It was over a month ago when I decided to embark on a project testing MOSFET Rds using the Rohde & Schwarz NGM202 power supply when I got a comment from a reader about the possibility of testing some “cheap” PWM solar regulators as they suspected that they may have some “fake” MOSFETs with high Rds values.
Needless to say, I decided to take on this challenge as I was interested in it as well. After all, my undergraduate degree is in photovoltaics and it’s a rather big surprise to me to find such controllers now hitting the AU$10-20 mark with quite generous current ratings up to 30A. Would such units be any good?
For testing, I decided to order two items – a 10A version and a 30A version to see what the differences between the two are.
The 10A version cost AU$12.86 including postage and GST. As it was shipped from China, it took around 34 days to arrive. The 10A version arrived in a cellophane bag. It contained the unit and the user manual (not seen in the picture above).
The unit feels quite flimsy and is mostly made of plastic. The front panel label is poorly fitted and the internal LCD is slightly angled. The unit claims “MCU Control, Build-in timer, SET voltage and FULL protect”. Three buttons adorn the front for user input. A row of screw terminal block inputs are provided for solar, battery and load. There are also two USB ports on the front for charging or powering USB devices.
The ratings are on a label on the top of the unit – it claims to be compatible with 12/24V with a 10A rated current, maximum 50V PV voltage and 130/260W power capability. The unit itself feels rather flimsy and is slightly curved on the rear.
The 10A version weighs 127 grams, which is exceptionally light for this sort of product.
The 30A version cost a little more, being AU$16.20 including GST and taking an identical amount of time to arrive from China. But instead of being just bundled in a cellophane bag, this one comes in a colour print cardboard box. It claims to carry ISO accreditation, but no mention of manufacturer. It also carries a CE logo, but that’s very dubious.
The box features a blue, white and black printed design. They don’t seem to have mastered the use of spaces in the title, but also seem to not understand the difference between “C” and “G”, advertising features such as LGD (for LCD) and SOG (for SoC). I do like the rear that says “all necessary protections equipped” – I guess this depends on what you define as necessary, but it’s very non-specific. There’s also a surplus space in the word “parameter” in the next line.
The side of the box very clearly state that it is Made in China. On one side is an area for a specifications label which is not attached, instead checkmarks are placed in the boxes on the opposite side.
The same style of rating label adorns the top of the controller. This one has a rated current of 40A with a maximum power of 390/780W.
The controller visually looks the same as the 10A version. Even the front label has a slight angle to it, although it seems to be better adhered. The user manual is a single page colour fold-out leaflet with Chinese on one side and English on the other.
It’s not the easiest to understand with a number of typos and poor English expression, but it’s sufficient to understand how to set up the controller. One hint I have is that you need to press and hold the first button to “unlock” the value to be adjusted with the up and down buttons and then press and hold the first button to set. Aside from that, everything is pretty much self-explanatory.
The terminals themselves are rising clamp type, accessible from the bottom. They are recessed somewhat into the unit, but because of the flimsy construction, tend to be difficult to properly secure as it moves at the slightest pressure applied to the screws. The row of terminals is also mounted at a slight angle in this example.
The body is relatively flimsy plastic – the design has a “raised” section for the terminal block. There are four mounting holes provided.
The rear of the unit is made of a black metal plate, secured by four screws.
The 30A unit weighs 132 grams, or 5 grams more than the 10A version. This suggests to me that there is a difference between the two units.
Taking apart the two units side by side, it’s clear that the 10A version has just three MOSFETs while the 30A version has five MOSFETs. The internals are, otherwise, very much identical. A look at the oily “splodge” left on the rear plate shows that the thermal pad is only making contact at the edges due to the non-flat angle of the MOSFETs on the PCB. This arrangement doesn’t ensure consistent clamping force and does not ensure good heat dissipation – the rear plate appears to be steel which is not anywhere near as good as aluminium.
The thermal pad itself is very “thin”, with adhesive on the side facing the MOSFETs and a “greasy” side facing the rear panel.
The two PCBs are basically identical, although the silkscreen suggests the PCB was manufactured by different plants. The design is known as W88-V3.0.
