Tested: HW-432 XL6009-based 3-32Vin 5-35Vout “4A” Boost Converter Module

When supplying power to electronics projects, it’s usually the case that one voltage won’t satisfy everything. As a result, the use of power conversion is often mandatory – either you can step down the voltage (using a linear regulator or a buck converter) or you can step-up the voltage (using a boost converter).

The HW-432 is a boost converter module that can be found online at a wide variety of marketplaces. It is built around an XLSEMI XL6009 400kHz 60V/4A switching controller configured as a boost converter. Most marketplaces rate it as a 3-32V input for an adjustable 5-35V output at a 4A maximum current. This seems a bit ambitious, but let’s see how the module performs when it is put onto the test bench.

The Unit

It arrives sealed inside a static shielding bag.

The module is a familiar size as it seems to have a layout similar to a prior design utilising an LM2596. This one is based on the XL6009 and is often touted as the superior alternative. The circuit appears to be a textbook implementation, comprising 220uF 35V SMD electrolytic on the input, 33uH shielded inductor, a 10k multi-turn trimpot for voltage adjustment, SS34 Schottky diode and 100uF 50V electrolytic on the output, paralleled with a small ceramic capacitor of unknown value. The bottom resistance in the feedback divider is a 330 ohm resistance and is paralleled with a ceramic capacitor of unknown value – this is likely to stabilise the feedback signal against switching noise.

The high current paths can be seen to be via-stitched to meet current carrying capacity. Most of the top plane carries the positive connections, while the rear plane carries the ground on this two-layer board. Unfortunately, the XL6009’s heat is dissipated only to a small patch on the back-side through stitching vias – this isn’t likely enough to allow for the full current rating.

The tallest components on the board are the multi-turn trimpot and the SMD electrolytic capacitors.

Of course, at this price, you’re not going to get a Bourns multi-turn trimpot. I’ve seen a few different “copy brands” in my time – this one is a Baoter 3296 from XCHC Electron.

While a lot of the design seems straightforward, the 4A rating may be a bit misleading for the uninitiated and a bit generous. Can you get 4A continuously from the boosted output? Almost certainly not – the output goes through a 3A-rated Schottky diode that’s going to cook if you tried to push 4A through. Instead, the 4A rating most likely refers to the current flow through the inductor from the low-voltage side – but even this seems generous. With a typical Rds of 0.11ohm, the power dissipated by the XL6009 would be at least 1.76W and that’s going into a set of thermal vias to a plane on the rear that has an area not much bigger than the D2PAK package itself. To add further insult, if we assume that it’s doing a 2:1 step-up, 2A will need to go through the SS34 Schottky which is right next to the package, which will have a voltage drop of at least 0.3V leading to another 0.6W of heat right next to it.

As a result, the absolute maximum current you can get at the output depends on the worst-case (lowest) operating input voltage and target output voltage – after all, power out <= power in. For example, you have a 4-cell AA battery pack which might be 4V at the lowest and you’re expecting 12V output. 4V * 4A = 16W input, so therefore, at 12V the current you can maximally obtain assuming no losses = 16W / 12V = 1.333A. This is in the ideal-case but in practice, it’s likely going to be a bit less due to self-losses in the circuitry but also due to thermal constraints. At low step-up ratios, the output 3A Schottky will limit the output (as it will likely burn-up at higher currents), so if you get an answer greater than 3A – take it as 3A because of this design choice.

Efficiency and Voltage Stability

I decided to run the module through my scripted “efficiency surface” tests which probed the board at every 0.01A load step at every 0.5V input voltage step. The script has the above power limits programmed into it and was to respect the 3A Schottky limit too. Test equipment used are a B&K Precision Model 8600 DC Electronic Load and Rohde & Schwarz HMP4040 Programmable Power Supply.

The board was wired using four-wire connections at both input and output, to ensure voltage drop on connecting wires was negated. The converter was given the benefit of operating in still open air, so that natural convection can provide cooling, although no forced cooling was provided.

