1. Home
  2. Electronics Design
  3. Problem with TI TMS320DM6446 and crystal not starting

Problem with TI TMS320DM6446 and crystal not starting

Has anyone had any difficulties with the crystal used for the USB peripheral interface on the TMS320DM6446 not starting up? The leading production board failure that our board supplier reports is that this particular crystal circuit is not staring, sometimes replacing the crystal 'fixes' the board, sometimes not. Based on what we've seen so far it doesn't appear to be a generic 'crystal not starting problem', but seems to be quite specific to the TI part.

I was just wondering if anyone here has happened to encountered this problem. Details for those interested are listed below.

Kevin Jennings

=============== Details of the design:

- Crystal circuit layout is reasonable, the crystal and caps are located close to the 6446, short traces.

- The 6446 has two crystal input circuits, one to drive the processor core, the other is strictly for USB. We're not observing any problems with the crystal used to drive the processor core, only the USB one.

- There are a total of four crystals on the board, they're all the same supplier (different frequencies only). The other three crystals do not have any reported startup issues.

- Per the TI spec, both the 'processor core crystal' and 'USB crystal' I/O pins are powered from the same supply voltage (DVdd18). There is nothing indicating that the DVdd18 supplies are in any way segregated for the USB pins, so it would seem that since one crystal circuit works and the other one doesn't that it wouldn't likely be a power supply issue (speculating).

- The TI part in question is TMS320DM6446ZWT. Pins 'F18' (name M24XI) and 'F19' (M24XO) are the 'USB crystal' pins in question.

- The USB crystal that won't start is CITIZEN AMERICA CORP. CS10-24.000MABJ-UT

- The crystal capacitor load is spec'ed at 18pF which is implemented on the board with two 33pF caps, allowing for ~3pF of stray capacitance.

Some other observations:

- Most boards do not exhibit this problem. However, of the boards that end up in the 'not working' pile, the crystal not starting is the most frequently observed problem. On the couple of boards that we've investigated, the odds of the circuit starting up properly are ~20% or so.

- Touching with a scope probe, will sometimes get the circuit started, though many times it does not.

- Cycling power will sometimes get the circuit started.

- Waiting ~30 seconds will sometimes get the circuit started.

- Measuring the frequency error that results from using the two 33 pF caps and comparing that with the frequency error that results from instead using two 18 pF caps indicates that the ideal value to use on the board is ~30pF which suggests that the load capacitance is about right.

- The odds of the circuit starting up is the same whether using the

33pF or 18pF capacitors which again suggests that the crystal load capacitance is not the issue.

- When the circuit does not start, both crystal I/O pins sit at 0 Volts. Since the 6446 should be implementing an inverter function, this seems to suggest that maybe that inverter is either not biased into it's linear operating region, or the gain of that inverter circuit is too low or the inverter is not being powered.

Based on all of the above, the root cause of the problem would seem to be

- Defective crystals (going to try swapping the two physical crystals to see if that is the case)

- Defective 6446 (looking for input from someone who might have seen this)

- Unidentified design error (any other suggestions on what that error could be?)

Reply to
KJ
Loading thread data ...

Sounds like you are shipping time bombs.

We don't use crystals any more: only crystal oscillators.

John

Reply to
John Larkin

Some crystal oscillators require a 1M resistor across the crytal. Did you try that? What does TI say about this?

--
Failure does not prove something is impossible, failure simply
indicates you are not using the right tools...
                     "If it doesn\'t fit, use a bigger hammer!"
--------------------------------------------------------------
Reply to
Nico Coesel

Did you form a T network on the I and O pins with the caps, with the T to common.

I had an issue where I had to use a set of caps that were offset from each other to force an imbalance.

How stable are the rail lines?

Reply to
Jamie

rom

I'm not sure what offset you mean that you forced to not be balanced.

Possible power issues are still under investigation. At this point, about all I can say is that both crystals that connect to the 6446 are powered by +1.8V, one crystal circuit works, the other one doesn't.

Kevin Jennings

Reply to
KJ

The resistor across the crystal is to force some form of 'digital' invertor to operate in a linear region. This can also be accomplished on chip with a transistor. Since TI says to only connect the caps and no resistor, the implication is that the 6446 takes care of the biasing on chip.

Kevin Jennings

Reply to
KJ

Have you tried a driving device reset true and see what happens. OSCPDWN reset value is for OSC power off. Granted you say that sometimes touching a scope probe will cause the xtal to start, but I think I would still verify and verify again that the firmware is correctly initializing the USB. I once found a problem (not this TI device) where one of the ISRs was clobbering a peripheral control register at random intervals.

--
Joe Chisolm
Marble Falls, Tx.
Reply to
Joe Chisolm

what I meant was, the cap values used on the IN/OUT of the crystal terminals are not the same.

Also, board lay out can cause issues.. You maybe getting a problem from an adjacent circuit near by and/or the back side of the board. Near by parasitic currents and capacitance plays havoc when board space is tight!.

I had this problem when trying to get a lot of smt components with a uC, inside a hood of a 9 pin RS-232. The IO on one side was effecting an input of the uC on the other side due to board lay out. I was able to stuff a R in there to fix the first batch of boards. I redesign the rest.

Reply to
Jamie

Sounds like bad code me to ;)

Reply to
Jamie

But it doesn't work... reliable... Just try the resistor.

