RoadTest: Review the LinkSwitch-TN2Q Non-Isolated Buck Power Supply RDK-707Q
Author: Gough Lui
Creation date:
Evaluation Type: Semiconductors
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?: I did not find a direct match, but the LT8315 has similar voltage range and no-opto, but is a flyback isolated topology with a lower output current. The MP9488 has a lower voltage range and is no-opto with a buck topology, but also has a lower output current. Neither appear to be AEC-Q100 qualified for automotive applications.
What were the biggest problems encountered?: Board had insidious failure of some sort, invalidating all results. High-voltage DC testing is quite difficult - combining multiple power supplies can have its own downsides.
Detailed Review:
By Gough Lui – November 2024
There’s no denying that electric vehicles (EVs) are not just part of our future – they’re part of our everyday. On my way to work, five years ago, I’d be lucky to see just one. Today, I’m seeing more than ten. It’s a truly exciting development. Those who don’t own one themselves (like myself) might be surprised to know that, despite having a large, high-voltage traction battery, they also have a lower-voltage auxiliary battery to power many of the critical control systems and accessories, taking the form of a familiar 12V sealed lead-acid battery that is recharged from the traction battery when the car is running.
The limited capacity, charging rate and lifetime of these auxiliary batteries limit the amount of load that can be placed on them, so heavier non-critical loads such as heating/cooling tend to be driven directly from the traction battery which is available when the car is running. But there may be other loads which may be worth migrating to the traction battery side – the major complication being that traction batteries frequently range between 200V to 800V DC with a majority at 400V today.
The Power Integrations LinkSwitch-TN2Q family targets such automotive applications. Being somewhat of a unique device, this RoadTest actually began with a direct question from the organisers which culminated in my application being submitted.
Thanks to element14 and Power Integrations for the opportunity to undertake this challenging RoadTest.
You may not be familiar with Power Integrations, but I’d guess that the majority of people might actually be using a Power Integrations-enabled product without knowing it. I first remember encountering their products in USB-C chargers, both GaN and traditional Si-based units. The Power Integrations InnoSwitch family and PowiGaN technologies are a favourite amongst even lower-cost solutions on the market for their good performance, reasonable cost, compact size and simplicity of implementation with a single-chip that does almost everything. Unless you’ve taken apart a few USB-C chargers of your own, you probably wouldn’t know just how Power Integrations might have already been part of powering your life … but I digress.
As mentioned in the opening paragraph, EVs have an auxiliary 12V sealed lead-acid battery for powering critical loads and accessories, but this comes with downsides as the battery has limited capacity, charging rate and lifetime. Some loads just make better sense to run from the traction battery as they are only required when the car is running, but the challenge is that traction batteries frequently range between 200V to 800V DC with the majority around 400V nominal. The big difference between voltage usually poses challenges for switching converters in achieving high efficiencies and the automotive operating environment can be harsh with vibration, electrical noise, high temperatures and long operating lifetime requirements.
The Power Integrations LinkSwitch-TN2Q family targets such automotive applications, offering a switching controller capable of 30V to 550V DC input, 270mA-850mA output depending on model, configurable in buck, buck-boost and flyback topologies, operating at 66kHz with frequency jittering and AEC-Q100 qualification. The chip itself is optimised with a minimal bill-of-materials (BOM) requirement, low no-load consumption below 50mW, protections including OVP, OTP, SCP and integrated 750V MOSFET for surge withstand.
For testing, they have provided the (already, fully qualified) RDR-707Q reference design board, which is configured as a non-isolated 9.75W buck power supply with 30V - 550V DC input and 15V 650mA output. This is built around the LNK3209GQ, the higher-current variant of the LinkSwitch-TN2Q family.
The solution itself has a minimal BOM count as the majority of the work is handled within the single chip – no external MOSFET, no feedback optocoupler are required. This simplifies the solution and allows it to be more compact and reliable. As higher voltages are being switched, the peak inverse voltage can get quite high, so two diodes are used in series in some places to reach the necessary voltage withstand without compromising on reverse recovery time.
