Posted by Eric Huang on October 31, 2017
USB 3.2 brings us from USB 3.1’s 10 Gbps to 20 Gbps.
The always cunning and clever USB-IF did a lot to make this work well.
USB 3.2 only works with the new Type C connector. The Type-C connector and Cable provide two lanes, Lane 0 and Lane 1. The standard allows use of both lanes simultaneously at 10 Gbps. 2×10 Gbps = 20 Gbps.
Preserving the existing PHY signaling speeds allows:
The data coming in and exiting the USB Type-C port along 2 lane bonded ports means the PHY needs to be capable of bringing in two lanes (or more) of data at a time. The SERDES needs those two lanes, and the PCS needs to be modified to handle two lanes of incoming traffic. This brings up the concept of skew. If two lanes of data come in at the same time, then they have to come in pretty close to the same time to be useful. For example, when watching a video through USB, if the audio is coming in on Lane 0 and the video is coming in on lane one, the two need to be aligned**. If it’s mis-aligned, it’s possible that the video and audio will not look correct when viewing. During dialog the mouths would be moving, but would not align with the sound. To deal with this, the USB-IF specifies a de-skew of 6400 picoseconds. His means data from Lane 0 and Lane 1 that “belongs together” must arrive within 6400 ps of each other. Belongs together means in the same microframe of data.
*Cable lengths are not actually defined in the USB specifications. It specifies the amount of signal loss allowable between the transmitter and receiver. The USB-IF defines the qualities of the Cables and Connectors to try to achieve a 1 m length for USB 3.2 and USB 3.1, and a 2m length for USB 3.0. The fact is if the design has a great USB PHY on the receiver especially but both ends, good connectors, and an over specified cable, it’s possible to have less loss over the cable, and have the receiver interpret bad data and “fix” it to make it readable. This means the bit error rate stays down. The issue here is the materials for the USB cable will make the cable more expensive. (I couldn’t say how much). Also, to make a better PHY it might be physically larger and take longer to build and take up more area on a chip. The reason you might want to build a better PHY is to reduce the number of customer returns because the device doesn’t work. Consumers might buy a cheap device that absolutely must work with a PC. A PC might have a much better PHY so it can interoperate with a 2 dollar flash drive or a $3 cable. Consumers have little tolerance failing products and certainly not for USB ports.
** Audio data rates are slower than Video data rates so it’s unlikely this would be an issue, this is for illustration purposes only. IF the data is separated too much, it’s likely this will be registered as a bit error, and the data will be requested to be re-transmitted. This will slow down the overall data rate assuming some data gets through some of the time. It’d be possible to buffer the system to account for this, but the expectation is the system should not have to deal with this if all the parts, the USB Host, the USB peripheral, and the USB cable are all USB logo certified.
If you made it this far, I’ll be publishing every Wednesday.
Next week Audio Device Class 3.0 Part 2
Week after that Type C Primer Part 2
Two XKCD comics for you. Enjoy