TI E2E Community
Digital Power Forum
UCD9222 interface with two TMS320C6678 DSP's
We seem to be having issues interfacing the TI UCD9222 PWM controller to two of the TI TMS320C6678 DSP’s in our design. If we connect up the PMBUS interface to the UCD9222 we see our temperature rise up to the 125 shutdown point and also our current draw seems to start off high and rise to close to the turn off point as well. We had been seeing this issue while running our temperature test but at ambient we can get the temp limit to trip and shut things down when we remove our fan from the system. The temp limit is turning off the part first but the current limit looks like at some point it would also turn the part off if the temp limit didn’t trip first. Something just doesn’t seem right to me as at ambient the current draws idles around 6-7A’s before we have much of anything running on the DSP’s. Our design should be a copy of the TMS3206678 EVM board and also use the default UCD9222 PWM xml project file except we copy Rail #1 over to Rail #2 to have two identical 1.0V outputs from the device to drive two TMS320C6678 DSP's.
If we power up the TMS320C6678 EVM board and load our project onto it we see at least a 1-2A lower current draw than our design at amibent.
Our Modified UCD9222 configuration file
Our Design Failure displayed in Power Designer
This looks pretty normal at first glance in that it seems like the UCD7242 is running hot with about 10A output current. Is there any way to hold the DSPs in reset/slow clock etc to change the output load to see the supply run at lower currents (and lower temperature)?
One thing to double check is the layout of the racing strip package and thermal vias. They are key to getting heat out of that package.
It might be good to measure curent input as well and do a quick efficiency calculation just to see if things seem reasonable (IMON is unconnected so I can't tell what input current is)
You may end up going to a higher current part or doing soemthing creative in cooling if the DSPs are truly running at this load, but we may as well look closely through things first.
If the other DSP is pulling close to 10A too this might just be a thermal issue.
It certainly wouldn't hurt to get a scope plot at the UCD7242 of Vin, Switch nodes, vouts and if you have a current probe, lift the output side of the inductors and put a wire loop in to measure output current. Just to make sure things look sane there too.
David so we did see mention of a 15A UCD74110 part in the EVM documentation but as far as we can tell this part was never released. Is there any other suggestions here if we need to move away from the UCD7242? I don't suppose there are any footprint compatable drop in replacement parts for the UCD7242 are there?
I am not sure this comepletely makes sense to me still though as the EVM kit seems to run fine we our code but then when we jump to our board we gain at least an amp of current usage. Is this because we are using two DSP's and not a single one like the EVM kit? Even still I would have expected just the two rails to be using around 5-6 Amps each and not the single rails to increase 1 Amp each.
Another item of information we have is our two DSP's are getting cooled to around 60C so we don't think we should be seeing any issues with the DSP's at this point. We did noticed a higher temp DSP but would this help us any if we are already within the spec anyway?
I am not aware of any duals that would be footprint compatible with higher current output.
The discrepancy between the two scenarios is interesting - is this measured the same way on each? When you say you gain an amp is that input to the system or output on a given rail? A scope plot of switch node vin/vout may still be useful.
There new power stages currently in preview status are: http://www.ti.com/product/ucd74120 This is a 25A max and there is a UCD74111 is on the same page and it is the 15A. version.
When I am referring to us gaining at least 1A of output according to the Power Designer interface I am reffering to an ambient temperature steady state with our software not running fully on both the dev kit and our own board. On this test we see about 5-6A on the EVM dev kit and 6-7A on our own board. The only thing we should try is to shut off one of our two DSP's so we would mirror more closely the EVM kit I guess. But even then I would expect the two different power rails to match the EVM reading anyway and I would have expected the readings to be 5-6A on Rail #1 and Rail #2 not 6-7A like we are seeing on our design. Both the EVM and our board should be running the DSP's at 1GHz. The EVM kit doesn't appear to heat up the UCD7242 part as much as you can touch it and it definately doesn't feel as hot as the one on our board. We don't have fans directly on top of the DSP's like the EVM kit but we have two fans blowing over the tall DSP heatsinks on our board and per our mechanical guy we far exceed the specs found in the datasheet for recommended air flow.
I can work on measuring our Vin/Vout but I guess I am still questioning the EVM referance design then as it appears the UCD7242 parts are not designed to support the DSP above a certain workload point? We are not running above the noted 1.25Ghz in the design notes which then states more than 10A of current might be required. Our buck and controller are placed between the two DSP's on our board so they probably will get a little hotter than the EVM kit which they are on the side of the DSP.
Is the UCD7242 thermally shadowed by the DSPs (do they block the airflow over the driver)? What thickness of copper is used for the PCB?
Would you be able to share either a PCB file or a screen shot of the PCB layers around the UCD7242 driver and the schematic for this section?
I could definately share board files I think if needed. We use altium designer and could provide the gerber files I believe. But attached for now is a 3D pdf of our board layout. You will see the on the bottom edge of the board between the two DSP's is the UCD7242 part at the very bottom and directly above it is the PWM UCD9222 controller. I didn't do the layout of this PCB but it appears the copper thickness is 1/2oz on a 12 layer board.