While intended for automotive, this does seem like something that could be just at home on the mains as well – provided you whack in a bridge rectifier and filter capacitor to turn the AC voltage into DC. Doing so may actually yield a non-isolated low-voltage DC supply with lower standby currents than achieved using crude capacitive dropper and Zener-shunt designs. However, being a non-isolated design, this means that the output ground is directly connected to input ground which could be a safety concern if a user makes contact with the output – so these are best used for powering self-contained systems with no direct user contact.
The board arrived inside a box with a colour cardboard wrap-around. The design indicates the product contained inside, but doesn’t actually show an image of it. I suppose that’s fine given this is not going to be a retail product. Information is available online and a QR code is provided, linking you to https://qrs.ly/c2e66hn (likely a link-shortener for tracking and redirection purposes).
The box inside contains the board, packaged inside a static shielding bag.
My board is serial 141 and is pre-manufactured on a 4-layer green PCB. The size of the solution, in the white square, is indeed rather compact (a millimetre scale is shown above). However, a downside of that compactness is simply the proximity of all the test points to one another, making connections slightly tricky to make without shorting out. The design of the board not having any additional sense connections is unfortunate as well, as this would make for less accurate voltage measurements. The outer rim of the board has six screw holes with only dielectric, intended for mounting to a heatsink with a thermal pad for higher-temperature tests. The components on the board are apparently AEC-Q100 qualified, and all of the components are SMD with the exception of the test-points from Keystone. There is a lot of via stitching which is all solder-filled for better heat dissipation – a good idea but perhaps not one that will make many PCB manufacturers happy (small drill bits wear out quickly).
From the underside, it seems this board was manufactured Week 38 of 2022 by Unice.
Looking at the side profile, we can get an appreciation of the height of the components. The Coilcraft inductor is the tallest item on the board along with the output polymer electrolytic. But one shouldn’t underestimate the size and bulk of the input ceramic capacitor as well – one that has to handle 550V+. It is nice that they also included a Bourns 1.25A fuse on the board just to protect against catastrophic failure. I certainly hope I don’t come to rely on that!
From the outset, I knew this particular RoadTest would prove to be a challenge. With a board that has a 30V - 550V DC input range, there would be little point in reviewing the board if I were to stick to the “safe” low-voltage DC regime of up to 60V. Furthermore, even my SMU can only reach 200V at the most, which still is far from the 400V nominal of most EVs today, and that’s not considering a number of other technical issues. I could use rectified and filtered mains, but the presence of 50Hz ripple may affect the results and would be very difficult to automate tests. Another option would be to use a bank of batteries … perhaps even having a string of 9V batteries in series, but that would get expensive and they would not hold their voltage as they get consumed during testing. I would have to get creative … so this chapter focuses on the various preparations which took around a month to complete.
The test report suggests the above test set-up, which is something I hope to replicate as closely as possible. However, the value of the bulk capacitance is not actually detailed in the report to my knowledge, and while its intention is to stimulate the “stiff” traction battery supply, if it were truly necessary for the operation of the converter then it should be part of the solution, so I decided to omit it. Another reason is that most DC supply outputs have some level of capacitance already and handling high-voltage capacitors with a charge could be dangerous.
️Warning️ High Voltage DC Can Kill!Before I proceed, I must remind all readers that high-voltage DC can kill. It can be a danger in multiple ways – if one makes contact with high-voltage DC, their muscles are likely to clench. In the case of hands, this could mean that you end up “gripping” onto a cable, improving the connection and are unable to let go. In other cases, if the current travels through the heart, it could stop the heart and lead to death. Furthermore, high-voltage DC can also sustain arcs, which could lead to generation of ozone (a health hazard) or even to fire which could lead to loss of life.
