RoadTest: ROHM 8-Channel Multi-Rail DC/DC Converter Board
Author: Gough Lui
Creation date:
Evaluation Type: Development Boards & Tools
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?: Various AEC-Q100 qualified buck converter ICs
What were the biggest problems encountered?: No jumper shunts provided, silkscreen could be more useful, minor documentation errors in labelling, board design makes evaluation of individual converters difficult and no price for the board available.
Detailed Review:
Rohm REFRPT001-EVK-001 8-Channel Power Tree Evaluation Board RoadTest
By Gough Lui – October/November 2021
Power conversion has often been an afterthought and clean power is something designers take for granted. Power-focused RoadTests have not attracted all that much interest, but as someone who has interest in power conversion and some relevant equipment, I decided to apply for this RoadTest even though I’m not in the automotive industry and am not designing any ADAS systems. This particular RoadTest poses some interesting challenges including the sheer number of power converters on this board! I proposed delivering a “monolithic” RoadTest report, so apologies for the length as there were a lot of interesting results!
Thanks to Rohm and element14 for selecting me as a RoadTester for this very niche evaluation board. This review was supposed to be released to coincide with the upgrade of the element14 Community platform, but it seems they've had a bit of a hiccup, so it's being released on the Jive platform instead. I hope this review is interesting, informative and insightful – if you liked it, please leave a like; if you have questions, leave a comment; and share it with others who might be interested.
The power requirements for modern electronics can be quite complicated, including ever-lower voltages, tighter regulation and higher efficiency requirements. In the automotive space, this is especially complicated by the need to ensure high reliability in harsh environmental conditions which include electromagnetic noise, voltage/temperature fluctuations and vibration. Keeping quiescent currents and electromagnetic emissions low is also important, as that can also mean less battery drain, better fuel economy and reduced likelihood of malfunction due to interference. The operating environment can be unforgiving, so parts used in automotive context often need to be qualified to Automotive Electronics Council (AEC-Q100, AEC-Q101, AEC-Q200, etc.) standards to ensure reliability.
The Rohm REFRPT001-EVK-001 is an evaluation board intended to evaluate a power-tree solution targeting infotainment displays and ADAS ECUs based around Rohm’s DC-DC buck converters, a low-dropout linear regulator, a MOSFET power switch and a pair of voltage supervisor ICs. This particular reference design is said to meet CISPR25 Class 5 with all converters operating, meaning stringent EMC control (although this is not something I will be testing).
This design is known as a “power tree” because it is arranged as a cascade of DC-DC converters. Two primary converters are present on the board – a BD9P105EFV-C providing 5V/1A and a BD9P205EFV-C providing 3.3V/2A. The secondary converters are “leaves” which take the output from a primary converter and further down-regulate it to a lower voltage. From the 5V rail, a BD9S201NUX-C provides 1.25V/2A and a BD00IA5MEFJ-M LDO provides 3.3V/0.5A. Loads can also be powered directly from this 5V rail, unswitched. From the 3.3V rail, a BD9S201NUX-C provides 1.8V/2A, a BD9S300MUF-C provides 1.5V/3A and a BD9S400MUF-C provides 1.0V/4A. Finally, an RV4C020ZPHZG MOSFET load switch allows for the 3.3V rail to be switched.
The power-tree configuration offers some key benefits to sensitive loads – having two stages of regulation improves the power-supply rejection ratio, as each stage has a high rejection ratio on its own and the capacitance between the two stages serves to buffer short transients as well. It also spreads the heat dissipation and could in some cases improve efficiency depending on careful converter choice and design. The major downside is increased component count and costs, and the likelihood that efficiency would be reduced as the power is transformed twice (e.g. a 95% efficient converter followed by a 95% efficient converter is effectively 90.25% efficient).
While the datasheet specifications for each of the converters is quite generous – for example, the primary converters can handle an absolute-maximum input of 42V, the board has not been designed for this with primary capacitors rated to 35V and the operating conditions table listing a supply voltage range of 9V to 16V. Similarly, while the secondary converters often have high current capabilities based on the datasheet, the rating for this evaluation board is more limited due to thermal constraints – the 1.25V rail provides up to 1.25A; the 3.3V LDO up to 0.2A, the 1.0V rail up to 1.5A, the 1.5V rail up to 1.0A; the 1.8V rail up to 0.5A; and the load-switch up to 0.15A. The primary converters can achieve their full ratings, assuming they are operating alone.
To ensure the power is within tolerance, two BD39040MUF-C voltage supervisor ICs are used. Each IC is capable of monitoring up to four outputs using resistive dividers and generates outputs that reset connected chips and indicate the power-good status of each rail. The supervisors can also perform watchdog timer functions, however, this is not used in this design.