I could send someone the board files as needed not sure I want to just post them though. If I send to someone who would it be best to forward on to?
It would be helpful to see the copper layout, specifically the copper area that the UCD7242 part is heatsinking to. 1/2 oz copper is pretty light for a power design of this current level, so that may be a contributing factor. We'd need to see how much copper area the part is heatsinked to and what layer it is on.
As far as the higher current draw goes, I am not fully understanding the differences between your board and the eval board. If you have an extra DSP, it seems probably that the current would be higher.
One more thought. Have you tried reducing the switching frequency? This will reduce the power dissipated in the UCD7242.
I've got it now. You've got two 1.0V rails coming out of UCD7242. Each 1.0V rail independently powers a DSP. Can you verify that these 2 rails are not tied together anywhere?
Here are the actions we discussed:
1) Reduce switching frequency to 700kHz and note the temperature change on UCD7242
2) Compare rev of DSP between eval board and your board
3) Please measure input currents to the different boards
4) Measure output voltage on both 1.0V rails with an oscilloscope. Compare to eval board. Scope the output voltages on the 1.0V rail to see if there is anything fishy going on there. Please verify your PMBus readings with onboard measurements. I'd like to know if the DSP is changing the core voltage on the eval board and possibly not on your design. This measurement will also show you if there are any noise or ripple issues
My initial thought was that your board is running more power through the UCD7242, so it’s running hotter, causing a higher RDSon of the FETs and lower efficiency. However, it appears that you are measuring the output current of the core voltage rail, so that extra current is going to the DSP. There may be some copper losses on the way to the DSP, so you could measure the voltage output of the power supply and compare to the voltage input at the DSP and see how much difference there is. That could account for some loss.
My second thought was the accuracy of the UCD7242 current sense element may be affected over temp. The datasheet claims a +/-5% accuracy over load, so that would not account for the 1A change. I am investigating this further to see if there have been any cases of current measurement inaccuracy with this part.
You have confirmed that you are loading the same firmware on your board as well as the eval board.
Please take a look at rev of the DSPs to ensure there’s no difference there.
You confirmed that you are using different inductors and output caps from the eval boards, please send me the part #s or datasheets or post here.
Your board layout does not have much copper to dissipate heat from UCD7242 and copper thickness is only 0.5oz, so that is a contributing factor to the high temperature of the part. An improved layout with thicker copper and larger copper area is recommended.
I suggest that the simplest work around to reduce temperature would be to reduce the switching frequency through the config file. You are at 800kHz currently. I suggest to drop to 700kHz and measure the difference in temperature. This will tell you whether this is worth pursuing to a lower frequency or not. The lowest frequency that you can go to is dependent on your inductor size and current draw. Once you send me your inductor, I'll have a better idea of how low you can drop this frequency and possibly what inductor you may need to change to.
You clarified that you are measuring Tmon (Tj of UCD7242) through PMBUS. You are at 60C on the eval board with no airflow and are at 60C on your board with airflow. If this is true, it would negate my ideas that temperature difference is causing inefficiency or measurement inaccuracy. I still think measurement inaccuracy could be an issue, but may not be temp related.
From the initial layout notes on page 24 of the UCD7242 datasheet I don't see that our layout is that horrible I guess. The major flaw in our layout would probably be they do suggest 1oz as a minimum if not 2oz copper on the top layer. They state multiple 10-mil (0.25mm) diameter vias should be placed next to Vin and GND and in the datasheet they show on the UCD7242 test board around four vias right under the part on 12V in which we have and are actully larger (0.6mm) diameter vias and they have around 8 vias on GND right under the part where we only have 6 vias but we do again use a larger diameter via.
Attached is the polymer output capacitor datasheets we are using both 330uF and 220uF parts from Panasonic.
Attached is the Coiltronics (HCM0703-R47-R) 0.47uH inductor datasheet we use.
Also I believe both the EVM and our board are using the same exact prototype DSP parts as the part numbers appear identical:
The dev kit DSP is marked:
2010 TI 82
Our parts are marked:
Copyright 2010 TI 2 G1
All content and materials on this site are provided "as is". TI and its respective suppliers and providers of content make no representations about the suitability of these materials for any purpose and disclaim all warranties and conditions with regard to these materials, including but not limited to all implied warranties and conditions of merchantability, fitness for a particular purpose, title and non-infringement of any third party intellectual property right. TI and its respective suppliers and providers of content make no representations about the suitability of these materials for any purpose and disclaim all warranties and conditions with respect to these materials. No license, either express or implied, by estoppel or otherwise, is granted by TI. Use of the information on this site may require a license from a third party, or a license from TI.
TI is a global semiconductor design and manufacturing company. Innovate with 100,000+ analog ICs andembedded processors, along with software, tools and the industry’s largest sales/support staff.