Internally, the capacitors used are not of reputable make – Chongx, JEC, Jwco and MT. The main controller resides underneath the LCD and is an unmarked 20-pin chip. Aside from that, there are a pair of LM258 Op-Amp ICs and an ON semiconductor MC34063A switching regulator controller. The input seems to be protected by a 47V MOV, so it’s strange that the maximum input voltage is listed as 50V as that might cause some degradation of the MOV over time. The MOV does not appear to be fused in any way, so if it does “let go”, a fire or shattered hot bits of MOV could be the result.
The inside of the front casing shows the mold used seems to be quite messy. Nothing unexpected from cheap Chinese electronics, I suppose. No clue as to who actually makes these units though …
To fairly test the unit, I decided to create a basic standalone photovoltaic system using some spare parts left over from my PhD and others which could be adapted for use. In some ways, it’s a demonstration of what not to do if you want to have an efficient setup for long term use.
Solar input is from two unbranded Chinese 50W PV panels. These were connected in parallel using bullet crimps to 0.75mm^2 mains flex leads with a mains plug (since I have no MC4 connectors or solar cable lying around). The parallel arrangement was had by using some cheap double-adapters to hook the two together into a 10m extension lead, also 0.75mm^2. My calculations suggest that the voltage drop would have only very slightly limited the maximum current when the battery is approaching full, but for demonstration this would be sufficient.
Snaking the cable through the window, it was hooked up to the 30A version of the PWM solar regulator via another double-adapter (due to gender differences) while the exposed pins were used as a way to measure PV voltage. A Century PS12180 12V 18Ah sealed lead-acid battery with about 10Ah of effective capacity (at five years of age) was used as the storage. Various MR16 downlights were used as a load – initially a 50W incandescent, then a 35W incandescent followed by a 10W LED unit. I added a few small switches to isolate the panel, battery and loads from the regulator to ensure proper sequencing and in case of operational emergencies. The voltage on the PV panels, battery and load were monitored using a B&K Precision DAS240-BAT Multi-Channel Recorder.
During the first day, the battery state was probably only around 30-40% charge to begin with. For the most part, the regulator was just connecting the panels to the battery – the voltage difference is in the voltage drop in the wiring. The PWM effect is not much seen on the first day which was partly cloudy, but the over discharge protection cut off the load at about 10.8V as expected (midpoint between battery and load voltage is the controller’s view of the voltage due to wiring losses). The second day was full sun, where the PWM was much more active in protecting the battery from excessive voltage, maintaining 14.5V (configurable). Part of the reason is that the 18Ah battery can only accept charge current up to about 5.4A according to the datasheet – its internal resistance was pushing the terminal voltage up. I was not able to determine the timing at which it switches from cycle equalisation charge to float charge – in winter, we might not get long enough periods of sun for this to happen.
With a smaller 10W LED load, the unit allows the light to run for a lot longer before it cut off. The load can be seen to activate at a PV voltage of about 8V, as stated in the datasheet. On the whole, this seems to be exactly as claimed on the datasheet.
One thing I learned the hard way was not to reconfigure anything on the DAS240-BAT while recording. In this case, after three days of testing, towards the end I tried to turn on the screen sleep feature. After that, when I went to stop the recording, I was greeted by an error and the recording lost all the data after changing the setting. Just another unfortunate quirk in the firmware on the DAS240-BAT.
The regulator never really warmed up under such a low current load (estimated at ~8A peak), which is a good sign. Unfortunately, the design of the regulator is a bit strange – the load is automatically switched on as soon as the battery is connected and must be manually turned off before connecting the load. The timer feature is not a clock-type timer, but instead is a “x hours after dusk” timer, detected by the PV voltage falling below 8V. The first time this is configured for 0H for dusk-to-dawn operation, it seems the load may stay on for some reason. I had to configure a certain number of hours and then back to 0H for it to operate correctly. But given the price, it seems to do a good job of preventing the battery from being “cooked” or over-discharged to death even if the accuracy is to about 0.1V.
Part of the reason I decided to test the regulators in an actual system was because of my testing of the regulator in bench-test situations.