All of the Rohm semiconductor solutions used on the evaluation kit are AEC-Q100 Grade 1 (all except the load-switch MOSFET) or AEC-Q101 qualified.
The board was packaged in a mostly-white-and-grey colour print cardboard box, with the striking red Rohm Semiconductor logo tastefully offset by grey traces of varying widths. Perhaps one of the nicer evaluation board boxes I have seen.
Inside, there is a single page leaflet with the necessary disclaimers and reminder to obtain information via their website. The grey foam is especially plush, but does not fill the void of the box completely, so the board is somewhat free to rattle inside the box. This is not ideal, but is unlikely to cause any permanent damage.
An ordinary metallised-foil static shielding bag wasn’t good enough for Rohm, so they opted for something akin to Velostat. Nothing about this unboxing has been typical so far!
The board itself is a high-quality four-layer construction which seems very clean of flux residue and having excellent soldering consistency. The evenness of the small stitching via through-holes gives me the feeling this is probably Japanese made and not inexpensive to manufacture, but will also serve the purpose of ensuring the lowest impedance to reduce noise radiation. Despite this, the common mode filter (CMF1) and primary 3.3V terminal block (P33) are not fitted, although they were not part of the documentation. There also seem to be some extra footprints for capacitors (CA1, CE1, C12, C22, C40, C43, C72) which may have provided even better EMI performance depending on the performance of the remaining bypass capacitors. Inputs and outputs are by 5.08mm terminal blocks, with gold-plated colour-coded test loops provided. The output of each voltage supervisor has also been thoughtfully arranged in an 8x2 2.54mm raised pin header that would be compatible with certain logic probes, however, unfortunately they did not fit the Rohde & Schwarz RT-ZL04 because of clearance issues around the connector.
The underside of the board has very few components – a few resistors and mostly thin, narrow traces that connect signals to their respective points. The majority of it forms a ground plane which is likely to help shield the noise generated by individual converters.
The P5V and P3V converters essentially use the same design with different values to achieve the required output voltage.
Likewise, the S1V and S1V5 outputs use a footprint-compatible chip from the same family and feature virtually identical layouts.
The S1V8 and S1V25 outputs use the same IC and same design as well. Just above the 1V25, part of the voltage supervisor indicator LEDs on the board are visible, with the supervisor responsible for P5V-derived rails having no fourth LED as there is no rail monitored in that position. The resistive dividers that provide the input are seen to the bottom of the image.
The voltage supervisor of the P3V outputs can be seen to have all LEDs populated and the resistive dividers above, as a mirror-image layout. Above this, the QL1 load-switch P-MOS transistor can be seen, with a smaller QL2 N-MOS acting to pull the gate of QL1 low. Finally is the S3V3 LDO which has the largest package of all.
While no included documentation about the board is provided with the kit, the documentation is available from the Rohm website and is quite comprehensive. The documentation consists of a User’s Guide with PCB layout, parts lists and schematic, an EMC Test Report to CISPR25 (although not officially certified) using the board in spread-spectrum, forced PWM mode and an Application Note which contains a full series of results regarding output ripple, conversion efficiency, regulation, quiescent current, transient response and thermal characteristics. Some of these experiments are repeated in this RoadTest, so this gives a useful measure to compare results. It is also nice to see that most of the component ICs have SPICE models, simulators, symbol/footprint and 3D models available.
The current user guide (63UF071E Rev.001 OCT.2020) and Application Note does have some subtle errors, such as the capitalisation of supervisor being “SuperVisor” in the block diagram, the part number of the MOSFET load switch being tabulated as RV4C20ZPHZG rather than RV4C020ZPHZG and the image having Monitor-terminal-1 and Monitor-terminal-2 labelled in reverse compared to the tabled signal assignments. While the naming convention of the rails seems good (i.e. P5V = primary 5V, S1V25 = secondary 1.25V), the exception is with P3V which should be P3V3 (as it’s a 3.3V rail, rather than 3V).
The design of the board also makes evaluation of individual converter performance quite awkward as there is no way to nicely isolate a single converter from the rest of the circuit, the consumption of indicator LEDs and the supervisor IC. While gold-plated test points are provided, they are relatively small and closely spaced which makes it tricky to use effectively. The arrangement of the selection jumper pins along the edge of the board is convenient, but it could have been done in a more compact manner while being labelled more intuitively so that the evaluator doesn’t have to refer to the documentation to check which pins correspond to which converter. The fact they are arranged to be read from the edge of the board, so that the top row of jumpers effectively appears upside down when the logo of the board is the right way up is unique, but perhaps not a good feature. Neither is the fact that for some converters, a mode of + means forced PWM (i.e. no energy saving) while in others, it means “SLLM” (i.e. simple light-load mode, energy saving) which can lead to confusion. The omission of the P3V output header is understandable given the documentation makes no reference to it, but it can prove useful for testing the primary 3.3V converter alone, so I would have preferred it to be fitted.
