Author: Gough Lui
Evaluation Type: Evaluation Boards
Did you receive all parts the manufacturer stated would be included in the package?: True
What other parts do you consider comparable to this product?: N/A
What were the biggest problems encountered?: None encountered.
RoadTest Review of Texas Instruments TPS56C215EVM-762 Evaluation Module
Thanks to element14 and Texas Instruments for selecting me as one of the RoadTesters for the TPS56C215EVM-762 Evaluation Module. I saw that this particular RoadTest didn't attract too many applicants, so I thought I'd give it a shot seeing as I also managed to snag the B&K Precision Model 8600 Electronic Load which would be very useful in trying out this product, and surprisingly, I was selected. Because of the delays in postage on both RoadTests, it didn't quite arrive in the timeframe I had planned, and thus time ran a bit short as I had other commitments as well. That being said, I took some extra time today to try and round it out, but as you will see later, in reality, this isn't an easy product to make a RoadTest review on.
When building designs involving high performance microprocessors, FPGAs and data converters, power supply management is definitely one of the many issues that need to be considered. Such loads often require high currents at low voltages with wildly swinging current demands due to the high-speed digital nature of operation, an area that more "basic" power converters struggle with.
In many cases, the requirement of the silicon is a voltage less than the supply voltage (e.g. 5V, 3.3V, 2.5V, 1.8V, 1.2V, 0.8V etc.). This necessitates using a buck switching converter, for better efficiency over a linear regulator. Basic converters that I've dealt with as a hobbyist generally don't perform too well at low voltages, and have significant trouble maintaining a stable output. Many of these simple designs are also fairly bulky, running on lower switching frequencies and thus requiring large magnetics as well.
While the premise of switching converters is not new, the control algorithms and integrated circuit fabrication and design have moved on. This evaluation module (EVM) is based around Texas Instrument's TPS56C215 4.5V to 17V Input, 12-A Synchronous Step-Down SWIFT(tm) Converter. This is a 3.5mm x 3.5mm VQFN-18 SMD chip featuring adaptive-on-time D-CAP3 control mode with selectable Forced Continuous Conduction Mode (FCCM) or Auto-Skipping Eco-mode(tm) for high light-load efficiency. Other features include a 0.6v to 5.5v output voltage range, selectable switching frequency of 400khz, 800khz or 1.2Mhz, adjustable current limit settings, adjustable soft-start, enable pin, power-good signal and efficiencies up to 94% depending on the configuration. The chip also has overtemperature protection and undervoltage lockout, with output under/overvoltage protection features. It integrates the switching MOSFET within the package, resulting in only in inductors, capacitors and resistors required as the external support components.
If you were interested in designing using the chip, you could order a few sample chips, build your own design using the datasheet and test it out. But that's not always easy to do and get right. It's often a problem with more sophisticated ICs, as their performance can be very dependent on component choice and component layout. So rather than start off with just the datasheet, that's where the TPS56C215EVM-762 Evaluation Module comes in. Think of it as a starting point given to you from TI's engineers - they have designed a simple board to show off just what their chip can do. Often the designs for such EVMs are close to optimal, and the design notes with full bill of materials is provided. The boards are also set-up with test-points to help you analyze how well the unit is performing under your test conditions.
This particular module is being sold by element14 at AU$89.43 at the time of posting. While it might sound strange that you essentially are "buying" a board that's trying to "sell" you their product, this could save a lot of time and headaches for designers especially if they find out early-on that the product isn't suitable for them, and it helps them understand what the expected performance would be if they integrate the chip into their design. Of course, people are still free to "go it alone" and work straight from the datasheet as was common practice.
It's important to remember though, that the EVM is not a finished product. It is not designed to be used as a product, as it hasn't undergone any certifications for safety or EMI compatibility. It's also not designed to be incorporated into finished products - so it's not there for you just to "plug in" the board into your final design. It's merely an evaluation board, for test purposes only.
As there is a wide variety of power converter chips on the market, each with its own appeal, I have not surveyed the market this time around. There are just too many options and intricacies involved, and paper specifications aren't always that applicable, as we will see later.
The unit comes in a large cardboard box which seems to be common with their other evaluation modules.