It is important to make it clear – what I have attempted here should not be replicated by anyone else, especially if they are not confident with working with high voltages or do not take the necessary precautions. What I have done does not meet all safety standards simply out of practicality and cost considerations, and should not be taken as an example of best working practices. Anyone reading this who chooses to take any action based on this information is doing so at their own risk and no liability will be accepted.
For my own safety, I had a few rules with regards to testing to keep me safe:
In an ideal world, additional safety could include the use of a test chamber with interlock to cut power, but I didn’t have any to hand. Furthermore, one should also adhere to all equipment safety requirements (e.g. maximum voltage to Earth) and observe insulation voltage ratings. However, I did have to bend the rules slightly (and at my own risk). For example, most of the high-quality Hirschmann test leads I had were “rated” at 60V because they use unshrouded banana connectors. The cable had thick silicone insulation and showed no issues handling 1kV on an insulation resistance tester, so I decided to use my own judgement here simply so I didn’t bankrupt myself buying a forest of test leads.
Let’s start with the “high voltage DC hero” of the house, my Keithley 2450 Sourcemeter SMU which is capable of 200V at 0.1A for a 20W rating. This isn’t quite enough to run the board up past 200V on its own, but it should be fine for running the board between 30V and 200V, right?
Actually, no. This is because the Keithley 2450 has a two-range system – 20V up to 1A or 200V up to 100mA. Therefore, at 21V, I would only have 100mA available, for a measily 2.1W – nowhere near enough for the 9.75W rating of the converter and that’s before accounting for efficiency losses and peak-to-average current ratio (or the “spikiness” of the current drawn by switching circuits). By now, perhaps an Aim-TTi SMU4201 with its Powerflex ranges would be nice, but I didn’t win that RoadTest, so too bad.
The next issue lies in the fact that SMU outputs drive all quadrants, thus, they’re really not designed to operate in series with other instruments to boost voltage. In fact, as soon as the SMU turns its output off, it will see the rest of the voltage of the series-connected supplies across its output and that’s not going to make it happy especially if its output-off mode is not “high impedance”.
We also often hear the story that SMUs are like a “very high end” power supply with more quadrants. But perhaps we don’t stop to think about what issues that might entail. In early quiescent current testing of the board (as soon as I received it), I encountered lots of issues getting it to run on the SMU output even at high voltages because of problems of stability. The SMU regulation loop is “fast” but not fast enough and the spiky current characteristic of switching converters can interact with this to create persistent oscillations. Unfortunately, for me, this meant that getting a stable voltage/current reading was nearly impossible at a number of points because the SMU and the input capacitance of the board were playing a game of charge-and-discharge.
That’s when I decided to “convert” my SMU into more like a PSU. First was to get rid of the other quadrants – the supply can only source positive and not sink. That’s easy – whack a diode in the way, or in my case, make it two so that it’s not a single-point-of-failure. The next would be to make the output series-friendly – this means limiting the reverse polarity and essentially installing a bypass diode into the mix to allow current to flow by even if the output is turned off. Finally, we also need some bulk capacitance. My supply of high-voltage rating big capacitors was limited, but two 47uF capacitors is probably better than nothing.
The circuit was built on some Veroboard and pushed into service, being held inside a plastic yoghurt tub for safety. There is no discharge bleed resistor on the capacitor, so care must be taken to avoid a shock. This was a deliberate design choice to avoid skewing measurements overall – already the capacitors and diodes will add some leakage current. But because the SMU no longer has a “direct” connection to the circuit, I did not use any of its measurements going forward, instead relying on another high-precision power supply to provide the current (all-in-series so current should be identical through all branches) and a 5.5-digit multimeter to provide the voltage.
In order to reach 550V in a safe, reliable and controllable fashion, I had to call in practically all of my power supplies. As you will see, this was a bit of a co-ordination exercise, but also one of careful planning. It wasn’t just the output voltages I need to think about.
