And we’re back!
After passing the mother of all blogstones, I hope it’s had time to sink in with everyone, because it’s time to replace it on the front page. Just because I wander away from this site for a week doesn’t mean I haven’t been doing anything siteworthy. In fact now I’m once again in a situation where I need to backpost like 3 weeks after the fact. Luckily, this time, it’s all loopy engineering content! That doesn’t mean I’m done with the 2.00gokart coverage – it will hopefully appear on Make soon, and I might actually be wrapping it up in a more presentable style for a conference next year. We shall see.
Beyond Unboxing began with me taking apart a derpy Hobbyking controller. While it will not end with me taking apart a different Hobbyking controller (at least, I hope not…), if you haven’t gleaned from the title yet, this is once again about them!
No, Hobbyking hasn’t come out with a 1000 amp controller yet. That will be the day. What I got instead is five different “200 amp” controllers!
Here’s how it happened. A long time ago, I had one of their “190/200A” controllers (which I blurbed about very briefly in that first ever Beyond Unboxing post). I wanted to do something with it recently, but it turns out I must have given it away to some eager frosh or senior design project a while back. So in a fit of rage, I ordered 1 of every “190/200A” controller type on the website. Seems logical, right? EIGHT! HUNDRED! AMPS! to make up for my lost two hundred amps, which was probably more like 15 or 20 to begin with.
Why did I actually get all of these? I can’t really say. You’re not going to ever pressure me to actually use these. I suppose it is curiosity driven – how far does the Law of Chinese Packaging Inertia (“If the Chinese product looks the same, it probably is the same”) go, and what minor differences (if any) exist between these four models? Just how terrible is that board sandwich arrangement anyway? And which one of these is actually worthwhile – if, hypothetically, for some reason I became extremely disoriented and tried to run one in a real application?
Here are the contenders:
“Birdie” 180A, no BEC (for some reason, the BEC enabled version is advertised as 190A). Codenamed Birdie hereafter.
“SS Boat Series” 200A, no BEC, or Boat
“Red Brick” 200A, no BEC, or Brick.
“SS Series” 200A, no BEC, or Simple. Couldn’t think of a good B name for this one… besides maybe Bad. But, I haven’t objectively proven badness yet, so I shouldn’t stain my own results!
I bought them all without BECs (5V supply on the center servo cable pin for external devices like receivers) since the first one I got was the non-BEC Birdie and then the rest had to follow suit. Seriously, that’s the level of science we’re dealing with here.
Furthermore, it’s important to note that all of these cost something around the $30 to $40 range. I know that more legitimate “200A” controller options exist, but they were more money than I cared to spend on something just to crack apart, the designs are all different (knocked off or not), and I’m sure they, like, work. This rowdy bunch of value-engineered misfits, however, might be a different story. To summarize, I am not saying anything at all objective about how well they will work for any particular application; with this kind of stuff, it either works and keeps working or destroys itself right away in a very noticeable and possible deflagrative fashion. I’m literally just hacking them apart to see how different they are. In fact, I gutted them before any one ever ran a motor.
My test will occur in three parts, as per usual for motor controllers. First, a visual inspection and snide commentary on the construction and materials choice; second, I probe the gate drive circuitry, one of the most critical subsystems on a motor controller, to see which one might have the best switching characteristics; and finally, I crank up the DC input voltage until the FETs spill over their brims – what is the true absolutely maximum voltage these can handle? I was specifically including this last one because some of them are advertised for up to 6S lithium operation (25v peak), some to 7S operation (29.4v), and some of them have anecdotes on their Hobbyking review threads of operating up to 8S (or 33.6v)!
Anecdotes on a Hobbyking reviews thread might in fact be the most tenuous possible factual claim that exists on this planet.
I lined these up in a row and labeled the servo cable with their designations before tearing into them. It’s like a corpse tag for things which never really lived. From now on, unless otherwise indicated, the order of tests and the lineup in pictures from top to bottom (or left-to-right) is always Birdie, Boat, Brick, and
Bitchtits
Simple.