But perhaps my biggest gripe was that my board had no jumper caps supplied! That is a very unfriendly move, especially given how many are required to configure all the converters and their operating modes. I literally had to turn my room upside down to find precisely enough jumper shunts. This is not the first board I’ve received that didn’t have jumper shunts included, but it seems to be a silly omission, especially on this scale. Perhaps my board is an earlier edition, as the model number seems to be a label from a label-printer adhered to the board.
Tests of the whole solution quiescent current and voltage supervisor behaviour requited quite a few probes and connections to be made to the board.
Input was provided from a Keithley 2450 SMU and the input voltage was swept using KickStart 2 to produce an I-V curve of quiescent current. With the default configuration that had all converters in “force PWM” mode, the quiescent current ranged from about 30-44mA in the range of 8-16V. Using the power-saving modes on all converters which featured them dropped this down to 12-22mA which is a significant saving. The remaining power consumption is quite likely accounted in the power indication LED and the voltage supervisor output LEDs. Once the outputs were all disabled, the quiescent current of the whole solution became very low to the point that the current induced due to the voltage sweep charging/discharging capacitors dominated. Unfortunately, I did not slow-down the sweep to get a better measurement.
Note: Image Gallery Below. If it does not load, check your browser ad-block and enhanced tracking protection settings.
| {gallery} Supervisor Power On Behaviour |
|---|
|
S3V3 LDO: Power Up Behaviour |
|
3V3 LSW: Power Up Behaviour |
|
S1V25: Power Up Behaviour |
|
S1V0: Power Up Behaviour |
|
S1V5: Power Up Behaviour |
|
S1V8: Power Up Behaviour |
Each of the above screenshots from the Rohde & Schwarz RTM3004 Oscilloscope demonstrates the behaviour of the supervisor lines (as labelled) versus the input voltage (C1, yellow), P5V (C2, green), P3V (C3, orange) and channel under test (C4, purple). The supervisor behaviour across multiple power-ons is very consistent, resulting in an almost static-looking digital trace plot. The time taken for the output rails to come up varies, but were within about 13ms. The voltage supervisor seems to have various power good signals coming up in sequence a little after the rails become valid, although the XRSTOUT behaviour does not seem to make sense as it goes high first. Perhaps a reset is issued later and not captured in these traces.
Note: Image Gallery Below. If it does not load, check your browser ad-block and enhanced tracking protection settings.
| {gallery} Supervisor Power Down Behaviour |
|---|
|
Loss of Input: Power Down Behaviour |
|
P3V Disabled: Power Down Behaviour |
|
P5V Disabled: Power Down Behaviour |
Repeating this for a loss of input voltage shows that XRSTOUT goes low as soon as any power good fails. I also chose to disable each primary rail separately to see the impact it has on the outputs from that particular voltage supervisor.
One of the main tests in this review was to measure the per-converter efficiency and voltage regulation, but this required quite a bit of work to achieve.
While the evaluation board has a rather fancy name and niche application, at the end of the day, it is just a set of DC-to-DC converters and can be tested in the same way. As a result, my RoadTest proposal suggested that I modify the board to isolate each converter so that it can be tested individually and the performance of the overall board can be inferred from these experiments.
However, as noted previously, the board is not designed such that individual converters can be tested accurately, so I had to crack out my Dremel-clone and do some cutting and desoldering. This step permanently damages the board and I do not recommend doing this unless you really must.
Modifications included removing both voltage supervisor ICs and the associated resistor dividers for each rail to remove the quiescent draw that would occur. In the process, the debug headers were maimed by the hot air. The main power input terminal block was desoldered and relocated to the P33 header to provide a P3V output for testing purposes. The B- input was used as the input ground while the input Schottky diode was removed as testing would focus on the converter efficiency without the impact of these ancillaries. The input power trace was cut into half to provide a power input for the P3V and P5V converters separately. The “leaves” of the power tree were “pruned” to have the P5V output available on CAN5 and P3V output on P33. Solder resist was removed and cables attached to each of the secondary converters to allow for providing individual input, which caused quite a mess of connections to be made. It should be noted that the logic-high signal for each set of secondary converters either comes from P5V or P3V, so these were supplied separately to ensure that the converters would come up, but may have contributed a few “free” µA to the IC. In the case of converters with enable signals from P3V, the trace actually connects to a via near QL1.