--
Failure does not prove something is impossible, failure simply
indicates you are not using the right tools...
                     "If it doesn\'t fit, use a bigger hammer!"
--------------------------------------------------------------
Reply to
Nico Coesel

Un bel giorno KJ digitò:

I had a similar problem with a different TI DSP. In that case the oscillator started every time, but sometimes the PLL took too much time to stabilize (I shut down the DSP and woke it several times per second, therefore the startup time was important). In my case the solution has been to add the 1M resistor, even if TI doesn't advise that explicitly. The solution has proven right over thousand of devices, though.

Perhaps if the PLL of your DSP doesn't lock up within a certain amount of time, the USB section is "disabled". I'd give a try with the 1M resistor anyway.

--
emboliaschizoide.splinder.com
Reply to
dalai lamah

It was! I think about 4 different people with 4 very different coding styles and no standards. Took me days to wade through the various pieces. I cannot point the finger too much though, I've done the same thing. ;)

--
Joe Chisolm
Marble Falls, Tx.
Reply to
Joe Chisolm

Well, maybe it does. But maybe not. I second Nico's suggestion. Take one of the worst boards and just try it. Also, does this processor offer the ability to select different burden caps inside? If yes then those code settings might be off.

If all that fails I suggest you post the layout, only the area with the crystal and the bypassing in the vicinity.

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.
Reply to
Joerg

The data sheet shows the "burden" capacitors as external. One mistake often made by amateurs is not following the directions for the right values of those capacitors, dependent on _frequency_.

Also, depending upon who designed it, the oscillator may or may not include the drive-limiting series R. Particularly problematic if the designers are located in that up and coming "English-speaking" democracy... you know the one I mean... Microchip is fond of doing that ;-)

My rule of thumb is to use an RC from the output of the oscillator, providing 45° of phase shift at the crystal frequency, the other capacitor is not nearly as critical (particularly with high-Q resonators) unless you overdo it and lose loop gain.

...Jim Thompson

--
| James E.Thompson, P.E.                           |    mens     |
| Analog Innovations, Inc.                         |     et      |
| Analog/Mixed-Signal ASIC\'s and Discrete Systems  |    manus    |
| Phoenix, Arizona  85048    Skype: Contacts Only  |             |
| Voice:(480)460-2350  Fax: Available upon request |  Brass Rat  |
| E-mail Icon at http://www.analog-innovations.com |    1962     |
             
 I love to cook with wine     Sometimes I even put it in the food
Reply to
Jim Thompson

The resistor might turn out to be the solution, but first I need to get to the bottom of what the actual problem is. Trying a resistor, whether it gets the board going or not, won't get to the bottom of understanding just what is the ailment...might provide another clue, but not the root cause.

Kevin Jennings

Reply to
KJ

Have you tried the crystals from the different batch?

Have you tried the 1M resistor across the crystal?

Are you sure this is not a silly problem with the board etching, leakage through the unwashed flux, etc?

Couple of times I had problems with the unstable operation of the microprocessor crystals in the small packages. FWIW increasing the load capacitors to 47pF helped the issue. Never had a problem with the HC-49 types.

Vladimir Vassilevsky DSP and Mixed Signal Design Consultant

formatting link

Reply to
Vladimir Vassilevsky

They do recommend that optional 1M resistors for the USB oscillator on page 104:

formatting link

Wasn't it you who once said in a design review "And how does it start?".

Also, according to the datasheet the caps must be tied to M24VSS, not to the GND plane (goose bumps showing..). Somewhere in there I've read that overtone crystals are not supported. Might be worth checking.

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.
Reply to
Joerg

Page 103 shows the 1M resistor. Also, check into whether you have an overtone crystal and whether that's supported for the USB oscillator. Possibly it isn't allowed. Are the caps connected to M24VSS? Should not be tied to the ground plane, for whatever reason.

formatting link 

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.
Reply to
Joerg

Many times... mostly about regulators that loop back upon themselves... unless there's a kick-start circuit somewhere, you have to pray for leakage ;-)

My favorite blooper was from an Atmel design team from Germany... an RF amplifier with no DC loop to set its output Q-point... it worked on their simulator... their engineer even argued with me over it :-(

The typical pi network isn't selective enough to pick an overtone. In those cases I prefer a loose LC. My favorite actually runs the crystal at series mode... unfortunately it take three pins :-( But it also provides a nice sinusoidal signal.

...Jim Thompson

--
| James E.Thompson, P.E.                           |    mens     |
| Analog Innovations, Inc.                         |     et      |
| Analog/Mixed-Signal ASIC\'s and Discrete Systems  |    manus    |
| Phoenix, Arizona  85048    Skype: Contacts Only  |             |
| Voice:(480)460-2350  Fax: Available upon request |  Brass Rat  |
| E-mail Icon at http://www.analog-innovations.com |    1962     |
             
 I love to cook with wine     Sometimes I even put it in the food
Reply to
Jim Thompson

Actually, they just say it is 'optional', but you're right, the PCB should've had provisions for that 'option'. Not sure how many times I've looked at that figure, maybe next time I'll clean the glasses first or something...RTFM

Now I did read that part and had already checked that the guy who designed it, got that part right.

Reply to
KJ

ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here. All logos and trade names are the property of their respective owners.