The stars in this setup are:
Ever wonder what it means when we say that a power supply has “floating” outputs? The most common answer is that the outputs are not referenced to ground so they can be combined in any way to get whatever voltage is needed. But this is only partially true. If you look closer, many power supplies have marked next to their terminals “max 250V to ground” or something similar. This is because, while the outputs are floating, there is a limit to the voltage rating of the insulation that separates the output from the chassis ground, and in many cases, there are also transient suppression ceramic capacitors that bridge between ground and the outputs that have a voltage limit as well. As a result, it’s not possible to get “any” voltage you would like. Some power supplies are less tolerant than others – the HMP4040 can only go 150V to ground, while the QPX750SP can go 300V to ground. The worst part is that some instruments don’t make this clear – the B&K Precision Model 8600 has no indications about what it can handle, so I’ve decided to take a gamble and say that it can handle 300V based on the fact that its slightly-unrelated Model 8500 predecessor is said to handle 300V.
This also brings up another point which some people might be curious about – why do some power supplies have an Earth terminal on the front? Well, as it turns out, it’s useful for enforcing the ground point in the system by connecting a banana jumper cable. This means that there is a known fixed earthing point. Without it, the output of the supplies being floating could “drift around” somewhat and this could end up exceeding the rating-to-Earth of some of the instruments and eventually cause damage. Earthing of all the instruments is otherwise handled by the power cables, living in a country with three-pronged outlets.
As a result, the setup I have in mind will give me +250V with respect to ground, and -300V with respect to ground, for a total of 550V between the two connections (which is what the board sees). As the board is not otherwise connected to ground, it won’t see any difference, but this also means the DC electronic load will see its negative terminal at -300V respect to ground and its positive terminal at -285V with respect to ground when operating. In hindsight, it would probably have been better to invert the stack, so as to have +300V to earth and -250V to earth to reduce the strain on the DC electronic load.
This is a somewhat scary moment for me – if I didn’t get everything right, I might just have the biggest test equipment accident ever and take all of my gear out of commission, leaving me powerless (pun intended).
It wasn’t easy to get set-up, as my ordinary test “rack” (shelf) setup had to be rearranged and recabled to make this happen. Furthermore, I had to write an automation program that would speak to all eight instruments and command their twelve total rails to meet my aims.
Of course, this is further complicated by the fact that series-connecting all these power supplies is likely to introduce poor voltage accuracy. The reason is simply because inactive rails (or even those set to 0V) don’t behave exactly like pieces of wire. Instead, when off, most power supplies have an anti-parallel diode (for series-ing) that prevents the input from going reverse polarity and hurting the circuitry or output electrolytic capacitors. As a result, when the rails are off, a diode-drop is going to be seen across the output. We all know diodes have an I-V curve, which means that the voltage lost will be dependent on current draw. Further to this, even when supply rails are turned on at 0V, not all supplies can bias the diode accordingly to compensate for its drop, and even if it attempts to, it could result in instability or oscillation.
As a result, my script has a “trimming” feature where it requests a particular voltage, checks the value on the multimeter, computes the difference and commands the SMU to boost its output by that amount to bring it into line. But since diodes are non-linear, this is done iteratively until the output is within 1mV of target, in which case the trim value is memorised, and experiments continue until a spot-check shows that the voltage has drifted by 25mV in which case a new trim is initiated. It’s not going to be as “accurate” or as fast as a single rail that can be set in one go, but it’s the best way to compensate for the situation I’m in.
Despite my best attempts, it seems that problems still remained. One thing I didn’t bank on was the fact that different power supplies have different transient response capabilities. As a result, during a certain part of testing, it seems the spiky load from the reference design interacted with the HMC8043 which couldn’t keep its output steady, wavering by up to 10V at an instant and throwing up a sporadic “sense” error. As a result, there were sections of the results which have been “censored” as the input conditions were not as requested by the script.
The script also couldn’t account for an issue with temperature – the shelf I’m using isn’t the best for heat dissipation and with the arrival of summer-like temperatures in Sydney, we suffered an occasional overheat spoiling the test. Relocating some of the equipment managed to fix this.
