with stackup
Do not route first and “calculate impedance later.”
The stackup determines
- trace width;
- spacing;
- reference-plane distance;
- propagation velocity;
- loss;
- via geometry;
- manufacturability.
A strong workflow is
\[\boxed{ \text{fabricator stackup} \rightarrow \text{impedance targets} \rightarrow \text{routing rules} \rightarrow \text{SI verification}. }\]30. Controlled impedance
For high-speed links, trace geometry and dielectric construction are chosen for a target impedance.
Typical design types:
- single-ended microstrip/stripline;
- edge-coupled differential pair;
- broadside-coupled pair.
Do not copy trace width from another PCB: impedance depends on the actual stackup.
31. Return current
High-frequency return current flows near the signal path on its reference plane.
A split/gap under a high-speed trace forces the return current to detour, increasing loop inductance and radiation.
Therefore:
Route high-speed signals over continuous reference planes.
When changing reference planes, provide an intentional return-current transition.
32. Differential pairs
Important quantities include
- differential impedance;
- common-mode impedance;
- intra-pair skew;
- pair spacing;
- coupling;
- asymmetry.
Avoid obsessing over geometrical symmetry while ignoring the complete electromagnetic channel.
A differential pair still needs a good reference plane.
33. Length versus delay matching
Equal physical length does not always mean equal electrical delay.
Delay depends on
\[t_d=\frac{l}{v_p}.\]If traces use different layers with different effective permittivity, time-domain matching may be more meaningful than simple length matching.
Current KiCad documentation explicitly supports both length- and delay-domain tuning for high-speed nets. KiCad PCB Editor.
34. Vias and stubs
A via is a 3-D discontinuity containing
- capacitance;
- inductance;
- unused barrel stub;
- antipads;
- coupling to nearby vias.
At high data rates, unused stubs can create resonances.
Mitigations include
- fewer layer transitions;
- blind/buried vias;
- microvias;
- backdrilling;
- optimized antipads.
35. Connector and BGA breakout
The hardest part of a channel is often not the long uniform trace.
Pay special attention to
- BGA escape;
- neck-down;
- via fields;
- connector pads;
- connector transition;
- ESD footprint;
- AC coupling capacitor footprints.
These localized discontinuities can dominate return loss.
36. Crosstalk
Coupling increases with
- long parallel routing;
- small spacing;
- weak reference-plane coupling;
- large edge rate.
Mitigate through
- spacing;
- orthogonal routing on adjacent signal layers;
- solid references;
- layer assignment;
- shorter parallel run length.
37. Via stubs and backdrilling
Backdrilling removes unused plated-through-hole barrel beneath the active signal transition.
It is common in
- PCIe;
- high-speed Ethernet;
- SerDes;
- high-rate storage/server boards.
It should be designed together with PCB fabrication capability and residual-stub limits.
Part X — Power integrity
38. Decoupling
A digital IC draws transient current.
The PDN must provide low impedance over the relevant frequency range.
A basic target is
\[\boxed{ Z_{target}=\frac{\Delta V_{allow}}{\Delta I}. }\]Use
- plane capacitance;
- local MLCCs;
- bulk capacitors;
- appropriate VRM control.
Do not select decoupling only by nominal capacitance; ESL/ESR and placement matter.
39. Power rails for high-speed interfaces
High-speed PHYs often have separate
- analog;
- PLL;
- core;
- I/O;
- termination
supplies.
Noise coupling into PLL/CDR/SerDes rails can become jitter.
The PCB layout must therefore connect SI and PI thinking.
Part XI — Flex and rigid-flex
40. Why use flex
Flex is used when the product needs
- folding;
- compact 3-D packaging;
- reduced connector count;
- motion;
- weight reduction;
- reliable interconnection between rigid regions.
A rigid-flex design combines rigid PCB regions and flexible interconnect in one fabricated structure.
41. Static versus dynamic flex
Two very different cases:
Flex-to-install
Bent during assembly and then mostly stationary.
Dynamic flex
Repeatedly moves during product life.
Dynamic flex requires substantially more conservative mechanical design.
IPC-6013 distinguishes installation-use and continuous-flex usage classes.
42. Flex layout rules
Good practice includes
- avoid vias in bend regions;
- avoid component pads in bend zones;
- route conductors roughly perpendicular to the bend axis;
- avoid abrupt width changes;
- use smooth transitions;
- use teardrops where appropriate;
- stagger traces in multilayer flex where the stackup requires it;
- coordinate coverlay openings and stiffeners;
- respect fabricator minimum bend radius.
