Inter chip-usb (ic-usb) questions

[MChapman]

Good day,
I have designed a couple USB 2.0 to serial devices but am only a recent graduate (as of last Saturday) and have been asked to interface with a cell module using ic-usb. I have read the ic-usb addendum to USB 2.0 but will need to make sure the device will be robust enough to last at least five years.

A few key elements that I am aware to design in are:
1. 90 ohm differential pair
2. Short traces (<10cm)

Design criteria:
1. Lithium battery powered.
2. The microcontroller (EFM32 Giant Gecko) and the Cell module are supplied from the same source at the same potential.
3. Trace length of D+ and D- will be <18cm

I used Saturn PCB Design Toolkit to help with the trace impedance. My trace width is 10mil (0.254mm)and height to reference plane is 7.5mil (0.1905mm). Using Coplanar Wave the impedance approximates to 61.412 ohms therefore I added a resistor in series with the D+ and D- traces.

Does anyone have experience with ic-usb? Is a 29 ohm resistor the only passive component needed on this short of a trace. I understand that on other USB devices a decoupling capacitor is often added but I believe that is due to the inductive impedance of the long cables often on the devices. Please correct me if I'm wrong.

I noticed that Texas Instruments has a dedicated ic-usb voltage translator (TXS0202) but since both ICs are running off the same 3.3v source I thought this would be overkill.

Attached is a quick schematic of what I intend. Since I'm not using Vbus to supply the cell module I'm tying VregIn and VregOut together with a 4.7uF capacitor as in Silicon Labs' "USB Hardware Design Guide" AN0046 - app note.

 

[Harald Kapp]

The IC-USB addendum to the USB2.0 specification states:

• "The length of the physical link is limited to 10 cm." You exceed this limit by almost a factor of 2!

At 10 cm and USB 2.0 speeds impedance matching should not be required. USB 2.0 high-speed operates at 480 MBit/s which equals 240 MHz in frequency. This translates to a wavelength of approx. 1.25 m which is way more than the distance between two IC-USB devices. Therefore it is not necessary to consider this short trace as a transmission line and you do not need impedance matching to avoid reflections.

If you want to match the impedances anyway, select a resistor with low inductance as a standard resistor may (depending on the manufacturing and trimming process) exhibit considerable inductive impeadnce at 240 Mhz.

 

[MChapman]

I'm glad you caught that. It's merely a typo the lenghts are actually 18mm not 18cm. Thank you for the advice on impedance matching.

If you don't mind I have another question. I have heard that the length of differential traces must be exactly equal and I have heard that there are tolerances that can be lived with. Do you have advice on drawing differential traces?

 

[Harald Kapp]

Intuitively it seems to be only logical that a differential pair needs to have traces of the same length, otherwise you'd introduce asymmetries which is counter productive. As to the tolerances I'm sorry, I have no clue.

 

[garublador]

I can't really say that we've done a through analysis of how good the signal integrity is in our designs, but we frequently allow (but not necessarily use) 500 mils (12.7mm) of uncoupled length on USB lines in our designs. I'd guess that if you're only going 18mm that you'll probably have an easy time keeping well below that.

 

[MChapman]

Thank you both fory your remarks. I'm glad to hear that designs work with 500mils tolerance. That would explain why my first USB design worked. I made a simple USB to uart using an FTDI chip. The traces were very short but not equally matched. After learning a little more about differential pairs I was puzzled at why something designed so badly still worked well.

 

[garublador]

If the signal lines are very short then they can't be all that mismatched.

Most of these numbers are difficult to quantify because they depend on so many other variables. The signal integrity will depend on mismatched lengths, impedance matching (including discontinuities), uncoupled lengths, proximity to noise sources and whatever combination of those you get anytime you go though a connector. So saying how good any one of those can be with your design still working is tough. On simple designs it's pretty easy to keep well within the guidelines without trying too hard. You start running into issues when you start running the signals from board to board and connector to connector though minefields of adjacent split planes and other close proximity signals and end up stacking all of the uncoupled and mismatched lengths together where you get real signal integrity issues.

 

[Harald Kapp]

In the light of the sometimes miserable quality of standard USB cables, I don't think you have any reason to worry.

 

[garublador]

In the light of the sometimes miserable quality of standard USB cables, I don't think you have any reason to worry.

Our layout guru here likes to point that out. We worry about hundredths of an inch when it comes to uncoupled lengths and then plug in the cheapest USB cable we can find. Ethernet is even funnier when you consider the lengths of some of those cables.