This resulted in a rat’s nest of wires to allow for testing of each converter with four wire connections to the input and four-wire connections made to the corresponding terminal block for the output to minimise error due to resistance in the wires. Initially, I used spade terminals with a panel-mount banana socket, but I found it easier just to use 4mm “bullet” female terminals as they are vaguely banana plug compatible (although leaving some of the contact exposed). This was a much quicker and cheaper way to achieve the necessary connections.
To test the converters most accurately and expediently, I used a Keithley 2450 SMU as a highly accurate, although power-limited source, while loading the output using a Rohde & Schwarz NGU401 SMU (on-loan) as it is also highly accurate while having a greater current range, thus suiting the output of a buck converter more. I roped in my Rohde & Schwarz NGM202 Power Supply as well, since it was “in the stack” to provide the bias voltage for the enable/mode select lines, with all instruments being commanded through a pyvisa script over Ethernet. As these instruments are all fast, I was able to make unattended measurements at a high resolution within a reasonable time.
Primary converters were tested at a voltage input of 6V to 20V to fully cover the specified 9V to 16V range with some extra margin. While higher voltages may have been possible, the Keithley 2450 SMU can only deliver 100mA above 20V which was not sufficient. Secondary converters were tested at 2.8V to 5.5V to fully cover the operational envelope, noting converters were hesitant or failed to start up at 2.7V. Tests swept the load in 1mA steps, with voltages on primary converters being swept at 0.25V steps and secondary converters at 0.1V steps. Output current was swept from 1mA to a value above the solution rated current by about 10%-30% to provide additional coverage. To ensure higher accuracy, the Keithley 2450 was run in 10PLC mode with the Rohde & Schwarz NGU401 run in 5PLC mode (the closest matching value). At each step, input voltage, input current, output voltage and output current were recorded. Tests were repeated for each converter mode combination including spread spectrum on/off and power saving/force PWM mode. The LDO and load switch were not tested in this regime, as their losses are well understood by characterising quiescent current and Rds respectively.
As a result, the total number of test steps were a whopping 946,400 steps taking approximately close to 10 days to collect the requisite data. The resulting data was analysed in a mixture of Microsoft Excel and MATLAB to produce nice surface contour plots of efficiency and voltage stability.
Note that the following abbreviations are used throughout this review – NS = No Spread Spectrum, SS = Spread Spectrum, FP = Forced PWM Mode, AU = Automatic Mode, SL = Simple Light Load Mode, EN = Enabled, DIS = Disabled.
For extra versatility, the efficiency tests were performed in a “grid-scan” configuration allowing for efficiency surfaces to be derived for each converter. This allows for “what-if” scenarios to be explored and back-calculation of whole-system efficiency from different loading scenarios can also be performed.
To do this requires knowing the current load for each secondary converter, converting that into a power by multiplying it by the voltage, then dividing it by the measured efficiency of the converter at the relevant operating point to know the input power to the secondary converter. Summing up all secondary converter power requirements, this can be converted back into a current by dividing it by the primary converter voltage and the efficiency of the primary converter can then be read off the graph depending on its operating point. Dividing the secondary total power by the efficiency will give you the primary converter power requirement and dividing this by the operating voltage will give you the current. It may seem a little complicated, but it’s very useful in case you decide to run the S1V25 converter from the P3V rail for example, or instead you want to run everything off a single P5V converter.
Please be aware that contours on these plots are not at evenly spaced intervals to best provide detail across the range of values in a single plot. Please refer to the value on the contours themselves.
The P5V converter running in fixed PWM mode reaches 50% efficiency at fairly low loads of about 30mA which is a good result. At the nominal 12V, it achieves a peak efficiency of around 91.4% at around 650mA with improved efficiency as the voltage difference reduces (as is customary). The use of spread spectrum appears to have a nearly immeasurable effect (it could be just measurement error). The measurements match well with the results provided in the application note.
A clear threshold is seen around 170-200mA where the converter switches into the power saving mode which helps boost efficiency, keeping it quite high even at no load. Measurements of over 80% at extremely low load indicate just how effective this power saving mode is, however, there is a cost to this as will be seen later.
Testing of the P3V converter encountered a current envelope limitation with the Keithley 2450 SMU running out of current towards the upper load current area at the lower input voltages, resulting in the “staircase” of blank values in the top-left corner of all graphs. The trend here is similar with 50% efficiency being reached at relatively low current levels of about 50mA. Peak efficiency at 12V is about 85.4% at around 700mA loading which is a bit less than for the 5V solution but again matches very well with the application note’s single curve.
The threshold for switching over to power saving mode is around 400mA and similarly, excellent efficiency values are seen to very low levels of load. In all, the efficiency of the primary converters are not bad at all and the power saving mode has quite an impressive positive effect on the results.