Never choose a universal bend-radius rule without the material stackup and fabrication process.
43. Flex copper and materials
Flex constructions commonly use polyimide and may use rolled-annealed copper for improved repeated-flex behavior.
Important mechanical/electrical parameters include
- copper thickness;
- dielectric thickness;
- adhesive;
- coverlay;
- stiffener;
- bend radius;
- grain direction;
- impedance through the flex region.
A high-speed rigid-flex link must preserve both mechanical reliability and electrical impedance.
44. Flex standards
Two standards worth recognizing are
- IPC-2223 — design standard for flexible/rigid-flex printed boards;
- IPC-6013 — qualification and performance requirements for flex/rigid-flex boards.
IPC explicitly identifies IPC-2223 for flexible/rigid-flex design and IPC-6013 for qualification/performance.
Official references:
Always check the revision required by your organization or fabricator.
Part XII — PCB design workflow
45. Schematic capture
A strong schematic includes
- clear power hierarchy;
- interface grouping;
- readable net names;
- decoupling intent;
- termination values;
- test points;
- pull-ups/pull-downs;
- reset/boot straps;
- connector pin mapping.
The schematic should communicate design intent, not merely connectivity.
46. Library quality
For each component verify
- schematic symbol;
- footprint;
- pin mapping;
- courtyard;
- assembly outline;
- solder-mask/paste geometry;
- 3-D model;
- manufacturer part number.
One bad footprint can invalidate an otherwise perfect circuit.
47. Constraint management
Create constraints before detailed routing.
Examples:
- impedance classes;
- differential pairs;
- max via count;
- length/delay matching;
- spacing;
- high-current width;
- sensitive analog keepouts;
- flex bend zones;
- component height limits.
Professional PCB design is heavily constraint-driven.
48. Placement
Placement determines most of the eventual routing quality.
Prioritize
- connectors;
- processors/FPGAs/SoCs;
- memories;
- SerDes/PHY devices;
- regulators;
- clocks;
- decoupling;
- remaining support circuitry.
Place the topology before drawing traces.
49. Routing order
A useful priority is
- clocks / reference;
- highest-speed serial interfaces;
- DDR/source-synchronous memory;
- sensitive analog/RF;
- power;
- lower-speed digital/control.
The exact order depends on system architecture, but not all nets are equal.
50. Fabrication outputs
Know how to generate/review
- Gerber or ODB++ / IPC-2581;
- NC drill;
- stackup;
- fabrication drawing;
- assembly drawing;
- pick-and-place;
- BOM;
- netlist;
- impedance coupons/requirements;
- flex bend/stiffener/coverlay documentation.
A PCB is not finished when routing ends; it is finished when fabrication and assembly intent is unambiguous.
Part XIII — CAD tools
51. Altium Designer
Common in product-development teams.
Know
- schematic;
- PCB editor;
- constraint manager;
- layer-stack manager;
- differential routing;
- length tuning;
- 3-D/mechanical view;
- rigid-flex regions;
- outputs/OutJob.
Altium’s current rigid-flex documentation covers substacks, board regions and bend definitions. Altium rigid-flex design.
52. Cadence OrCAD / Allegro
Widely used for complex/high-speed commercial electronics.
Important concepts include
- Capture/OrCAD schematic flow;
- Allegro PCB Editor;
- Constraint Manager;
- differential-pair rules;
- timing/length constraints;
- high-density routing;
- SI/PI integration with Sigrity.
The skill that transfers between CAD packages is constraint-driven board design, not memorizing menu locations.
53. Siemens Xpedition
Xpedition targets high-density and advanced PCB systems.
Siemens describes current Xpedition Advanced capabilities including
- high-speed constraints;
- routing automation;
- rigid-flex design;
- integrated SI;
- ECAD/MCAD collaboration;
- manufacturing documentation.
Official: Siemens Xpedition Advanced.
Rigid-flex reference: Siemens Xpedition rigid-flex design.
54. KiCad
KiCad is a capable open-source PCB platform.
Its PCB editor supports
- differential pairs;
- custom DRC rules;
- length tuning;
- differential skew tuning;
- time-domain delay tuning;
- interactive routing;
- 3-D inspection.
Official: KiCad PCB Editor.
55. SI/PI and electromagnetic tools
CAD layout is only one layer.
Useful analysis tools include
- Cadence Sigrity — PCB/package SI/PI;
- Siemens HyperLynx — SI/PI and high-speed verification;
- Ansys SIwave — board/package SI/PI;
- Ansys HFSS — 3-D full-wave structures, vias, connectors;
- Keysight ADS — channel, RF/microwave and SerDes analysis.