A little more discussion about these lovely pieces of kit: They’re all derivatives of the Suppo 200A-LV design and are all the same basic phenotype. The FETs are stacked in three layers, an extremely poor thermal design. The board traces are also long and skinny and the copper thickness does not look like a heavy 4 or 6oz type, so let’s be honest and make it clear we’re not ever getting 200 amps of anything from them. It doesn’t even make mathematical sense, much less physical; usually you can get away with something working physically that seems like it shouldn’t theoretically, which is in fact the entire foundation of 2.00gokart.
One incision on each was all that was required. The heat shrink tubing on these is the exoskeleton – without it, the whole thing falls apart. It holds the heatsink in contact with the (upper row of) FETs, so just hope you never run them hot enough to cause the heat shrink to fully recrystallize and lose stretch.
You can kind of tell already that the bottom three are the exact same thing. The Birdie seems to have a different signal board; as it was the most recently introduced line, I might say it’s a newer version of the design.
The biggest repeated complaint I see about this style of controller stems from the fact that they put the heat sink too far forwards. It rests on the output lead solder joints, pierces the silicone insulation layer eventually (or even right away), and the resulting short blows everything up. I’m honestly not sure why nobody has thought of maybe moving the heat sink 4mm back towards the capacitors; there’s plenty of space to do so.
Above, the silpad layer is already eroded through in one spot, and working on another. Vibrations or mounting pressure will probably blow this through in no time.
If you actually wanted to use these, it might be best to cut the stock heat shrink, move the heat sink back a few millimeters, then reshrink it. This sounds like just enough effort to make it worthwhile, and you’d only have re-engineered 57% of it!
The boat ESC has a goofy little aluminum waterblock that is made of one U-shaped flow channel – two holes drilled in one side, and then a cross hole connecting them, with the surface sealed with some kind of press-fit slug. This isn’t an unreasonable surface area to volume ratio for its size, however, and maybe if they stuck with one layer of FETs it would actually be effective.
The controller itself is also entirely dipped or brushed in what must be polyurethane floor varnish. It smells like “new hardwood floor”, there is no way to describe it otherwise. I assume this was a grasp at making it at least splashproof.
The Brick offered something extra – besides the dinky little heat sink on the FET side, it had a bit aluminum heat spreader plate on the signal board…
…that doesn’t seem to touch anything important in particular. Maybe it’s for the regulator (left component on the top board), but the aluminum touches it by a mere sliver! Maybe the gate drivers just get that hot? On this model only! Or maybe it’s to bulk up the thing a little to make it look more hardcore, like injecting your chickens with salt water, only less conductive.
I didn’t take any pictures of the SS non-boat model because it was… well, Super Simple. Exactly like the rest, but not even a weird waterblock or vestigial heat sink on the back.
One thing I noticed right away about Birdie is that it had a Fairchild FET part number. Specifically, the FD8896. However, a closer look at the devices shows that some are slightly differently shaped – square tabs versus round, and the markings and case molding marks are different. So these could be counterfeit devices made by a handful of nameless fabs, or they could be the infamous “pulled parts” that Chinese electronics recycling outfits are known for providing.
The assembly job on these controller power boards is clearly all manual. Some of the FETs aren’t sitting flat, and others are… are they even soldered in?
Below are high-resolution photos of the observe and reverse of all 4 ESCs:
Besides Birdie, all of the rest use FETs made by Nikos Semiconductor, specifically the P0603BD and P0603BDL (I could only locate the datasheets through another company, which is named just similarly enough for me to think they’re fronts for each other or a third operation). Brick and
Billybob
Simple use the BDL variant, but Boat used the plain BD. What’s the difference?
30 volts maximum Vds for the BD, versus 25 for the BDL. So at least one of these right away could satisfy the “runs on 8S” myth. The “Failchild” parts in Birdie are also 30 volt rated.