The secondary S1V0 converter does not seem to reach as high of an efficiency level, however, this is perhaps to be expected in part due to the voltage difference, the very low voltage output and the size of the converter. In fixed PWM mode at the nominal 3.3V input that this board provides, the peak efficiency is about 82.7% at about 600mA load. The application note implies the peak should be about 500mA but this is a very small discrepancy and overall the trend is very much the same. Enabling the simple light load mode does provide a small efficiency improvement for some of the light load scenarios, but curiously, above 3.6V, this does not seem to function well.
The S1V5 converter is able to muster up a little more efficiency. In fixed PWM mode at the nominal 3.3V input, the peak efficiency was about 87.2% at a load of 600mA. The application note again seems to state the peak efficiency is at about 500mA but the trend is the same and the efficiency percentages are in agreement. Enabling light load mode on this converter had a more consistent improvement in efficiency down to extremely low currents. It is impressive just how wide of a “decent efficiency” band this converter can manage.
The S1V25 converter has a much “peakier” characteristic with regards to efficiency, having a narrower band of efficiency. At the nominal 5V input, the peak efficiency was measured about 84.6% at around 550mA load. This is in agreement with the application note.
Finally, the S1V8 converter does not seem to be as peaky, perhaps because the operating current range is more limited on the whole-solution level. At the nominal 3.3V input, it reaches a peak efficiency of about 91.2% at 350mA which also agrees with the application note.
Overall, voltage stability for all converters were excellent under load and mostly exhibited ohmic tendencies suggesting resistive losses in the PCB and terminal block were likely the major contributors to the observed small voltage droops under load. The automatic power saving mode of the primary converters did cause noticeable jumps in the output voltage at a threshold voltage while the SLLM mode of some secondary converters did not exhibit this behaviour. The results are very similar to that of the application note, however, the application note did not exhaustively test all combinations of operating parameters and thus did not detail this behaviour.
In the fixed PWM regime, the P5V converter achieves fairly good voltage stability, with a near linear drop in voltage of only about 10mV as the loading increased likely due to resistive losses in the PCB itself. There was no observable effect from the spread spectrum setting, although the gradient change may be due to temperature changes in the environment.
Enabling automatic power saving mode results in a noticeable discontinuity in voltage regulation where higher voltages are observed in the power saving mode, falling steeply in the change-over to a fixed PWM regime. The changeover point is about 180-200mA of load. Highly sensitive devices which need a very stable voltage may prefer fixed PWM mode despite the efficiency penalty to avoid this.
As the P3V converter is of the same IC as the P5V converter, the results are vaguely similar with more noticeable linear voltage drop-off with current increase of about 37mV in this test. This could be the reason why the P33 terminal was never fitted to the board. Again, spread spectrum had no major effect on voltage stability. Note the lack of test data in the upper-left corner of the graph due to the Keithley 2450 SMU reaching its 1A sourcing limits.
Enabling automatic power saving mode also results in a noticeable discontinuity in voltage regulation where higher voltages are observed in the power saving mode, falling steeply in the change-over to a fixed PWM regime. The changeover point is at about 400mA load.
In fixed PWM mode, about 7mV drop was observed across the full range of loading which is very acceptable and likely reflects resistive losses in the traces of the PCB and terminal contacts. Enabling simple light load mode (SLLM) did not change this behaviour in any noticeable way.
As the S1V5 rail is based on a similar IC to the S1V0 rail, a similar trend is seen with excellent voltage drop under load of just around 5mV and no effect from SLLM at all.
The S1V25 rail demonstrated around 10mV voltage drop across the load current, which is again not bad and likely indicative of resistive losses in the test PCB.
The S1V8 rail has the same architecture as the S1V25 rail and showed fairly good results as well, with a droop of around 5-7mV under load.
Per-converter quiescent current was measured using the Keithley 2450 SMU and the KickStart 2 software performing an I-V curve sweep. However, because the converters have significant capacitance on their inputs, the number of sweep points was more limited and instead, high capacitance mode was enabled at 10PLC with a 15s measuring delay to allow for the capacitors to do most of their charging/discharging at each voltage transition. The results, however, are occasionally affected by oscillation and may not be 100% accurate especially below the µA level because of the use of ordinary PVC wire and in an ordinary “noisy” home-office environment. Tests were performed in each of the operational and in non-operational modes and co-incident curves have been labelled to avoid confusion.
With the primary converters operating in fixed PWM mode, the quiescent current was around 7-10mA with a small change depending on input voltage.
When the automatic power saving mode is engaged, it seems that the quiescent current for the P5V converter drops to about 40 to 110µA. The situation for the P3V converter is a little more unclear due to oscillation, reaching a maximum of about 50µA but measuring below disabled levels of current at high voltages (but this is likely an artifact of the measurement technique).
