Migrating from 100BASE-TX to 100BASE-T1 — What Changes in the AUTOSAR Stack

100BASE-TX uses MLT-3 (Multi-Level Transmit-3) on two separate wire pairs — one per direction. Because transmit and receive are physically separated, the PHY needs no echo cancellation.

100BASE-T1 uses PAM3 (Pulse Amplitude Modulation, 3 levels) on a single wire pair, bidirectionally. Transmit and receive signals overlap on the same line — the PHY has to subtract its own transmit signal using an adaptive filter to isolate the receive signal.

This has direct consequences for the PHY chip: a 100BASE-T1 PHY is internally significantly more complex than a 100BASE-TX PHY, because it has to implement an adaptive echo canceller, a near-end crosstalk filter, and trellis-coded PAM3 encoding, where TX gets by with simple 4B5B encoding and MLT-3.

100BASE-TX supports Auto-Negotiation per IEEE 802.3 Clause 28 — two connected ports automatically negotiate speed and duplex mode by exchanging Fast Link Pulse bursts.

100BASE-T1 has no Auto-Negotiation in the classic sense. There is only one speed (100 Mbit/s) and only full duplex — so Auto-Negotiation would add little value anyway. What it requires instead is a static master/slave role assignment that must be fixed before commissioning.

Since there is no Auto-Negotiation, every 100BASE-T1 link requires exactly one side configured as master (drives the clock) and the other as slave (synchronizes to the master’s clock). This role is fixed at system design time and stored in the ARXML — it does not change automatically at runtime.

The details of this configuration, the relevant ARXML parameters, and typical topology decisions (e.g., gateway always master, endpoints always slave) are covered in depth in the dedicated Master/Slave post in this series.

100BASE-TX supports PoE (Power over Ethernet, IEEE 802.3af/at) for office applications like IP phones or access points. 100BASE-T1 has no equivalent PoE — instead there is PoDL (Power over Data Line, IEEE 802.3bu), a separate, automotive-specific concept with its own voltage classes. In most vehicle topologies, though, PoDL is barely used so far, since ECUs already get power from the KL30/KL15 vehicle power supply.

The most obvious point: 100BASE-T1 is implemented by a different PHY chip than 100BASE-TX, and every PHY chip needs its own EthTrcv driver (a BSW module, close to the MCAL, supplied by the semiconductor vendor or the Tier 1).

The EthTrcv API itself (EthTrcv_Init, EthTrcv_GetLinkState, EthTrcv_SetTransceiverMode, …​) remains stable upward — that is the entire point of the AUTOSAR layered architecture. But the driver implementation underneath it is entirely new: different register maps, different init sequences, different power-up timing requirements.

As described above, the static master/slave assignment replaces Auto-Negotiation. In the AUTOSAR stack this shows up as a new ARXML parameter (usually modeled as vendor-specific, under EthTrcvConfigSet or similar) that simply did not exist for 100BASE-TX because it was not needed.

The full treatment of this topic — including typical topology patterns and failure modes from misconfiguration — is in the Master/Slave post in this series. For now: master/slave is not a software convention but a physical necessity for clock synchronization of the PAM3 signal — and an ARXML parameter that doesn’t match the real cabling results in a link that never comes up.

100BASE-TX PHYs commonly use Energy Detect as a WakeUp mechanism: the PHY detects that a link partner is active purely from the presence of signal energy on the line, and can derive a WakeUp event from that.

100BASE-T1 PHYs instead predominantly use a dedicated WakeUp pulse — a defined signal pattern (in practice often aligned with the IEEE 802.3bp/bw WakeUp specification, or vendor-specific) that a transmitting node actively generates to wake a sleeping node. Energy Detect alone is not reliable enough on 100BASE-T1, because noise and reflections on the line can more easily cause misinterpretation than with a clearly defined pulse pattern.

This is not a purely academic distinction for the AUTOSAR software developer: the ARXML parameters for WakeUp configuration (e.g., enable, debounce times, polarity) differ between the drivers, and carrying over the TX configuration unchanged frequently configures a WakeUp mechanism the new PHY doesn’t even offer.

Beyond the points already mentioned, the ARXML configuration differs in several further details:

The classic one: both sides of a link are configured as master (or both as slave). Since 100BASE-T1 has no Auto-Negotiation, there is no automatic fallback — the link simply never comes up.

A common root cause: in complex topologies with many ECUs, the master/slave assignment is maintained in a table that isn’t consistently reconciled against the actual ARXML of each individual ECU. An ECU gets swapped or reflashed — and ends up with the wrong role by mistake.

100BASE-T1 requires a line impedance of 100 Ω (±15 %) and careful differential-pair matching on the PCB. If the impedance matching isn’t recalculated during the hardware redesign (new PCB layout for the new PHY) — for instance because layout rules are carried over from the TX design without review — reflections appear on the line.

The symptoms are rarely a complete link failure; more often:

This class of failure is particularly unpleasant because it often goes undetected in bench-level functional testing (short cable, stable temperature) and only surfaces during vehicle durability testing or in the field.

As described in the WakeUp section: if the TX configuration is carried over unchanged, the WakeUp mechanism ends up either disabled or set to a mode the new PHY doesn’t support. The system works flawlessly under continuous operation — the failure only shows up once the ECU goes through its regular sleep/WakeUp cycle.