It’s often the case that FET maximum Vds ratings have a margin built into them of 15-20%. You would still be advised against using FETs anywhere near their Vds maximum, however, since any cresting of this limit will result in the FET entering avalanche breakdown and destroying itself very quickly. R/C ESCs are often run at the limits of their semiconductors because they do not perform regeneration (i.e. the voltage coming from the controller will never exceed battery input voltage), and you can get away with this plus the manufacturer’s overhead. For fast reversing and regeneration applications, my rule of thumb is still 100% overrating if reasonable – that’s why RageBridge uses 60V rated parts and I advise it to be run no more than 36v nominal.
However, we can already see that for Birdie and Brick, 7S lipos might be a bit bogus. The former says 2-6S right on the sticker and website, but the latter says 7S! It’s also important to note that these cheap controllers tend to use whatever FETs they can get. So it might be completely unsurprising if I buy another batch and find that more of them have the “Failchild” parts, or the next boat ESC has BDLs. This factor probably contributes to the legendary 8S lipo anecdotes.
The bottom line is, trust nobody.
Here’s a high res version of the reverse side showing the driver chips and signal side layout. Once again, Birdie is obviously different from the rest, which are clearly clones.
Pretty much everyone uses the IR2101 half-bridge driver chip. I’m stoked to see they even bother with driver chips – why can’t Jasontrollers move on in that direction too instead of a nest of discrete diodes and BJTs? The Brick, though, uses the IR2106, which is an identical chip, but looks to be a few milliamps weaker and probably a few millicents cheaper.
It’s interesting to see that all of these derive gate drive power from 12 volt Flash memory programming power supply chips. They’re charge pump based supplies, so the current is pretty minuscule compared to, say, the switching boost converter of RageBridge, but it’s enough to get the job done. Birdie uses the LTC1262 and the rest use MAX662A. Like the two gate drivers, these are pin-and-function-compatible chips. This is pretty crafty, I must say.
So this whole setup suddenly got more interesting, because the component variations come out to the following:
Birdie: FDD8896 on IR2101 drivers
Boat: P0603BD on IR2101 drivers
Brick: P0603BDL on IR2106 drivers
Bongwater
Simple: P0603BDL on IR2101 drivers
I could now get a reasonably good cross-correlation on which FETs plus which drivers might yield the best switching times.
Let the science begin.
I soldered little pigtails directly to the gate drive outputs to view them, and set a servo tester to run the motor at a constant speed. I just wanted to take a peek at the drive waveform. The voltage was set at 25 volts for all of these tests. Again, I have no agenda to prove that one is necessarily worse than the other, but the faster the switching time, the less switching dissipation for the same frequency. I figure all of these have sub-milliohm resistance when on since there are so many devices in parallel, and that Chinese ESCs are well known to have very weaksaucy gate drives, so the losses from turning on and off might actually amount to something significant.
If you’re wondering how all the alphabet soup in a FET datasheet affects its operation, Renesas / NEC has a good power MOSFET datasheet reading guide and design notes. Specifically pay attention to the righthand column in the first few pages.
Switching profile of Birdie showing about 1us switching time from Vth to end-of-Miller-region
The metric for “switching time” that I use is the time between points:
When the gate voltage Vgs hits the threshold voltage Vtg
When the “Miller plateau” ends and the Vgs begins to swing exponentially closer to the gate drive supply voltage.
MOSFET switching loss calculations are actually not simple things. There are many dependencies – current through the device, Vds (across the device), gate drive voltage, temperature… There is no “one equation” that can capture all the intrinsic behaviors. Vishay has a good read on how switching times are affected by all these parameters, if you care for the guts.
Extrinsically, though, I usually use a two-triangular approximation for power dissipation . In between the two time instants I mentioned, the Vds usually falls from your supply voltage to nearly-zero in the first half, and the Ids (current through the device) ramps up from zero to whatever your max current may be in the second half. This looks sort of the like the graph in Figure 4 in the Vishay appnote.