Know what problem requires 2-D/3-D extraction versus a simple transmission-line model.
Part XIV — Bring-up and debug
56. Board bring-up order
A disciplined sequence is
\[\boxed{ \text{shorts} \rightarrow \text{power rails} \rightarrow \text{clocks} \rightarrow \text{reset} \rightarrow \text{JTAG/debug} \rightarrow \text{low-speed buses} \rightarrow \text{high-speed links}. }\]Do not begin by blaming PCIe if the PHY PLL supply is wrong.
57. Instruments you should be comfortable with
| Instrument | Use |
|---|---|
| DMM | rails, resistance, continuity |
| Oscilloscope | clocks, power noise, eye/jitter, signal quality |
| Logic analyzer | I²C/SPI/UART/protocol timing |
| TDR/TDT | impedance/discontinuities |
| VNA | channel S-parameters |
| BERT | BER/jitter tolerance |
| Protocol analyzer | PCIe/USB/Thunderbolt transaction/link debug |
| Current probe | rail/load/transient behavior |
58. Common failure patterns
PCIe
- trains at lower generation;
- narrower negotiated width;
- repeated Recovery;
- lane-specific errors.
USB / Thunderbolt
- orientation-specific failure;
- USB2 works but SuperSpeed fails;
- device falls back to lower speed;
- PD works but data does not;
- marginal cable/connector behavior.
I²C
- SDA stuck low;
- wrong address;
- missing ACK;
- slow rise time;
- level mismatch.
SPI
- incorrect CPOL/CPHA;
- setup/hold violation;
- clock ringing;
- wrong CS timing.
UART
- wrong baud;
- TX/RX not crossed;
- wrong voltage standard;
- framing error.
HDMI
- low-rate video works, high-rate fails;
- display detection/EDID failure;
- HPD/DDC issue;
- ESD/connector loss limiting FRL.
Part XV — What “proficiency” should mean
59. Interface proficiency
You should be able to explain for each interface:
- electrical topology;
- clocking;
- initialization/training;
- speed generations;
- power/sideband signals;
- PCB constraints;
- compliance metrics;
- likely failure modes;
- debug approach.
60. PCB/flex proficiency
You should be able to
- read and create a stackup;
- define controlled-impedance rules;
- route differential pairs;
- manage length/delay constraints;
- understand return paths;
- review via/connector discontinuities;
- design a PDN and decoupling strategy;
- create flex bend/stiffener/coverlay constraints;
- perform DRC;
- review fabrication/assembly outputs;
- work with a PCB fabricator;
- correlate board measurements with SI simulation.
61. Interview-level summary
A strong answer to “Do you know PCIe/USB/I²C/SPI/UART/HDMI and PCB layout?” should connect protocol and physical design:
A digital interface is only reliable when its protocol, PHY, clocking, power, connector and PCB channel are designed together. For high-speed interfaces I treat the package, vias, traces, connectors and cable as one channel, define stackup/impedance and loss constraints before routing, and validate with SI tools and measurements. For low-speed buses I still check voltage levels, pull-ups, timing and edge integrity. For flex/rigid-flex I add mechanical bend, coverlay, stiffener and fabrication constraints to the electrical design.
Part XVI — Official and practical references
62. Interface specifications
- PCI-SIG — PCI Express specifications
- PCI Express Base Specification overview
- USB-IF — USB4 Specification v2.0
- USB-IF — USB Type-C specification
- NXP UM10204 — I²C-bus specification
- Intel — Thunderbolt Technology
- HDMI — HDMI 2.2 Specification overview
SPI and UART are implemented across many device families rather than governed by one single modern public specification; use the controller/target device datasheets for timing and electrical limits.
63. PCB / flex references
- IPC-2223D — Flexible/Rigid-Flexible Printed Board Design
- IPC-6013D — Flexible/Rigid-Flexible Board Qualification and Performance
- Altium — Rigid-Flex Design
- Siemens — Xpedition Advanced
- KiCad PCB Editor
64. Related pages
Signal Integrity & Power Integrity · High-Speed SerDes & IO · Silicon Interfaces & SoC · ADC & DAC · Arduino Boards · Measurements & Instruments
Related Ansys/Synopsys simulation software
See Ansys Electronics & Semiconductor Software for HFSS, SIwave, Q3D, Icepak, HFSS-IC and chip/package workflows.