A few years ago, while managing the power management product line at work, I started an initiative with the development team to optimize new products by achieving ESE. ESE stands for Equations = Simulations = Experimentation. The idea is centered on the engineering goal of product design to verify that the systems design equations match the simulation results and ultimately the experimental results. When these three items match, not only do you understand a system, but you have the best chance to optimize a better solution. I’ll have to say that in today’s mad dash to get new products out the door, achieving ESE is not always possible. But to break through the ordinary and have a chance for the extraordinary, I would say this is a requirement. Since this power supply is just a fun design for an upcoming nixie tube clock project of mine, I have the time to achieve ESE. While in Part 1, I described the equations and simulations, in this Part 2, I collected experimental results to complete the design. In the process of finalizing the design, I was able to discover a couple of key design improvements and I’ll share these changes with you. The updated schematic, BOM, Kicad Layout, and design files are located at Github. The Updated PCB board (rev 2) can be ordered from oshpark.com
Here is a quick video showing six IN-4 Nixie tubes being powered by a 5v iPhone charger. (Enable Pin Pulled Low)
Nixie Tube Power Supply Specification
- Input Voltage (Vin): 5v (Can also work from 12v input)
- Output Voltage (Vout): 166.7v
- Output Load Current at Vin=5v (Io): 6 x 3ma = 18ma
- Maximum input power (Pmax): 4.1Watts
- Power Inductor: 33uH
- Typical Duty ratio at full power: 72%
- Inductor Ripple Current (@Pmax): 2.6A
- Efficiency: 73% (as high as 81% with larger sized inductor)
- (18mA load, 5v input) (1.1watts of Loss at 3Watt output)
- 82% efficiency at 12v input (As high as 87% with larger sized inductor)
- Temperature Rise (18mA load):
- L1 Inductor: 22 degree C (15 degrees C at 15mA)
- Q1: 10 degree C
- D1: 7 degree C
- PCB Footprint: 14.5 cm2
- Maximum Component Height: 10mm
- Enable Pin to remotely disable the supply
- Soft Start Time (output voltage rise time): 25ms
- Input Connector for 5v: USB Micro b
Key Design Tweaks and Discoveries from Part 1
- Reduced the gate drive resistor, R6, to zero ohms for Improved Efficiency.
- Eliminated Turn off losses and improved efficiency by using a 560pF snubber capacitor for C15.
- Reduced switching frequency to 40kHz by increasing C7 for Improved Efficiency.
- Replaced the Schottky diode, D1, with an Ultrafast Silicon version for improved efficiency.
- Discovered the inductor Core Material losses account for at approximately 50% of the total losses.
- Reduced the OCP Limit, R12, to 2.9 amps to limit current during overload conditions.
- Increased the Rating of the input capacitors so the power supply can also operate from 12v.
- Fixed the pin layout of the MMBTA42, the layout had the emitter and base swapped.
Below is the final schematic, both in JPG and PDF. I also updated Part 1 of the blog, Designing a Small Footprint, Low Profile 5v to 170v Nixie Tube Power Supply (Part 1) and the Github location containing the Kicad Layout, Schematic, and Bom so only the latest version is being published today.
Eliminating the Power Fet Turn Off Loss
One of the design tweaks discovered during the experimental verification was that the DCM boost converter can reduce the power FET turn-off loss by adding a snubber capacitor, C15 in parallel with Q1. And because the boost is operated in DCM, the energy stored in C15 will be recycled back to the input! With this snubber capacitor, a typical power FET, like Q1, in a switching converter will produce power dissipation during turnoff. This is simply due to the fact that during turn-off, the power FET must block a relatively high voltage (i.e. 170v in the case our converter) and still conduct current until the current commutates from the power FET, Q1, to the power diode, D1. One way to reduce this loss is to reduce the gate drive resistor, R6, to zero ohms. This was a nice step, saving 0.3 Watts of loss. However, even with R6 equal to zero ohms, the image below shows the power dissipation in Q1 when the power FET is turned off is still approximately 0.48 Watts.
The top trace shows the power FET drain voltage rising from zero volts (full on state) to 170v (full off state). The bottom trace shows the same current through the FET. If we calculate the total energy dissipated during this period (ie. integrate the voltage times current) and then multiply by the switching frequency the turn off losses are calculated as follows:
Where I_L is the 2.4A peak current and F is the 40kHz switching frequency, T_off is the 60ns turn off time. P_FETOFF = 0.49Watts. However, by adding the correct value of C15, this power loss can be reduced to zero.
Using a value of C15 = 330pF reduces the losses as shown.
While not zero, the image shows that the current in the power FET has been reduced below 1 amp (See channel 1: The scope is looking at the 0.1ohm sense resistor voltage).
When C15 = 560pF, the power dissipation during turnoff is near zero watts.
At the end of the turn off transition, the energy in the snubber capacitor equals:
You may wonder where the energy stored in C15 goes? For a DCM boost inductor, this energy is recycled back to the input. Can you figure out where in this picture below that recycling occurs?
The combination of the R6 gate drive resistor and the snubber capacitor saved a total of 0.8watts of power.