With both converters disabled, the leakage current is quite small, in the 3-4µA region although some of that can be down to leaky insulation between wires and flux contamination on the board. It is so small to be difficult to measure accurately.
The secondary converters showed similar results where the controllers were from similar families. The S1V0 and S1V5 rails in fixed PWM had a quiescent current about 10 to 22mA. The S1V8 and S1V25 rails in the enabled mode had a quiescent current about 6 to 17mA.
Engaging the simple light load mode on the S1V0 and S1V5 rails drops the quiescent current down to 500 to 780µA. The S3V3 LDO when enabled has a quiescent current measuring about 250-350µA.
With the switching converters disabled, they all measured similarly with currents between 28-55µA which seems curious and potentially anomalous. The LDO had the lowest quiescent current of all, being essentially almost immeasurable given the limitations of my test set-up, which is the expectation.
While these figures may not match the claimed “0µA” typical or 10µA maximum shut-down value in the datasheet, part of the reason may be leakage in non-ideal components such as capacitors, but also potentially due to the test setup which may have some parasitic paths even though care has been taken to try and minimise them. Having a dedicated “per-converter” test board would have helped troubleshoot this further, but the values are not necessarily bad as they stand.
While the application notes provide oscilloscope screenshots from transient performance tests, there was a considerable lack of information on the transient itself and the measured performance, instead relying on one to interpret the result visually. In order to test this a little better, I relied on the Rohde & Schwarz RTM3004 Oscilloscope with a 10x passive probe and spring ground contact to measure the transient response with the transients supplied by the B&K Precision Model 8600 DC Electronic Load at a rate of 250Hz. The transient rise and fall times are recorded with peak-to-peak voltages at various transient slew rates.
The fastest transients had a rise time of about 9µs and a fall time of 11.4µs, oscillating between 0.1A and 0.9A and caused the P5V converter output voltage to swing around 271mV peak-to-peak (5.4%). Considering the speed of the transient, this is not a bad result as some other slower converters would have collapsed entirely to zero volts. A medium-speed transient was generated with a rise and fall time around 64µs which resulted in an output voltage swing of 192mV peak-to-peak (3.8%).
With automatic power saving mode enabled, the power saving mode can be seen to be activated on the low-level of the transient resulting in the “noisy” waveform at low current loading. It seems that entering and exiting low-power mode is essentially seamless, without any significantly greater dip or excursion compared to fixed PWM mode. The swing was about 263mV peak-to-peak (5.3%) on the fast transient and 193mV peak-to-peak on the medium transient (3.9%).
The fast transient had a rise and fall time approximately 11µs between 0.2A and 1.8A. This caused the P3V converter to swing 312mV peak-to-peak (9.5%) which is somewhat significant although the transient is very severe. A medium-speed transient with a rise and fall time around 128µs resulted in a more sensible 162mV peak-to-peak swing (4.9%). The difference in the transient speed is due to limitations in setting the electronic load.
With automatic power saving mode enabled, the swing was about 336mV peak-to-peak (10.2%) on the fast transient and 175mV peak-to-peak on the medium transient (5.3%).
As no effect was seen from enabling SLLM, only forced PWM results were recorded for the S1V0 converter. At the fastest transient I could generate between 0.15A and 1.35A, the rise and fall times were about 11µs which resulted in a 245mV peak-to-peak swing (24.5%) which is significant. It is quite hard to keep regulation tight when the output voltage is this small and the transients are this fast. The medium transient case has more relaxed rise and fall times in the 96µs range where the excursion was much more limited at 93mV peak-to-peak (9.3%) which is significant but much better. Improved performance is likely to be recorded for even slower transients.
As this output is based on a similar converter, SLLM results were not recorded as no difference was observed. The fastest transients between 0.1A and 0.9A measured at around 11µs rise and fall times resulted in 193mV peak-to-peak deviation (12.9%). For a medium-speed transient with about 64µs rise and fall time, this resulted in a more comfortable 94mV peak-to-peak excursion (6.3%). Again, these are very strenuous and demanding transients – real applications may not see as fast rise or fall times.
For the S1V25 converter, the transients oscillated between 0.125A to 1.125A. The fastest transients had a rise time of about 9.5µs and fall times of about 11.4µs. The peak-to-peak perturbation measured 180mV (14.4%). The medium-speed transient has a rise and fall time about 78µs and resulted in just 57mV peak-to-peak movement (4.6%).
Finally, the S1V8 converter was subjected to a 50mA to 450mA transient. The fastest transients had rise times of 9.3µs and fall times of 13.4µs resulting in a 108mV peak-to-peak deviation (6%). A medium transient had rise and fall times around 31µs reduced this to just 48mV peak-to-peak (2.7%).