So this simplified switching transient model that only captures the resistive element (The FET not being a perfect pass-through for a split-split-split second), and it doesn’t account for the actual power dissipated by slurping all the capacitors in and out. But my contention is that in a relatively low frequency motor control scenario, those losses are nearly inconsequential – we’re not building megahertz-scale DC/DC converters, these ESCs switch between 8 to 32khz max – and the currents involved tend to be higher and of a more pulsed nature than a DC/DC which aims to supply a near constant current when possible.
It works out like this:
No, it’s not mathematically rigorous. But the equation’s there at the bottom. That’s the ‘average resistive dissipation in watts for a known V and I load’. For most little things, it works out to be single digit watts or less. For instance, using that approximation, RageBridge at 50% duty cycle (switching each fet equally) pushing 36v and 40 amps will contribute 4.5W of heating to the FETs, on top of raw conduction (I²R) loss which would be, and that’s with my near 100 nanosecond switching time. RB switches at 32,000 Hz, and this extra frequency really ups that number.
Again, this doesn’t count capacitive slurping losses (charging and discharging Ciss and Coss), nor diode conduction losses which are extremely important for brushless motors and bidirectional DC motors. This is like, clicking a switch and a light bulb on and off.
Switching profile of Brick, the worst one at nearly 2.5us
The tests seemed to confirm Birdie as the frontrunner for switching speed, at around 1us.
Boat and
Broslice
Simple both hovered around 2us, but Brick was the straggler here, with the weaker drivers feeding the BDL type FETs which have much higher gate capacitances according to the manufacturer.
I’m not sure why there’s such a discrepancy between Birdie and the rest – maybe those “Failchild” FETs are actually legitimate devices and their drain to gate capacitance is much tighter managed. In practice, you’re likely to not really notice the difference between such minutia, since the power dissipation by I²R during motor driving is likely to be far higher than clicking the things on and off. And with the FETs as stacked as they are, it will probably light them on fire anyway.
The next test was where I got brutal. I lined them up again, set the motor to a gentle low speed cruise at around 1 amp, and then slowly turned the bus voltage up 0.1 volts at a time until the current suddenly increased and the motor started sounding really bad. This was the sign that the FETs have entered avalanche breakdown mode.
At this point, I had a few seconds to click a picture and then rip the controller off the supply before stuff started vaporizing. The controllers would heat up very fast – too hot to touch in only a few seconds. I recorded the rough voltage that this happened at.
Above,
Buttpirate
Simple is getting the breakdown treatment (I started proceeding backwards through them once I was done poking drives) and lost it at around 33v.
Next up, Birdie, with its 30 volt fets, showing a higher breakdown voltage since it uses 30V rated parts instead of 25.
Brick has 25V rated parts and broke down similarly to Simple at 33v.
Boat with its 30v rated shady Chinese FETs got the highest breakdown voltage at 37v. I was legitimately afraid of grenading the input capacitors or the single 7805 type regulator feeding the logic at this point.
What if…. I paralled all of these boards together?
So there we have it. I, again, have no emotional attachment or agenda to enforce with these, and you are free to use or not use them. I just have shown the process of the sausage being made, or perhaps this is more like the examination of the sausage under a microscope. I, in fact, love Chinese sausages.
Let’s put all of this in perspective: The fact that these devices were manufactured, the PCB printed and plated and drilled, then the whole thing assembled most likely by hand, and you can buy them for $30-40 and they actually do work (I did verify that they did indeed turn motors), is a phenomenon in and of itself. They are intended to be super low cost, so the workmanship and materials choice will be lacking. I’m not pointing out that they’re actually shitty – we know this when we make a choice to buy them.
For amusement and science, I borrowed a FLIR camera and took thermal profiles of Birdie to illustrate the fallacy that is 17 layers of FETs:
Science just got serious. The right side bank of FETs is the low side of the bridges, and it’s the side which is being switched at PWM frequencies, so it will experience more heating regardless of load. The controller is running a left over giant EDF from Deathcopter at about 30 phase amps here, but well out of the airstream.