I decided at first to try and simulate a PV input with the internal resistance feature of the NGM202 on one channel, while simulating a battery on the second channel but also attaching a bank of resistors to cause the virtual battery to “slowly” discharge naturally. Unfortunately, the PWM regulation on the charger is so fast, the NGM202 can’t keep up between charging/discharging of the virtual battery. Adding a capacitor to the load to try and slow things down helped slightly, but not enough as the NGM202’s simulated battery was not properly “accepting” charge. Other results, however, can be obtained from bench testing.
Low Voltage Disconnect
Testing the low voltage disconnect on the 10A version, the output load can be seen to be disconnected at almost exactly 10.8V. However if the battery voltage rises, even above the claimed 12.8V reconnect point, the load is not re-energised. This suggests that this regulator is not compatible with external charging that does not come via the PV input as it doesn’t reset the over-discharge protection. Definitely good to know but can cause issues in some special applications.
USB Port Output
The unit has two USB ports which seems to be a nice convenience. The datasheet says that they can deliver 2.5A maximum, but they are run from an MC34063A which has a datasheet rating of 1.5A. Somehow, it still didn’t look like it was sized enough to deliver that.
As the ports are paralleled, I decided to modify the 10A board by soldering wires directly to the battery input and to the pins behind the USB connector. Pairs of wires are soldered to produce a four-wire Kelvin connection so that voltage drop in the wire is cancelled out.
The port can really only deliver about 580mA before the voltage falls too far. At short circuit, it delivers a hair above 1.1A. Nowhere near the 2.5A claim.
Efficiency was determined by taking the output divided by the input. To determine the input, the quiescent current without the USB switching converter is taken away from the total input current to accurately determine input power (see next section). The results suggests that the converter has about 75% efficiency, making it very average at best. It is really not good enough to be truly useful.
As supplied, the datasheet claims the quiescent current to be <10mA. Using the NGM202, I determined that the average is about 12.91mA at a voltage of 12.8V (near a full-charge battery) which is more than what is claimed. This will reduce slightly as the voltage reduces. It means the regulator would cost about 310mAh a day.
But seeing as the USB output is absolutely rubbish, I decided to go ahead and desolder the MC34063A regulator IC entirely. This managed to drop the quiescent current to 9.63mA, which means that the switching regulator costs about 3.28mA when idle. It’s also good to know that nothing inside the regulator seems to be reliant on the switching converter 5V output, so it can be removed without affecting the rest of the operation.
Thus we reach the crux of why I was invited to test these regulators in the first place – the MOSFETs used within. Are they real, or are they fake? To try and answer this question, I first extracted all the MOSFETs using the “solder blob” and pull method.
For those who paid attention during the teardown, it is clear that the MOSFET packages were not all the same. From the rear, differences in the tabbing (notched, unnotched, notch size) are apparent. From the front, things are absolutely bizzare.
The 30A unit has four MOSFETs claiming to be SM7501N which is a Sinopower part number but it’s clear that there are at least three different types based on the plastic package “dimple” marks alone – the leftmost is operating the load control and is one type, the next two in control of battery reverse flow protection are a second type, while the right-two are a third type and in control of the solar PWM. This is corroborated by looking at the height of the plastic part of the package as well. It seems that the construction of the unit takes this into account – each “parallel” pair of MOSFETs are of the same “visual” type. Even the fonts vary on the markings in a very noticeable fashion. This could happen in some legitimate cases in the case there are multiple plants making the same MOSFETs but to have one product use parts from multiple plants is highly unusual.
All MOSFETs are surprisingly scratched-looking from the front. This suggests to me that these parts may have been recycled – likely to have been re-marked during this process and are perhaps not well treated. I’ve never heard of Sinopower branded transistors, so perhaps that brand and part number is a victim of a remarking operation, or they could be themselves part of one. I have no idea.
The MOSFETs on the 10A controller are even more bizzare. There seems to be an attempt to cover up the part numbers by grinding them away, but they did a bad job on the third MOSFET and I can see it’s an FBM80N70P. This MOSFET is actually quite similarly rated to the above, but with a slightly lower Rds. Looking at the dimples, it seems all have central dimples, but the rightmost one seems to have the larger dimple. The texture of the metal surface on the middle one is different, and the hole seems marginally larger. I’ve assumed the three are all the same MOSFET type for test purposes.
