ESD damage to XMOS U6

Technical discussions around xCORE processors (e.g. xcore-200 & xcore.ai).
User avatar
mon2
XCore Legend
Posts: 1913
Joined: Thu Jun 10, 2010 11:43 am
Contact:

Post by mon2 »

Chris -

the Socay polymer fuse we are using in our designs is:

Part Number: SCF035-1206R
Quantity: 8 K
Unit Price: 0.032 USD/ pc

Lower qty pricing may be slightly higher than the above but should be the MOQ of a full tape & reel for proper automated assemblies. You should select the proper current / trip rating for your project.

Getting back to the defective boards - the USB interface pins are very likely interfacing with an internal PHY. Respectively, even if the USB portion of the design is damaged, guessing that the XMOS XTAG interface should be still functional. However, XMOS may know more on the specifics. Do review your power supply rails to see if they are proper and operational.

What are the details of your power supply inductors ? P/N ? Vendor ?

Are the damaged XMOS devices warmer than a working XMOS device ?

Is the damaged XMOS board drawing more current than a working XMOS board ?


User avatar
Ross
XCore Expert
Posts: 962
Joined: Thu Dec 10, 2009 9:20 pm
Location: Bristol, UK

Post by Ross »

Documentation should be in the usb lib documentation. If the device is bus powered then you don't need to connect it (since the pull up will disappear when its unplugged due to no power..)

For a self-powered device you *should* connect it if you care about passing a USB compliance test. You can run the USB lib in "bus powered" mode on a self powered device - this disables the vbus checking but is not advised.
CMJ
Member
Posts: 8
Joined: Thu Mar 03, 2016 6:09 pm

Post by CMJ »

Thanks for all the ideas and information. We have been examining one board that has failed and found out some more information:
  • There is no output at the main PLL output pin (note: we have OSC_EXT pulled low to select an external clock source).
  • The chip is not completely dead because we were able to connect to it from JTAG (not XTAG) and using that we could read and reprogram the connected serial flash. Reprogramming the flash did not result in the board coming back to life however.
Our current external clock uses a crystal so next we will try using a clock module to see if that helps. If anyone has any ideas what in our circuit or environment could have caused this PLL output to stop functioning we are interested in all theories!

Thanks,

Chris
User avatar
mon2
XCore Legend
Posts: 1913
Joined: Thu Jun 10, 2010 11:43 am
Contact:

Post by mon2 »

Chris -

1) does your external crystal based oscillator continue to oscillate on the non-working boards ? If not, can you share the details of your external clock source for a review ? Without a working clock, this design will not function and may be the root cause.

What is the external crystal value ? If known, which vendor, part number and the value of the loading caps ? Is your design featuring an external gate for this oscillator ?

In the datasheet, the recommended value is 24 Mhz - is that not oscillating ?

Update comment - forgot that you are using USB so you are obligated to use 12 or 24 Mhz as the value for the external clock source. On this note, IF you should consider to apply an external oscillator of 12 or 24 Mhz value, please review 18.10 of the datasheet which notes that the external oscillator must be of a 1V8 value (1.98V is the MAX but since 1V8 is the power rail for the XMOS device, consider to use a 1V8 oscillator. These are not so common but certainly available from offshore vendors. When penny pinching on BOM and assembly fees, you must weigh how many parts must be placed to operate the external crystal oscillator vs. a single device for the same purpose (ie. 3 parts vs 1 part). Varying with the placement costs, it may be wiser to use an external clock oscillator as a single part.

The caps used on crystal based oscillators can be a bear to get right so often it is easier to just source a full SMD oscillator. We can share the details of some very solid crystal and oscillator suppliers.

For crystals, you should be paying about $ 0.08 to $ 0.10 USD each (30 PPM).

For SMD oscillators, you should be paying about $ 0.30 to $ 0.50 USD each (30 to 50 PPM).

Contacts:

We have met with WTL at last year's HK trade fair and they have a very extensive line of devices. Recently started to consume their parts with success.