Here it is in the direct path of the airstream of the EDF at the same power level. You can see the top layer being stone cold (comparatively, anyway), but the center… Let’s not talk about that. Also interesting to see is the FETs closest to the big phase wire outputs are the coolest, since a big rod of copper is a very effective heat sink on its own. So big ESC wires are in fact beneficial.
I do legitimately think these things can flow 40-50 amps all day with good airflow hitting all the boards (not only impinging on the top layer heat sink) – then convective cooling against the exposed metal and long board traces should aid it. And the fact that at least some of them can run well above 30 volts means that they could be well suited for an 8S A123 system (25.6v to 28.8v). It also means they can seemingly all handle 7S lithium batteries, but I haven’t verified this in any extended, heavily loaded operation. And all of this could very well change with the next manufacturing run when I’m sure they’ll pick a handful of obsolete IR or Infineon parts out of the box. The firmware on all these is comparable – the startup tone is the same, the stick calibration is basically the same, and the starting routine for the motors behaves the same.
I’d wager that these would be a great choice for a super low budget 8S A23 or 6-7s lithium polymer powered scooter, good to maybe 1000-1500W (which is already way more power than you need in a scooter…), using a 50mm motor and with a fan appended onto the heatsink. It has the momentary amperage overhead for acceleration and making up for the lack of current control. The cruising power of said scooter will likely be low enough that the thing can stay at thermal equilibrium (even if the bottom fets are conducting through the middle or top ones to cool). This is a sub hundred dollar rig.
Anyways, if you *made* me use one, I’d probably pick Birdie because it seems to have the newest board layout, maybe the most legitimate devices, and the cleanest switching profile. However, Boat also has promise for the highest breakdown voltage – if the 30v parts are consistent, you could conceivably fan the mythical flames and run 8S lithium polymer… but why would you do that….
Instead of using these, I think I’ll spend an extra ~$60 and go for the big guns:
Yes! On the other end of the 200 amp legitimacy scale, this is the Trackstar 1/5 scale ESC. This thing is huge, in a huge fashion, and clearly designed for more rigorous ground applications where it’s going to get beat on. The cost? About $110. Can’t argue with that.
It advertises operation up to 8S, so I’m going to guess it has 35V caps and 40V fets (the next rung up).
Let’s start cracking it open. I’ll say pre-emptively that this is an entire different class of hardware than the cheapo 200A controllers, so you shouldn’t try to compare between them too hard.
First of all, this thing uses a mishmash of different hardware to keep it together. You have Phillips heads somewhere and hex head buttons of various sizes and lengths elsewhere. The hell?
And the button heads are no known common size. Seriously – 2mm strips, 5/64″ strips, but 2.5mm and 3/32″ are all too large to fit. 2.25mm??
I resolved this problem by slamming a T10 torx driver into the screws until I forged Torx corners into them – they’re rather soft. This worked extremely well.
The top and bottom caps are coming off, and the first set of bus capacitors are peeking out at me. 1000uF each at 35V.
This thing also has two “distal” buscaps, each of 560uF. This is great, because otherwise the very long layout of the board means the FETs at one end will not have any bulk capacitance to draw pulsed currents from.
The driver board is minuscule and fell out as soon as I started yanking on the heat sink (which is therm-o-taped on). On the reverse side, the gate drive chips, main processor, and support parts.
And on the obverse, the switching converter for the internal logic supply and a small boost converter! for the gate drive. Wow, this is getting serious. This is the exact same (not parts, but execution) rig I have on RageBridge.
High-rez photo of the meat and potatoes of this controller (the FETs would be the wine and cheese). The processor is a SiLabs chip, I think a C8051F367 (based on the fact that “Silabs F367″ returns that exact chip…). For a while now, most of the higher end Hobbyking equipment has been using SiLabs chips, because they have more powerful timer options and ADC options than the same class ATMega, it seems.