The spine of the box indicates which particular module is contained.
Opening the box, we have some leaflets on the top with foam egg-crate style packing keeping everything safe.
The documentation is merely some disclaimers about the EVM itself.
The EVM is fairly small and light and comes packed inside an ESD shielding bag.
The board itself feels of excellent build quality. I know most people like shiny things, but this board is distinctly matte and heavy. By heavy, I mean, this is a four layer board with 2oz copper. It's physically heavier and sturdier than most boards I've dealt with, and this is most probably to optimize its thermal performance. Connections to be board are made with terminal blocks that can be screwed into, but in my experience you should take care to support the blocks as they are tightened or else they will strain the solder connections to the board.
The underside of the board does not have any components at all.
The footprint of the chip itself is an impressive 3.5mm x 3.5mm which makes it almost as big as the capacitors that are used to filter the input and output. Considering this includes the switching MOSFET, this is pretty impressive.
The whole layout is not very big, and part of the reason is due to the high switching frequency which allows for the use of smaller inductors. Ceramic capacitors are used to keep up with the high switching frequency. A jumper is provided to toggle the enable pin.
Test points are provided as soldered down loops, which are good for test clips that can hook onto them. Unfortunately, in my experience, some of these hooks are a little close to capacitors or other components which requires some careful positioning to avoid short circuits.
The board itself is of good physical build quality, and the parts used seem reasonable. The design of the board appears to optimize the performance thermally, which might not be realistic in some end user scenarios. However, the important things such as the documentation for the board and the chip itself are both well done, with an ample amount of information about the performance and design principles which should help anyone intending to design with this chip. There are some interesting minor inconsistencies - such as the board claiming a capacity of 12A in some places, and 10A in others, whereas the chip could do up to 14A although not continuously at the risk of reduced lifetime due to electromigration in the solder.
However, hooking it up and powering up was very easy, and the unit worked immediately. Because of the low voltage and high current, remote sensing was used on the load to ensure an accurate voltage read-out. The voltage wasn't quite spot-on, but this was not unexpected since the voltage setpoint is determined by two resistors (R7 and R9) and their tolerances.
I tested the conversion efficiency using my Manson HCS-3102 power supply that can supply power to a voltage resolution of 0.1V and current read-out with a resolution of 0.01A. I had intended to use a higher specification power supply, however, as the EVM requires a power supply with 4A capability or greater, the Keysight E36103A was excluded. The results obtained by stepping the Iout on the B&K Model 8600 to values between 0.5A to 12A, while recording the current drawn for voltages of 4.5, 5, 5.5, 6, 7.5, 9, 12, 15 and 17V.
The graph is necessarily not that accurate, however, conversion efficiency was very similar to the graphs in the datasheet, slightly exceeding it at times. My results are to be considered preliminary due to the limitations in the metering especially voltage/current on the input. I did not choose to add more accurate metering, as putting in a current meter in the supply loop could affect the performance of the converter due to the nature of the load.
I decided to try and test the various protections on the board. For overcurrent protection, I raised the current of the load as high as possible before it cut out - it seemed possible to extract about 14.9A maximum from the board before it shut down, with no damage even after repeated attempts.
Running the board at such a high load resulted in the board shutting down as well. Using an IR thermometer, the outer casing of the chip was approximately 85 degrees C, and I believe the unit may have tripped its internal thermal shutdown limit. No damage resulted, although it's important to note that the EVM uses 2oz copper and a four layer board with large heat-spreading areas to optimize the thermal operating conditions. The board surface itself measured about 55 degrees C. It seems likely for a more mass produced product with a denser layout that the full ~12A current may not be sustainable because of thermal limitations. That being said, the unit does seem optimized for about 5A of load, so it seems wise to design around that point instead. For the EVM, it seems that a level of 14.2A was sustainable even with an input voltage of 4.5V (lowest recommended operating voltage).
The undervoltage lockout is left as per the default configuration, and the board found to energize ramping the input voltage upwards at 4.3V, and shut-down with the voltage ramping downwards at 3.6V. This is fairly close to the expected values, with the UVLO being configurable as well, which will be necessary in case of higher output voltages.