Lori Sales I Shenzhen WTL Electronic Technology Co., Ltd
T: +86-755-8267 7582 ext: 801 M: +86 135 3042 0550
:wtl001@wtlcrystals.com:wtl-delly:WTL-crystals
W : http://www.wtlcrystals.com Alibaba: http://wtlcrystals.en.alibaba.com
A: Room 505,5th Floor,No 6 factory, Zhong Chu Building,4th Ba Gua Road, Ba Gua Ling, Futian District, Shenzhen 518029,China


Our long time supplier is - used on 500k-1M+ boards to date. No issues on QC but pricing is higher till we raised the concern. Private labelled parts (laser marked) for our company. Have visited their Shenzhen operation a number of years ago.

Foina Yeung
-------------------------------
TEL: 86-186 8242 0160
e-mail: foinayeung@hotmail.com
skype: foina-chinafronter

F.O.B. Shenzhen, China.

If you want a really solid product, consider silicon oscillators or MEMS oscillators which would allow for a pure solid state design in that your board could be hit against a wall without concern of damage to the crystal. Best to review the specs of such parts. We have not yet moved to this technology but may do so for some military accounts working on drone designs.

We have used 5x7 size for ages but they are pigs on the PCB relative to the current designs. For that reason we are now using 3225 = 3.2 x 2.5 mm sized parts. With the some negotiations it is possible to get your company name laser marked on the parts (room and font size permitting).


2) moving forward, you noted the use of JTAG to reflash the target flash device. If you apply the same method on a known good board - does the same procedure and format of the flash allow your working board to remain working ? Just concerned about the serial data stream being applied using this method. It is vital the JTAG method be solid else the XMOS does not have working firmware. Perhaps you can code a small routine to blink a local LED -> test this code on any XMOS devkit -> then apply onto a working custom XMOS board you are producing to confirm the code still works using the JTAG tool -> only then proceed to apply onto the non-working board. In the end, be sure the LED is able to flash in every test. Again, no clock = no love and the design will fail to operate.
CMJ
Member
Posts: 8
Joined: Thu Mar 03, 2016 6:09 pm

Post by CMJ »

Thanks for the detailed response mon2. We had to order some 1V8 oscillators as we did not have any in house, so we will try it tomorrow. I will update back with more information.

Our crystal is 24MHz, and on the non-working board it does not oscillate. The caps are 33pF, and there's no gate. It is cheap but I've found crystals can be unreliable so if we have to switch to the modules that is well worth it.

Regarding JTAG reprogramming of the serial flash: yes this method is reliable on working boards. We have done this several dozen times with no failures.
User avatar
mon2
XCore Legend
Posts: 1913
Joined: Thu Jun 10, 2010 11:43 am
Contact:

Post by mon2 »

The value of 33 pf for the loading caps sounds quite high. Do confirm with the datasheet on the recommended values for these caps and also the load capacitance (Cl) of the crystal.

Be sure you are using a 'parallel resonant, fundamental mode of 24 Mhz crystal'.

http://electronics.stackexchange.com/qu ... ls-to-work

https://blog.adafruit.com/2012/01/24/ch ... ur-design/

Try 18pf on each leg; 12 pf on each leg if you are confident that the crystal is not defective from poor soldering or damaged due to drop or hard hit to the crystal, etc. The use of the external oscillator is a good idea to confirm that the root cause is the clock source. After the oscillator test, consider to tweak the loading caps.

Perhaps the 33 pf load caps are suitable for the parts used by XMOS (vendor & p/n specific to the build) but may not be ideal for your crystals.
CMJ
Member
Posts: 8
Joined: Thu Mar 03, 2016 6:09 pm

Post by CMJ »

Here is an update on what we've found after examining a few failed boards. On all boards that have failed, we observe that the USB Data+ line is damaged. Our theory is that our ESD protection on the USB Data+ and - lines is incorrect. It may be excessive because the protection devices clamps D+ to VBUS (as well as GND). We are going to spin the board to clamp only to GND and retest.
Post Reply