The driver chips are impressive for this class of hardware: the Intersil ISL6700, a 1.25A type, still pin-compatible with the other IR 210x chips. Whoa, it’s more than one amp! Classy.
A shot from the bottom for the semiconductors. These are Philips NXP PSMN-2R640 chips. Not bad. I see that actual design and part selection work has been done for these instead of sticking with good ol’ IR strategies.
How are they laid out? Ready for more FET spam?
That’s just half of them.
Here’s the other half. The total tally is 10 devices per leg, for a total of 60 devices.
That’s six more than the cheapo 200A controllers, which have nine per leg (across three layers).
Adding little cockroach antennae to the FETs directly was way too hard, since the small soldering iron tip I needed to get in there could not dump enough heat to overcome the proximity to the giant 8 gauge wires and giant brass conductor strips. When I pulled the iron away after trying for about 10 seconds, it was still stone cold.
So I had to move these to the other side of the divide, the gate driver chip itself. Not as optimal but it will still give me a readout of the voltage rise.
The first thing I managed to do with this controller once I hooked it up was regen into my power supply, scaring it to death. Oops.
Once again, if your controller can even think of performing regenerative braking, you must must have a battery buffer or run from a fused battery! Power supplies cannot generally sink regeneration current and the voltage will rocket up. This could destroy power supply and motor controller instantly.
Luckily, the BK 1902 designers probably accounted for boneheads like me, so it actually reverts to a shutdown mode. This poor supply has been back-fed at least 5 times!
Here’s the gate drive output. Unlike the cheapo ESCs which switch the PWM on the low side (ground referenced) FETs, this one does it on the high side. Because the power supplies for the high side gate drive on the cheapo ESCs is a wimpy chargepump, they have no choice but to do this – switching the FETs many times a second would flatten the supply in short order. But with this controller’s dedicated boost converter, it can use the more common “legit” topology. The reason I call it “legit” is because it’s the easier way to also include regeneration by switching the low-side FETs in sequence with the high sides, again a matter of robust power supplies and which FETs can be switched many times a second without sapping it.
The thing to notice is how underdamped it is. It might be in part due to measuring before the gate resistor, but a closer inspection reveals it’s also just very aggressive:
The Chinese have finally learned what an underdamped gate drive is. Help us all.
The rise time waveform on the gate driver side is very sharp, but again, this is before the gate resistance. I do want to try snatching the actual gate signal eventually, but even the fact that this waveform is underdamped (wavy) means the drive is no longer limited by the gate driver current – if the driver is giving it all, then even on the before-resistor side there will be a slow rise. So, nothing can quite be concluded yet from this, but the fact that it has a 1.25 amp driver is already a massive improvement.
I need to run this fucker in a real application, because I really like its potential. The firmware is much, much smarter than the cheapo ESCs or even most other hobby ESCs I’ve dealt with. Let me show you the “squeeze test”. No, I’m not kidding:
The starting routine on this begins to dead-reckon the motor’s drive frequency, but it can detect pole slip (the motor just cogging about) and immediately begins slowing down that frequency more and more until the rotor finally couples in. The final dropout speed is very low – like 5 or 6 eHz, and remains audible the whole time. As soon as it detects rotor coupling, it wangs (that’s a technical term) the rotor from my hand. This process takes maybe a second or so. Finally, starting and low speed running is also using a low frequency (2khz) PWM mode. I’m really not sure why this is done – once the speed reaches a certain amount, it audible cuts to 8kHz PWM. If I load the motor (like by squeezing it) to below that dropout speed, it reverts to the low frequency PWM.
There you have it – one thousand Chinese amps, of which maybe 250 are legitimate! Based on the parts and thermal design of this controller, I’m actually willing to give it the benefit of the doubt. The parts quality (if the devices are all real, that is) is fairly high end and the PCB, despite being also the primary heat sink, is heavy and has big brass bars all over it. This thing is pretty weird and worthy of more investigation, but I can’t really say much more about it unless I put it on something – like a silly go-kart.
That will come soon.