Quiescent current was tested with the output terminals unloaded, and the supply voltage varied while the current into the board was measured by the Keithley Model 2110 5.5-digit DMM. The results show about 0.65mA of quiescent current which is fairly stable at the higher voltages.
|Voltage (V)||Quiescent Current (mA)|
Regulation is one of the key difficulties when it comes to switching supplies, as it ultimately depends on how fast and accurate the feedback within the converter logic works. I started testing load regulation, and unlike the datasheet, I went for the full torture test.
I started with the most severe test conditions - alternating 0 and 12A load at a rate of 200Hz at 10% duty cycle, with a 5A/µS slew rate. Ultimately, it did show that the voltage did waver 200mV or so, but this is quite good when considering the power supply that is running this board can't handle that sort of slew rate anyway. In reality, the actual performance may also depend on the inductance of the wires (kept as short as possible), the capacitance on the output, etc. However, it's still a good result, and one which most devices can probably handle (~11% undervoltage at the low point). It also shows that the OCP wasn't being too sensitive and prematurely cutting-off the output, as I have seen with some lab supplies.
Being a little more gentle, now I'm using a 0.5A/µS slew rate, and the voltage only wavered about 115mV this time.
At an even more gentle 0.05A/µS, the 12A current change only resulted in a voltage deviation of about 72mV peak to peak. It's important to remember this includes any noise in the switching converter output as well. In all, this performance was much better than I thought was possible. It's always nice to be surprised this way.
As for line regulation, I measured the output voltage using my Keithley Model 2110 5.5-digit DMM under zero load and 12A load under a range of voltages:
|Voltage (V)||Zero Load (V)||12A Load (V)|
These measurements seem to show the voltage is fairly close regardless of input voltage, although under load, the voltage is slightly higher than it is without load. In all, that's still exemplary performance, especially considered that the output voltage is set by resistors, so the absolute voltage isn't really the factor of merit.
In order to supply other than 1.2V, a change to the R7 value is required. The formula is given in 1.3.1 of the documentation as Rtop = Rbot * (Vout - Vref) / Vref. As a result, for 5V out, a 73.3kΩ resistor is required.
Unfortunately, the EVM uses SMD resistors and in fairly close proximity to other components, which isn't as friendly as it could be for modification. As I had no 10-turn precision trimpots available, I just grabbed a regular 100kΩ trimpot and soldered it in on leads rather crudely.
The modification worked, and it was possible to obtain 5V output (or close enough) at 12A for 60W from such a small converter. Very impressive. Below are images at various input voltages - namely at 9.5V, my 5A power supply was at its absolute limit providing 7A.
In the case of a 5V input, the output was very unstable. Because I did not attempt to modify the UVLO by populating R1 and R2, the UVLO is defaulted to its ~3.6-4.3V level, thus is too low to prevent the converter attempting to start up when the voltage is insufficient.
An input of about 7.3V was necessary for a 5V stable output at 2A loading, or roughly 2.3V overhead.
The Texas Instruments TPS56C215 4.5V to 17V Input, 12-A Synchronous Step-Down SWIFT(tm) Converter was quite an interesting converter. In many ways, it had exceeded what I thought possible - for the size, for the current, and for the transient conditions tested. It shows just how far progress has been made within the realm of switching converter control and integrated circuit design and fabrication. The converter itself is a small 3.5mm x 3.5mm chip requiring only passives externally, with an integrated MOSFET and featuring OVP, OCP, OTP, UVLO, power-good output, enable input with selectable FCCM/cycle-skip operation modes and switching frequency.
In testing, it could not really be faulted at all, and even when pushed to its limits, no permanent damage occurred. Protections operated satisfactorily, and the efficiency of the converter was very good to excellent. I did not seek to evaluate customization of the UVLO parameters, nor checked the enable/disable performance or the power good output signal.
The finer points of the performance are really not that important, as the final performance obtained will depend on your board design, IC configuration and component choices. However, the EVM board itself appears to be well designed, well built from quality parts with a quality feel. I suppose that is expected, as this board is there to help sell their IC, and I think it does a great job of it. Just be a little careful with the test-points and screwing in the terminal blocks.
