---
title: "Retimers and redrivers"
question: "Why is there a chip beside every Thunderbolt port?"
url: "https://nearfieldnotes.com/videos/retimers/"
site: "Near Field Notes"
video: "https://nearfieldnotes.com/media/retimers/full.mp4"
duration: "2:44"
date: "2026-11-16"
note: "https://nearfieldnotes.com/notes/retimers/"
---

# Retimers and redrivers (video)

Why is there a chip beside every Thunderbolt port?

2:44, with sound. [Read Note 42](https://nearfieldnotes.com/notes/retimers/) for the full answer, with figures, conditions and sources.

## Transcript

On this MacBook Pro, every Thunderbolt port has its own chip, a centimeter from the connector.

Here's why. I simulated 20 Gb/s on ordinary laptop board. The signal passes the spec for 70 mm. These ports are about 130 from the processor.

Two kinds of chip can fix that. Here's what each one does, why USB4 only allows one of them at the port, and the cable with a pair built into its plugs.

Each bit lasts 50 ps. A board trace loses the fast part of every edge, so each bit smears into the next.

Overlay every bit and you get an eye. USB4 measures it right at the port, and it has to stay 240 mV open.

At 10 mm of trace, my simulated eye is 363 mV. At 70, 245. Past that, it fails.

That's 7.5 dB of loss, the same figure the spec recommends for a laptop. At 130 mm the eye is down to 164.

Fix one is a redriver: an analog amplifier that boosts high frequencies more than low ones, to undo the trace.

With one beside the port, the same lane passes out to 210 mm. Three times as far.

But it never reads the data. Leftover smear and timing jitter ride through it, and down the cable.

So at the far end of the cable, the eye narrows with every millimeter of board: from 30 ps down to 20, the limit.

I tested what closes it. Noise and crosstalk each moved the eye by under 1 mV. It's the smear and the jitter.

Fix two is a retimer: a complete receiver and a complete transmitter in one package.

It recovers the clock from the data, decides if each bit is a 1 or a 0, and sends a fresh copy.

So the far-end eye stays 30 ps wide, whether the trace is 20 mm or 400. The board only has to reach the retimer.

This is Intel's, on another laptop's board. 6 mm square, with its own crystal beside it. Intel lists it at $2.90.

Now the rule. USB4 says if a laptop uses a redriver, a retimer has to sit between it and the connector.

In my simulation, a redriver alone still passes at 130 mm, with a narrower eye. The spec rules it out anyway.

At 10 Gb/s, USB 3.2 lets a redriver work alone.

So this MacBook Pro carries three retimers. The first M1 MacBooks carried two, reported as the last Intel parts left in a Mac.

An iPhone 15 Pro's port runs at 10 Gb/s, half the rate on each lane. Its teardown lists no such chip.

Framework left Thunderbolt off its Laptop 12, and gave retimer cost as the reason.

And Apple's 1.8 m Thunderbolt cable has a retimer in each plug. The 1 m cable has none.

No chip: 70 mm. Redriver: 210. Retimer: past 400, where my sweep ended.

Which is why a port 130 mm from the processor gets a chip of its own.

---

Source: https://nearfieldnotes.com/videos/retimers/