Unfortunately, after half-a-day of testing, my first module became toast despite the script being respectful of the limits.

While the XL6009 has internal thermal shutdown to save it from burning up, the Schottky diode did not. The thermal camera suggests that at the highest test loads, the diode outer surface was reaching almost 138 degrees Celsius. Of course, at such high temperatures, it’s not going to last all that long. The XL6009 was not much cooler either. As I surmised earlier, the board is a bit too ambitious on specifications and lacking in thermal dissipation.

Nevertheless, I revised the script to cap the test current at 2A for a second module and forced it to terminate the test early if the output voltage drifted too far (a sign of an over-stressed converter). At each Vout, only the Vin steps strictly less than the Vout were run. Three graphs are provided – an efficiency graph that covers all steps, an efficiency graph that looks at the low current steps and a graph of output voltage under load. Due to interpolation by the plotting process, there are some strange fringes on the efficiency graph – if you don’t see a result on the voltage plot, then the converter does not operate satisfactorily at that condition.

The results are as follows (note that I will only provide commentary on some of the graphs where new phenomena appear):

Vout = 5V

Not much boosting to be done when your output is so close to the minimum input voltage. The converter struggles at 3V – the output collapses at just over 200mA of load. It improves at 3.5V to about 900mA. But at 4V, while we have efficiency measures, the converter is unstable above 1.4A as if thermally stressed. Conversion efficiency is not bad – around 82-83% peak at 3V input.

There’s no clear low-power mode discontinuity in this converter, but the efficiency is quite decent even at 40mA load.

This converter, when thermally stressed, seems to result in the voltage increasing from the setpoint. I’m not sure if this is because of the heat affecting the multi-turn trimmer, bottom feedback divider resistor or due to drift of the internal reference inside the XL6009 (more likely).

Vout = 6V

It seems that 5V in for 6V out manages to run 2A of load acceptably, but likely at the thermal limit. At 5V in, the peak efficiency is 88% near 400mA of load, but that is a very small step-up ratio (so high efficiency is expected).

Vout = 9V

Increasing the output to 9V requires a higher step-up ratio and it seems that around 7.5V is needed to get 2A of output. The “thermal cliff” is much more apparent on this one – it seems that the voltage rises initially and then dips dramatically – perhaps the converter is shutting down or the Schottky is hot enough to be so leaky that it’s not rectifying properly.

Vout = 12V

A common output voltage, but the high step-up ratio means that operation with 3V in seems nearly impossible for any reasonable load. It’s more comfortable with 4V+ input, but again, there seems to be a line for tolerable load. At 5V input, it can just about handle 0.8A at 12V output. I suppose that’s enough to run a few computer-style fans at full speed from a USB 5V output.

Vout = 15V

Vout = 20V

Vout = 24V

Another very common voltage. By now, the graphs are more detailed due to more steps being present on the voltage axis. At a 2:1 ratio (12V to 24V), it seems it can handle 900mA at 24V, or about 21.6W. This isn’t quite what you might expect from the “4A” claimed current and might have to do with the inductor. Nevertheless, achieving 2A of load current is becoming thermally impossible. An efficiency over 90% is achievable, but for load currents around 0.5A.

For a more ambitious ratio of 5V to 24V (e.g. USB to passive PoE), you don’t get much current to play with at 0.35A or so at 78% efficiency, with the peak efficiency of 85% at around 0.17A. Maybe barely enough to run a light endpoint.

Vout = 28V

Vout = 32V

This is the highest voltage tested and the limitations on output current with regards to input voltage are much more apparent, but if there’s anything the graphs teach us, it’s to have reasonable expectations and perhaps the need to not just blindly believe specifications without thinking.

Quiescent Current

The following tests are performed on selected common voltages instead. Quiescent current was measured with a Keithley 2450 SMU.

Definitely not low-quiescent in my books, with the quiescent depending on the step-up ratio and output voltage set-point. Going from 12V to 24V, expect about 11-12mA of quiescent current. Going from 5V to 12V, expect about 14mA of quiescent current. Going from 5V to 24V, expect about 32mA of quiescent current.