However, then it was the turn of another power supply (the QPX750SP) to occasionally complain of a Sense error too despite having its terminals disconnected, but rather than have an unstable output, it would turn off its supply entirely resulting in an up to 80V loss of voltage across the stack. The conditions under which this occurs is not entirely clear, but it seems rapid slewing of voltage can cause this to happen.
Then, came the biggest annoyance of all – occasionally one instrument (I don’t know which) would stop responding to commands entirely, causing a timeout. Ultimately, this made testing a little more difficult, resulting in me having to manually restart testing and key in the start conditions only to watch it fail randomly some time later on. Eventually, I made the script control my mains plug attached to the rack – whenever it lost communications to any instrument, it would perform an automatic power cycle and reconnection, then resume testing from where it left off.
Another thing to consider is the need to make spot measurements. For this, I used my Rohde & Schwarz MXO4 Oscilloscope with a pair of Micsig High-Voltage Differential Probes to ensure safety – remember, ground is not the same as the 0V output of the circuit! In order to do this safely, I had to modify the scripts to be able to drive the input voltage and load current to whatever values I wanted. Because of the issues noted before, this could not happen instantaneously – the script had to take slow steps towards the target to avoid upsetting a power supply and causing it to shut down.
It was a tough RoadTest filled with lots of fails and restarts, until eventually, I had a set of data to analyse.
While I will provide my test results here, it should be noted that it was only after the conclusion of all testing that I determined that there was an issue with my board, thus the results are not representative of the LinkSwitch-TN2Q’s capabilities. The failure was a subtle one which seems to have existed from the beginning, which unfortunately meant that it wasn’t picked-up until after testing had completed.
Efficiency results relied on the Keithley 2110 digital multimeter providing voltage readings and the Rohde & Schwarz NGM202 power supply providing current readings. An average of 16 readings were used to reduce the impact of noise. Censored data regions in white due to power supply oscillations causing the supplied voltage to the converter being >0.1V away from the intended programmed set voltage.
Efficiency appears to peak around 74% in my testing, which is mostly in the high-load regime around 0.5-0.65A between 190V to 450V. These results are quite a bit lower (~5%) than the claimed full-load efficiency of 78% at full load with 400V input.
An efficiency of 50% is reached at around 55mA to 65mA of load depending on voltage.
The output voltage seems to vary between 14.7V (-2%) to 15.75V (+5%) depending on the load, with lower voltages as the current increases. This is within the indicated 5% load regulation, but only just so.
At low currents, we can see higher voltages especially at low voltage input.
While the converter itself doesn’t indicate the availability of over-current protection, it does have cycle-by-cycle MOSFET current limiting which may do the job. The unit does advertise short circuit protection, so in order to understand if the circuit would come to harm, I decided to perform I-V sweeps until the output of the circuit shut down to determine the current at which the output ceased to be generated. The circuit was running in open air, ambient of about 25°C with no heatsink.
While the circuit is claimed to be rated 650mA and the IC claims that 850mA is available at higher input voltages, it seems that OCP really only kicked in around the 0.95A to 1.1A range depending on voltage. This should give some headroom in case the converter is powering some slightly “spiky” loads.
Thermal images were taken using a Topdon TS001 adjustable-focus USB-C thermal camera. Emissivity was defaulted at 0.95 while the unit was provided a full 550V and under maximum rated load of 650mA for two hours. Ambient temperature is about 24°C.
The hot-spot on this appears to be 88.5°C, at a point near the freewheeling diodes. There are also hot areas near the IC, the inductor and the output.
From the side, it seems the inductor is identified as the hot-spot at 86.3°C. These temperatures, while seemingly high, are within range and are similar to those detailed by Power Integrations in Figure 204 which measured 85.7°C at their inductor and 82°C at the freewheeling diodes on a 27.7°C ambient.
Two attempts were made at measuring quiescent current and power – the first was with the unit as-received, using the Keithley 2450 SourceMeter SMU only. Readings are thus limited to about 200V. The second is based off the whole 550V-providing setup. The input voltage was swept to the unit while recording current under zero load (i.e. output disconnected).
