JL
JLink Technology Semiconductor Distributor
RFQ
RS-485 Wiring & Termination Design: Termination, Fail-Safe Biasing, Daisy-Chain and Stubs
RS-485ModbusTerminationFail-SafeIndustrialWiring

RS-485 Wiring & Termination Design: Termination, Fail-Safe Biasing, Daisy-Chain and Stubs

RS-485 instability is 90% wiring: termination not at both ends, star topology, no bias on the idle bus. This guide covers the three rules — termination = cable impedance (120Ω is typical, not mandated), a linear daisy-chain, and fail-safe biasing — plus stub/grounding/node-count/rate-length engineering rules. RS-485 is the Modbus RTU physical layer, funneling to W5500/DM9051 RTU→TCP gateways.

JL Reviewed by JLink Technology engineering team · Updated 2026-06-30

The three rules you must get right

RS-485 rarely "fails in the chip"; it mostly fails in the wiring. In practice, nine out of ten "RS-485 comms unstable" tickets come down to three things done wrong: (1) termination not placed at the two ends of the bus, (2) topology wired as a star or with branches instead of a linear daisy-chain, and (3) no fail-safe biasing on the idle bus. These three decide whether RS-485 is stable, and we take each in turn (always defer exact spec values to the datasheet of the transceiver you use; this article gives the engineering rules and formulas).

Termination: both ends, = cable impedance

Where do the termination resistors go, and how big? The rule: put one resistor at each of the two physical ENDS of the trunk only, and make its value equal to the cable's characteristic impedance Z₀. Standard RS-485 twisted pair has Z₀ ≈ 120Ω, so the most common choice is one 120Ω at each end — but remember "120Ω is typical, not mandated": the real rule is "R_T = the Z₀ of your cable," so use whatever value your cable is (a mismatch reflects a fraction (R_T−Z₀)/(Z₀+R_T)). Don't add termination at intermediate nodes — that over-loads the driver and drops signal amplitude. The cost: every termination is a permanent DC load; an RS-485 driver is specified to still deliver at least 1.5V differential into 54Ω (the two 120Ω terminations in parallel with 32 unit loads), so termination raises driver current. In noisy environments the single 120Ω can be split into 2×60Ω with a capacitor to ground for common-mode filtering (match to 1%).

⚠️ Fail-safe biasing: why an idle bus gives garbage

Why do you get garbage when no one is talking on the bus? Because when the bus is idle (all drivers disabled), lines A and B float, the differential voltage approaches 0, and the receiver falls into its "indeterminate region" (output could be 1 or 0), so noise on the line gets read as data. Note too: an ordinary "standard fail-safe" receiver only holds a defined state when its inputs are OPEN — once there's a termination resistor (or a short) across the bus, V_AB collapses to about 0mV and it drops into the indeterminate region anyway. Two fixes: (1) use a "true / advanced fail-safe" transceiver — it shifts the decision threshold negative (indeterminate band about −50~−200mV instead of ±200mV), so open, short, and terminated-idle inputs all hold a defined high with no external parts; or (2) add a bias resistor pair at ONE point on the bus (usually the master): pull A up to VCC and B down to GND, sized so the idle V_AB exceeds the receiver threshold (~200mV, more for margin). Key warning: the "4.7kΩ bias" you'll see online only holds with open inputs — once there's a termination on the line, that value becomes ineffective and must be recomputed together with the termination. This is the single most-common real-world trap.

Topology, stubs, grounding, node count, rate×length

The remaining engineering rules: topology — it must be a linear daisy-chain (one trunk, each node brought out on a "short" stub); no star/ring/tree branches (a branch = an impedance discontinuity = reflection); keep stubs as short as possible, rule of thumb is an electrical length under 1/10 of the driver's output rise time (slower-edge drivers tolerate longer stubs; look up the actual value in that part's datasheet). Grounding — besides A/B, run a dedicated signal-ground/reference; never use "earth" as the return path, because ground-potential differences (GPD) between remote nodes become common-mode noise that can exceed the receiver's common-mode range (the RS-485 spec is −7V to +12V) and cause "works today, breaks tomorrow"; for long runs use isolated transceivers + isolated power. Node count — a standard driver drives 32 unit loads (UL, 1 UL ≈ 12kΩ); with 1/8-UL low-load transceivers you can reach up to 256 nodes (but the bias network itself consumes about 20 UL, so the real count is lower). Rate vs length — inversely related; the common rule of thumb is "bit rate (bit/s) × length (m) ≤ 10⁸" (50m ≈ 2Mbit/s; TI uses a more conservative 10⁷), anchored by ~10Mbit/s at short range and ~1200m at low rate. Duplex — 2-wire half-duplex saves wire but needs DE/RE direction control and bus turnaround (only one driver active at a time to avoid contention); 4-wire full-duplex transmits and receives simultaneously, faster but more wire.

RS-485 = the Modbus RTU physical layer → transceiver + gateway

Why should a semiconductor distributor understand this? Because RS-485 is the physical layer of industrial Modbus RTU — an MCU's UART is 3.3V/5V TTL, so to reach the differential bus it needs a "TTL↔RS-485 transceiver" (with DE/RE direction control; auto-direction parts handle the switching timing for you). And to put an RS-485 segment onto Ethernet/SCADA, you build an RTU→TCP gateway: downward an RS-485 transceiver + MCU, upward an SPI Ethernet controller — most commonly the WIZnet W5500 (hardwired TCP/IP, an easy serial-to-Modbus-TCP path), or an equivalent SPI MAC+PHY like the DAVICOM DM9051. JLink Technology supplies the parts for an RS-485/Modbus gateway — the W5500, DM9051 and GD32 MCUs; tell us your node count, cable length, rate and whether you need isolation, and we will help plan it and reply with samples, datasheets and pricing. (Further reading on this site: Modbus RTU vs TCP, RS-232 vs RS-485 vs CAN, and the W5500/DM9051 gateway articles.)

Products mentioned

W5500 In Stock

W5500 Hardwired TCP/IP Ethernet Controller

W5500 is WIZnet's flagship Ethernet controller with a unique hardwired TCP/IP stack, enabling direct network connectivity without software TCP/IP implementation. Connects to the host MCU via SPI, features 32KB TX/RX buffer, supports up to 8 sockets — the most widely adopted Ethernet chip in the Arduino/Raspberry Pi ecosystem.

EthernetTCP/IPSPI
DM9051 In Stock

DM9051 SPI Interface 10/100M Ethernet Controller

DM9051 is a DAVICOM single-chip 10/100M Fast Ethernet controller with an SPI interface, eliminating the need for 8/16-bit parallel bus and greatly simplifying hardware design. Features 16KB SRAM TX/RX buffer, full-duplex flow control, and native Linux kernel driver support — ideal for network expansion on ARM or RISC-V MCU systems.

EthernetSPIFast Ethernet
GD32F303 In Stock

GD32F303 Arm Cortex-M4 Mainstream MCU

The GD32F303 is a mainstream Arm Cortex-M4 MCU from GigaDevice at 120MHz (with DSP/FPU/MPU), 128KB~3MB Flash, 32~96KB SRAM, 3× 12-bit ADC, 2× DAC, rich timers, and USB FS/CAN/SDIO. In LQFP48/64/100/144, it is commonly used as an M4 upgrade from STM32F1 or an STM32F303-class alternative.

MCU32-bitCortex-M4

Need these components or design help?

JLink Technology provides parts, datasheets, and engineering support.

Request a Quote