Because these MOSFETs don’t have nice pins I can wedge into a breadboard and because it is expected they will exhibit very low resistance of 6-11mΩ, I decided to build a test board using a bit of strip board and soldering each tested MOSFET for testing. Kelvin four-wire connections were used to the NGM202 and the test current was cranked to the full 3A capability of the unit to try and produce the cleanest readings of Rds vs. Vgs.
Of note is that the MOSFETs are rated at Id = 40A which is well above what I can achieve with ease. As a result, I can’t actually determine if the MOSFETs are capable of passing the 80A they claim or verify the Rds at 40A, but on the whole if the Rds is high then that should show at lower currents as well.
Because of the low Rds, another “issue” arises – the resistance of the test board itself may affect the results. It’s not entirely possible to get a “true” zero ohm correction, but what I did was to measure the test board fixture with the drain and source pads solder bridged together to form as close to zero ohms for the test fixture. This is reported in the below graph as “Null”.
Interestingly, the overall readings show that the MOSFETs are broadly similar, at least at the Rds rating condition of Vgs = 10V. In the case of the 30A controller’s MOSFETs, there were a few that were similar and a pair which were somewhat different, but that is well within expectations for manufacturing variation. The 10A controller’s MOSFETs were a bit different from each other – perhaps there are actually three types in there, but the Rds was actually lower than the MOSFETs used in the 30A controller.
Zooming in and subtracting the null line from the readings, it seems that all the MOSFETs measured within the expected range. The SM7501 is rated at 9 to 11mΩ – the tested results were about from 8.8 to 10.8mΩ. The FBM80N70P is rated for 6 to 7mΩ – the tested results ranged from 5.4 to 6.5mΩ. These results are probably slightly optimistic, as the null reading that is subtracted is not a true zero resistance – however, even without the null line subtraction, the MOSFETs actually test quite well and survived the torturous desoldering process as well.
After testing the MOSFETs, it was a bit of a game trying to suck out the very crusty solder from the PCBs. This required a co-ordinated effort with an iron on top and the sucker below … but I managed to re-mount the MOSFETs on both units. All of that soldering and desoldering and it seems that the MOSFETs survived!
Unfortunately, in the case of the 10A unit, I discovered during testing that I had a tiny scrap of desoldering braid on the load-control MOSFET. It shorted out the gate with the drain – this immediately destroyed the drive transistor and thus the 10A unit has lost its ability to run loads and perform overdischarge protection duties.
The generic Kw12x0 PWM Solar Charge Controller is a rather inexpensive piece of equipment but it doesn’t do a terrible job of being a basic solar charge controller. On the whole, it behaves as one may expect – protecting the battery from excessive voltage and overdischarge, with an integrated dusk timer function and USB outputs.
Testing revealed the controller did have some quirks – the load is energised as soon as the regulator starts up, the output is also turned on after configuring the regulator into dawn-to-dusk mode, the output does not reset its overdischarge protection if the battery voltage rises above the reconnection voltage, a quiescent draw a little above the claimed value and USB outputs that are severely anemic with regards to current.
However, despite all of this, the MOSFETs inside were a surprise. They appeared mismatched, scratched and likely to be recycled in some way. But testing revealed all MOSFETs were able to achieve the datasheet rated Rds at the test current of 3A. While the datasheet rating was made at 40A (and I have no easy, accurate capability to test up to that level), the MOSFETs were definitely well oversized for the application (e.g. parallel pairs of 80A rated MOSFETs in the 30A rated regulator) that it seems likely the regulator would run cool. This matches the relatively “ad-hoc” thermal interface between the MOSFETs and the rear panel which can have a very limited contact area (and hence effectiveness) due to the difference in angle between the package and the metal plate backing.
The build quality of the unit feels relatively cheap and flimsy, with terminal blocks sometimes mounted at a jaunty angle with screw heads that are easily chewed up, but what do you expect for $10-20? Something that works is already a big surprise to me.
The testing was not exactly casualty free, but I was still pleased to have managed to test all of the MOSFETs and reassemble both units, only losing the load-control MOSFET’s driving transistor due to a scrap of desoldering braid shorting out the gate and drain.