The results suggest a quiescent power ranging from about 660mW to 745mW which is a lot more than the claimed 50-60mW that we should expect. Given that the SMU and 550V measurements agree quite well (although small current measurements can translate into large power discrepancies at higher voltages), it seemed my measurement configuration was not to blame. Because of this, I did a more thorough investigation which uncovered some issues in the later section “What Went Wrong?”.
For this section, I’ll be making some spot measurements to check the voltage ripple, infer the switching frequency and observe transient response behaviour. Unfortunately due to the high impedance of the high-voltage differential probes, the Imonitor output from the B&K Precision Model 8600 was quite noisy, and at the voltage extreme, it even drifted suggesting perhaps that its inputs do not like being “floated” by 300V referenced to earth. Nevertheless, I present a set of oscilloscope screenshots to indicate what I saw under zero-load and full-load conditions. It should be noted, no additional capacitance was added to the probe tip unlike that used by Power Integrations, thus higher levels of transient may appear. A 20MHz bandwidth filter was engaged to reduce noise, with all measurements made DC-coupled with sufficient offset to bring the signal into view.
At zero-load, ripple voltage averaged 95mV peak-to-peak and the output resembles a sawtooth suggesting a low-power mode. At full-load, the ripple was more sinusoidal in nature, averaging 138mV peak-to-peak with a frequency of about 23kHz.
At zero-load, the ripple voltage measured 104mV peak-to-peak; at full-load, this increased slightly to 111mV peak-to-peak. The frequency appears to have increased to 32kHz.
At zero-load, the ripple voltage measured 148mV peak-to-peak; at full-load, this increased noticeably to 208mV peak-to-peak. The ripple itself had a sawtooth pattern to it. Frequency remained about 32kHz, again, not quite the expected frequency suggesting something could be wrong with the circuit.
Increasing to 380V, the no-load ripple averaged 200mV peak-to-peak; at full-load this was 267mV peak-to-peak.
At 450V, the no-load ripple averaged 222mV peak-to-peak; at full-load this was 282mV peak-to-peak.
Finally, at 550V, the no-load ripple averaged 257mV peak-to-peak; reaching 283mV peak-to-peak at full-load. Of note in all the latter cases is that the ripple value is dominated by a short switching spike, thus lower values may be attained with some filtering. But when considering the design brief stated 300mV peak-to-peak ripple as the target, this seems reasonable at all voltages. The frequency measurement being about 32kHz, however, suggests the chip may not be behaving correctly.
A look at the FFT showing harmonic spikes with some breath indicates the spread-spectrum (or frequency jittering as they call it) is working to reduce EMI emissions at single-frequencies.
For testing transient response, I set the B&K Precision Model 8600 up for the maximum 5A/us slew rate (which it probably did not achieve) at 1000Hz repetition between 10% load and 90% load.
Running at 550V, it’s clear the Imonitor output line isn’t quite behaving correctly, “drifting upward” likely indicative of some leakage current that’s upsetting the Model 8600. Perhaps it can’t handle 300V-from-ground gracefully as I had initially assumed. Nevertheless, the modulation of the voltage across the set-point is quite visible, resulting in 788mV peak-to-peak deviation. Just by eyeballing the graph, it seems that it takes the circuit around 200-300us to react to return to regulation after the transient.
With a lower input of 48V, the deviation is reduced to 472mV peak-to-peak but the transient response time seems to have lengthened on the “unload”. The response to loading is faster, around 200us. In neither case were there any instances of severe under or overshoot (e.g. dropping to zero) which is good and it is expected with a slower operating frequency that regulation loop speed would necessarily be more limited.
It’s clear from the results that a number of things don’t match up to expectations. Quiescent current was an order of magnitude off of the claimed values. What was going on? Is it my multi-power-supply-stack?