If it’s any redeeming factor, while the excursions in many of these fast-transient tests may seem sizeable, they are only present for a very exceedingly short amount of time – around 12.5µs in this case. Perhaps the load wouldn’t even notice given a correctly-sized local decoupling capacitor.
While the converters had excellent transient response performance overall considering the harshness of the transients I threw at them (which can collapse general purpose lab power supplies quite easily), I also did attempt to measure line influences, however, the converters were so quick that I could not really discern the effect of an induced line transient (e.g. a primary converter seeing the input jump from 9V to 18V). As a result, the results of these tests are not reported.
As I have done many Rds versus Vgs plots before, I decided to do the same with the load-switch MOSFET on the board. For this, the roles were reversed with the Rohde & Schwarz NGU401 providing the drain-source current and the Keithley 2450 SMU measuring the drain-source voltage and the Rohde & Schwarz NGM202 providing the gate-source voltage. As I have reversed the connections so as the numbers reported are positive rather than negative, the gate voltage is swept from 8V down to 0V in 1mV increments while power dissipation is kept limited to 1.5W as rated. The source current is stepped 100mA at a time from 0.1A through to 2A as rated.
For more accurate tests, capacitors either side were removed and the gate connection was severed by removing the resistors that connected it to an N-MOS acting as a driver. Unfortunately, as the test began, it was clear that the dremelling did the MOSFET no good and it seems it has been ESD damaged. The resulting MOSFET was leaky, acting as if there was a parallel moderate-resistance shunt allowing current through at all times even without gate voltage applied. Despite this, the test was still run.
The shape of the curve shows a characteristic very similar to the datasheet typical and below the maximum. As the channel is pinched off, the resistance rises slower due to the ESD damage to this MOSFET, but it seems very likely that the MOSFET was as good as, if not better than, the datasheet specifications. A key feature is the very sensitive gate which allows for easy logic-level control.
The per-converter ripple and noise was tested by supplying the converter with its nominal rated voltage from the Keithley 2450 SMU and loading each converter at the user-manual whole-solution rating using the Rohde & Schwarz NGU401 while the voltage output was measured using a Rohde & Schwarz RTM3004 at full bandwidth (500MHz) using a 10:1 passive probe with spring ground. Tests were done with both spread spectrum options, and FFT spectrograms were also collected.
With the P5V converter operating in fixed PWM with spread spectrum, there is a visible periodic noise that is dithered, resulting in a loaded 1A ripple figure of 26mV peak-to-peak which is quite good. The pattern and result seem quite different from the application note, which could be due to differences in the test setup.
The effect of the spread spectrum mode can be seen, with the 2.2MHz switching frequency peak and sidebands being dithered down into a band of emissions between 2.15MHz and 2.38MHz. There is also a lower frequency noise around 360kHz which may be related to spread spectrum operation or it could be emanating from some of my test equipment, but I did not investigate.
With the P3V converter also operating in the same mode, at a load of 2A, the ripple measured 22mV peak-to-peak which is also quite good.
A similar FFT behaviour can be seen for the switching frequency, but there is a broad peak about 120kHz and a peak near 2.6MHz as well.
I mentioned earlier that the power saving modes come with a bit of a downside. Aside from a noticeable threshold where the voltage suddenly changes, another result is that the ripple and noise figures change dramatically due to discontinuous operation. Forced PWM at no load results in a nice 14mV peak-to-peak noise, but this increases to 45mV peak-to-peak in auto mode with a noticeable sawtooth waveform at a frequency about 220Hz.
The ripple at 1.5A load in fixed PWM mode measured 11mV peak-to-peak which is excellent. A wider-band FFT shows the main emission is at about 2.2MHz with some harmonically related peaks visible.
For the S1V5 converter at 1A in fixed PWM mode, the ripple measured 12mV peak-to-peak which is yet another excellent result. Similarly to before, the main emission appears to be at the 2.2MHz switching frequency with very few spurs (mostly in the lower end).
The simple light load mode shares exactly the same disadvantage as the auto mode under no load – it results in a sawtooth waveform and a greater ripple, although in this case it was just 17mV peak-to-peak unloaded. The frequency is about 450Hz in this instance with some jitter. The resulting FFT in this mode shows no major spurs but does have a band of emissions around 200kHz.
The S1V25 converter was tested at 1.25A and exhibited a 19mV peak-to-peak ripple which is quite good. Harmonically related spurs to the 2.2MHz switching frequency are visible in the FFT.
The 1V8 converter is tested at 500mA and had a 12mV peak-to-peak ripple which is excellent. The FFT result is very much similar to the S1V25 result.