More importantly though, is the quality of documentation. As a designer, these boards are only useful for testing - when it comes to design, you rely on the design notes, bill of materials, layout information and specification data. Texas Instruments has always been good about this, and their IC datasheet and EVM datasheet both have an extensive array of comprehensive information sufficient for designers.
I think that this, among other TI chips, is definitely one worth considering if you're designing a device that has some silicon with very stringent power quality needs and a fairly sizeable load. It's not a chip that a hobbyist can readily use due to the SMD nature of it, however, its small footprint and integration means that devices do not need to dedicate large areas of board to providing a high quality power supply, enabling smaller devices.
I thank element14 and Texas Instruments again for choosing me to perform this RoadTest review, and thank them for supporting the community. For the latest on what I'm doing, feel free to visit my personal blog at http://goughlui.com, and have a happy new year.
Nice writeup and yes it is an impressive little chip.
Good review. It was particularly interesting seeing the efficiency for 12V as a contrast to what I was seeing with the 2-channel, switched-capacitor, 2MHz-switching board that Jan Cumps gave me to look…
Don't worry about the measurement if you've packed everything away. I'm just interested out of curiosity - how quickly the loop responds (a function of its bandwidth) and how it manages the climb back…
Don't worry about the measurement if you've packed everything away. I'm just interested out of curiosity - how quickly the loop responds (a function of its bandwidth) and how it manages the climb back (what the compensation is like, as viewed through the damping - the converter I looked at was a touch underdamped, agile but a bit skittish [like a greyhound racing down a track]). It was interesting to me, with the board I looked at, that it was less assured coming off load. I suppose that makes sense because it would have to use up what energy remained in the coils - ie it would coast, slightly out of control, for a moment or two. In the case of your board it would be even more complicated because it would also be managing the changeover to whatever the selected low-current regime was.
I think you're safe to use the AC coupling. Droop is only going to be a problem at low frequencies; worry about it if you're trying to judge how flat the top of a 50Hz squarewave is, don't worry at all with a transient that lasts for several tens of uS.
Thanks for the comment and apologies for the time taken in reply. It's been a busy time for me - but I do agree that might be a good thing, but sadly things have been dismantled and time doesn't permit at this stage to do it.
One thing I was aware of, and forgot to highlight, is that such waveforms are likely to be highly dependent on the amount and type (ESR) of the capacitance used in the final design, and the inductance of the connecting wire is likely to have an effect as well. As this is designed as a starting point for implementation ideas/fine tuning, the performance of the board itself might not match the performance of your design employing the same IC due to differences in trace layout, capacitor brand, etc.
To that end, I felt that with my limited 8-bit resolution 25Mhz scope (i.e. 256 step), just having the peak-to-peak voltages read out at the bottom would be enough. The scaling is rather awkward and the detail is somewhat lost, but with 8-bits to start with, the voltage resolution isn't really there although maybe zooming into the front or back of the transient waveform would have helped. You can already see the one pixel "flat" areas which blip up and down - that is the actual limitation in the scope measurement itself rather than the size of the window or display. Admittedly, I did try to do that but I then deleted the files as I thought it would overload the review with images (as I have a tendency to do this all the time).
I could have used AC coupling to knock the range down a little to make more use of the 8-bits I have available, but not knowing what the frequency response of the unit is when I'm in AC coupling mode, I didn't want to draw potentially inaccurate conclusions based on a reading I don't fully understand. As far as I'm aware, most AC coupling modes put a capacitor in-series with the probe input, and have a frequency response that could cause transients to appear differently in magnitude or shape.
Thanks for the great review. I found it instructional and well written.
I liked the graph on efficiency comparisons at different voltage -in's.
I appreciate your efforts.
RoadTest Program Lead
Good review. It was particularly interesting seeing the efficiency for 12V as a contrast to what I was seeing with the 2-channel, switched-capacitor, 2MHz-switching board that Jan Cumps gave me to look at (not too dissimilar).
Jan was a bit critical of the efficiency of that board but this shows that, with the large step down from 12V to 1.2V, and a high switching frequency, where the switching losses are inevitably higher, it wasn't too disgraceful at all.
I would have liked to have seen the transient waveforms in more detail. I can hardly see what's happening from your traces. (It's a little blip a few pixels high.)
Nice writeup and yes it is an impressive little chip.