Ripple and Noise

So far, the tests all seem reasonable, but this is where things become a whole lot less impressive. Measurements were made using a Rohde & Schwarz MXO4 with passive 10:1 probes using a spring ground and 20MHz bandwidth limit.

Vin = 5V, Vout = 12V

Unloaded, the ripple was about 35mV with irregular timing, likely a bit of energy saving going on.

At 100mA of load, the ripple grows to 124mV peak-to-peak with an odd rounded sawtooth shape. This is pretty close to 1% or about the amount that sensitive electronics (e.g. computer equipment) can deal with.

At 500mA of load, it reached 448mV peak-to-peak. That’s very significant and is perhaps too noisy for electronic loads. It might not bother basic things like solenoids or motors, but the amount of noise also has the potential to impact on radio-frequency applications causing interference to reception. The ripple was strong enough that it interacted with the DC electronic load’s regulation loop, causing the load current to be modulated (in blue)!

I did try increasing to 1000mA of load, but the converter output collapsed before I had the time to take the measurements. It was not thermally capable of sustaining such a load.

Vin = 12V, Vout = 24V

At a more gentle ratio, unloaded, the ripple measured 64mV peak-to-peak.

At 100mA of load, the ripple was 137mV which is fine for this output voltage.

At 500mA, ripple reached 372mV peak-to-peak. This may be a bit high for sensitive 24V electronics.

At 1A of load, the ripple reached 592mV peak-to-peak. The modulation of the load’s current draw is very clear – it seems the load isn’t responding fast enough to ensure that the drawn current remains reliably at the 1A set-point.

Transient Response

Vin = 5V, Vout = 12V – 50/450mA load-step @ 1kHz 50% duty

It seems when the load comes on, the converter seems to increase in noise while slowly recovering (>500us). When the load goes off, there is a short overshoot (0.3V, 30us) before returning back to normal. The peak-to-peak excursion measured 730mV.

Vin = 12V, Vout = 24V – 50/450mA load-step @ 1kHz 50% duty

The behaviour is subtly different at other voltages despite the methodology being the same. In this case, the converter’s response to the load coming on is taking about 120us to arrest it and start trending back to the normal. Despite not fully recovering, when the load goes off, the overshoot spike does not quite reach the steady-state and the converter appears to continue its recovery. This seems to suggest the converter is a bit more sluggish under this state.

Conclusion

As is often the case when dealing with Chinese-made modules, the specifications are often in-part misleading, overly-optimistic and prone to misinterpretation, especially for the less experienced. As with step-up switching converters, the output current depends on a few factors, including worst-case input voltage, output voltage and inductor current with the focus on balance (i.e. power out <= power in). But in reality, there are additional constraints, such as thermal and self-power consumption that need to be considered.

This module can be a trap for makers who don’t know what to expect and get carried away with the supposed 4A rating. Based on my tests, I wouldn’t ever want to run it past 2A as it was showing signs of thermal stress, but depending on the step-up ratio, the amount of load current it can handle might be as low as 100mA (or even nothing) before the output voltage collapses below the set-point. The point where it operates efficiently is now something that can be determined now that I have run it through the gamut of tests, but in doing so, I’ve discovered that the Schottky diode can cook itself causing the module to fail, thus the module is not foolproof. Small modules just don’t have enough footprint to adequately act as a heatsink.

The biggest trap, in my opinion, is just how noisy the module is. The power that comes out of it has significant ripple and noise, which can cause connected loads to malfunction if they are sensitive electronic loads. Not withstanding this, the noise could also radiate and cause issues with radios, affecting reception of the AM/shortwave bands or even more. Perhaps this can be improved with better layout and proper choice of ceramic bypass capacitances alongside bulk capacitance. The regulation loop is not the fastest, but still, unless you have severe load-oscillations, it might still be acceptable. Increasing the output capacitance will likely have an effect on load transient response.