On the whole, all converters exhibited low ripple voltages, relatively speaking. All appeared to have switching frequencies of about 2.2MHz with spread spectrum spreading this energy across a band to reduce emission intensity. Many converters showed harmonically related spurs to this switching frequency, but there was also some broad peaks and FSK-like emissions lower down although whether they originated from the IC or elsewhere in the test chain have not been determined. The behaviour of the discontinuous light-load operation is also shown to create sawtooth waveforms with relatively low frequencies.
As a final check of the performance of the over-current protection (OCP) and the voltage regulation beyond the whole-solution limits given in the User Manual, I decided to use the Rohde & Schwarz NGU401 SMU as a load and the Rohde & Schwarz NGM202 as a source to ensure we were not limited by the sourcing capability of the more accurate Keithley 2450 SMU used earlier for efficiency tests. All connections to both input and output were made using four-wire connections, although the outputs were tested at the point of the terminal block, so any resistive losses within the PCB would be captured as a slight voltage drop with increasing current.
In all cases, no converters were damaged and the output voltage declined almost linearly with load suggesting resistive losses. In most cases, the voltage declined first before the converter started to “hiccup” in overcurrent protection. The OCP thresholds were all above the rated operating current by a variable amount – the 2A-capable P3V rail actually managed around 3.25A before showing signs of stress, while the 1A-capable P5V rail managed a more sedate 1.2A. In this swept test, the S3V3 LDO managed about 1.2A of a 0.5A rating as well, although I suspect under continuous load, the accumulated heat would cause OTP to shut-down the converter.
The Rohm Semiconductor REFRPT001-EVK-001 Evaluation Kit is targeted at automotive ADAS, infotainment and ECU applications and demonstrates a power tree reference design using AEC-qualified components which is claimed to be CISPR25 Class 5 compliant under full rated load. The kit itself consists of a high-quality, well-manufactured four-layer board featuring a total of six switching converters, one LDO, one MOSFET load switch and two voltage supervisor ICs with supporting ancillaries. The design of the board features gold-plated test points, many configuration jumper pins and a logic-probe compatible header for monitoring. The board is supported by documentation including a User’s Manual, Application Note and EMC Test Report with PCB layouts and BOM lists also available.
While the design intentions seem good, the execution of the design leaves some things to be desired. As there is no easy way to isolate converters, testing individual converters requires potentially destructive modifications to the board. The logic probe and test point connections are a bit too close to other elements and the jumper pins are oriented to be viewed from the edge of the board on all sides resulting in an unusual orientation for the top row as compared to other evaluation boards. The silkscreen labelling could have been improved to identify which jumper sets belong to which converter and indication of default values would have also been appreciated. Inconsistency with whether a + setting activates force PWM versus energy saving is also a frustration. The omission of jumper shunts for my board also posed a very specific inconvenience which is compounded by the fact that there are so many configuration options. Perhaps for someone evaluating all rails and is more concerned with EMC, this design would be adequate, but for those who are power-centric, the design poses some challenges. Documentation also showed some inconsistencies in part numbering and mislabelling of the monitor terminals on the photograph, adding to confusion. The choice to label the primary 3.3V rail as P3V is perhaps inconsistent with convention which would otherwise suggest it should be labelled P3V3.
However, that being said, the tested converters were surprisingly good. Voltage regulation was excellent under load considering that the residual voltage changes may originate from resistive losses in the PCB although power saving modes do create some voltage discontinuities. Efficiency was good to excellent overall, with the light-load modes proving to be really effective even down to single-digit mA draws. Transient response when challenged with fast transients did not upset the converters, although voltage excursions may have been significant at times, they were also very short which limits their impact. Ripple voltage was almost universally low, increasing slightly in power saving modes. Attempts to measure line regulation did not provoke any output deviation and thus were not recorded. Analysis of the switching frequency shows the 2.2MHz frequency in the output with some related harmonics. For the converters supporting spread spectrum, there appeared to be no penalty in performance but the noise energy was spread as intended. Quiescent current for all ICs was excellent, verging on immeasurable with my set-up when the ICs were disabled. Most of the measurements corroborated well with the application note results, however, it should be noted that the application note is a little brief on some measurements. The only concern, if there was one, is the slight inconsistency with overcurrent protection thresholds.
In the end, the converters themselves are excellent in performance and the solution size for each converter is quite compact. Unfortunately, there is no price on evaluation kit itself but the individual parts seem to be fairly priced given they are automotive-qualified. This was a rather challenging and comprehensive RoadTest, so I hope you enjoyed the result and thanks again for selecting me as a RoadTester for this board.
Top Comments
Very impressive RoadTest. You really made good use of the SMU and I particularly liked the efficiency plots.
Very good road test report.
DAB
Thanks fmilburn!
Indeed, the SMUs were extremely handy in terms of their speed and accuracy, meaning the plots had fewer jitters and took less time to measure compared to using conventional instruments…