Testing with SMU alone shows slightly lower values than before but still not anywhere near the manufacturer’s claims. Compared to the SMU trace in the earlier graph which was taken when the board was received, things have not changed much, suggesting whatever issue I am facing was already present at the time the board was received. I then had a brainwave - this level of quiescent current should throw up something thermally.
As it turns out, the thermal camera suggests that VR1 and R6 are hot. This appears to be a TVSS diode and a limiting resistor of 100Ω.
A visual inspection didn’t show anything wrong with it at all. Perhaps the TVSS was ESD damaged and became leaky?
I decided to simply cut the trace between the two, opening the circuit and removing both TVSS and limiting resistor from the equation. This removes one protection for the output but shouldn’t stop the circuit from working.
It seems that with this modification, quiescent current has now fallen to within the expected range, at 38.85mW. Quite impressive. So I can now re-run my tests sans protection, right?
Turns out this was only the symptom of an even bigger issue. Now with this modification, the output was measuring 27V+ with a 30V input (with 25V-rated capacitors on the output). The circuit was not regulating its output nor bucking correctly. It seems the TVSS was simply “doing its job” and somehow it was shunting just enough current to make everyone happy and keep the voltage in-check. Or maybe the circuit simply isn’t happy with a DC electronic load’s ~200kΩ “off” state being the only load on its output. Regardless, this threw up the understanding that if the buck wasn’t bucking, then my electronic load would probably get fried as the input voltage increased, so I decided to end my experimentation here before anything else got damaged.
The Power Integrations LinkSwitch-TN2Q family is a highly-integrated solution targeting EV automotive applications, allowing for a way to transition loads that may have originally been designed to operate from the auxiliary battery to operating from the traction battery directly. The reference design has a 30V to 550V DC input range for a 15V 650mA output, non-isolated, boasting >78% full-load conversion efficiency at 400V, <60mW of quiescent power consumption, no optocoupler required, protections including OVP, OTP, SCP, 85 degrees Celsius ambient operation or up to 105 degrees Celsius with thermal pad and an AEC-Q100 qualified minimal BOM implemented in a compact footprint on a four-layer PCB.
Reviewing this product proved particularly challenging, with a lot of time spent on preparations and refinements to the set-up. Unfortunately, while I did complete a set of tests against my reference design board, by the time I came to analyse the recorded data, it became apparent something was wrong as the board was not meeting the expected performance metrics with regards to efficiency and quiescent current/power. The failure was rather insidious, as the unit was still generating a reasonable voltage at all times (i.e. within 0.75V of 15V) and could still deliver the rated current (if not more). Thermal imaging of the quiescent situation showed that the output protection TVSS was glowing hot. While my initial thought was TVSS damage through overstress (e.g. ESD) causing it to consume current even at quiescent, removal of the TVSS from the circuit resulted in the output flying up to 27V with an input of 30V, suggesting the circuit failing to regulate (perhaps not meeting minimum load) or an upstream issue (e.g. with the IC or regulation feedback).
As a result, despite months of hard work, the result of this RoadTest is inconclusive, as this board wasn’t operating correctly as received. This is perhaps not the result I would have wanted to present, but as the timeframe for the review has elapsed, I felt it was important to deliver my report nonetheless. I did learn a lot throughout the process regardless, however, and would still recommend interested parties to consult the comprehensive characterisation available in their official reference design report.
Thanks to element14 and Power Integrations for this opportunity. Feel free to leave a like or a comment in case you have any questions or thoughts.
Top Comments
Thanks for the reminder. I did suspect this, based on seeing another schematic in the datasheet with a 15kOhm resistor over the output instead of the TVSS + resistor combination. That is interesting to…
Thanks to your comment, I went and soldered a 4k7 resistor over the output from the underside to ensure the pre-load requirement is unambiguously met. The area at the top is just too crowded and the electrolytic…
Hello! second roadtester of this board here. From the datasheet it seems that it is not regulating at zero load: usually it is solved using preload resistor
(R4 from the schematics below: "Due to tracking…