Reducing the Switching Frequency
When the snubber capacitor was added, you will notice in the picture above that it takes some time for the current in the boost inductor to reach zero amps. Otherwise, if the power FET turns back on before this current reaches zero current, the boost converter losses will become quite high. To ensure there was an adequate margin for the inductor energy (i.e. current) to reset, the switching frequency was reduced to 40kHz.
Replacing the Schottky Diode
Replacing the Schottky diode saved an additional 0.1watt of dissipation. Schottky diodes are nice because they have a low on voltage (i.e. 0.4v) and are extremely fast. However, at high voltages (i.e. 170v), they suffer from a large leakage current and this gets worse as the temperature rises. Although an ultra-fast silicon diode has a 0.8v on voltage, the much lower leakage current in this design was the optimal choice of component.
The image below shows the fast transient response seen with this DCM boost converter. Chan 2, the AC coupled output voltage, drops only 74mV with a 15mA step load on the output. This phase margin looks to be greater than 45 degrees. A more precious analysis is not warranted.
Soft Start of the output voltage
The rise time of the output voltage is approximately 25ms, limiting the inrush current of the supply at startup or when the enable pin is pulled low. In the circuit, Capacitor C14 and R7 are adjusted to perform this function by limiting how fast the COMPensator pin on the UCC3803 can rise. R9 is used to reset C14 when power is removed.
Core Losses Account for 50% of the Power Dissipation
The biggest discovery during the verification was that core losses in the power inductor swamped all other losses. As a result, at full load, the efficiency of the 5v to 170v DCM boost converter reached an optimum 73% efficiency, different from the initial 90% efficiency estimated in Part 1. Through experiments and additional calculations, the big difference was largely contributed to magnetic core losses. These losses are quite difficult to calculate. However, through experimentation and calculations, the following loss distribution was obtained.
At full load, the efficiency was also measured as the input voltage was raised to 12v. The best efficiency of 82% at 3.1watts of output voltage was achieved as the input voltage reached 13volts. To achieve a safe 13v operating condition, the input capacitor voltage ratings were increased from 10v to 16v.
The following table shows the final itemized losses found in the final design at 3.1 watts of load power. This was obtained through both experimental and calculations means. The exact equations for each parameter are shown in the boost_Flyback_Design.ods file located in the Nixie Power Supply Github repository.
|FET Turn off||0.032|
|Inductor Magnetic Core
|Diode Rev. Recovery||0.000|
|Input Trace Resistance||0.022|
Battle of the Inductors
By sampling of several types of inductors, improvements in the core loss can be obtained, at the expense of a larger size and footprint. Three inductors were chosen, the original shielded Wurth Inductor, an unshielded Bourns Inductor, and a high-performance, High current Epcos Inductor.
With the Epcos inductor the efficiency increase to 81% at 5v, a saving of 0.4watt of power, mostly by improved core losses but some inductor conduction losses. At 12 input, the efficiency with the Epcos core increased even more to 87%. The Bourns performed worse than the Wurth inductor when mounted on the PCB, while slightly better when mounted away. This indicates that the stray magnetic fields were causing losses in the copper ground plane. Because the Wurth Inductor is fully shielded, this is not an issue and is the best overall choice for low profile and footprint. For the best performance, the Epcos was great but has a height of 13mm and a larger footprint.
The Three Sampled Inductors:
- Wurth, FIXED IND 33UH 4.2A 45 MOHM SMD, MPN: 7447709330 : 73% Eff.
- Bourns, FIXED IND 33UH 3.1A 65 MOHM SMD, MPN: SDR1307-330KL : 72% Eff.
- Epcos, FIXED IND 30UH 8.5A 16.5 MOHM, MPN: B82559A0303A0: 81% Efficiency
Reduced the peak current limit to 2.9amps
R12 was reduced to 4.12k to reduce the peak current to 2.9amps. The design consideration was to adjust the value so that it will start to limit at the point the boost converter cannot reset to zero amps completely (see the discussion on turn off losses). This limits the output power to approximately 3.5 watts with a 5v input supply. At 12v of input, this peak current can be increased to increase the maximum output power.
Summary and Next Steps
The DCM boost converter is a good topology choice for a nixie tube display. The transient performance is great, and the efficiency has been optimized using off the shelf magnetics and components. And for the size, I can’t see anything more efficient. A flyback converter could reduce conduction losses, but would need a higher switching frequency so the net result may not be any better. Maybe a more exotic soft switched flyback topology could achieve higher efficiency using custom magnetics. This may be a topic for a future blog. 🙂
I’m also working on a taking the latest 1.5″x1.5″ inch PCB layout and shrinking it to 1.5″x1″ along with four mounting holes, a USB input connector, a terminal strip out, and components on one side of the PCB. This work is in progress and also located on the Github repository in the KC_NixieSupply5vto170vMini directory.
In any case, I hope you enjoy this design and get a good use of it for your next Nixie Tube display. I’m going to use this supply for my upcoming nixie tube clock design. Doesn’t everyone need to eventually build at least one of Nixie Tube clocks? Drop me a comment below if you have any questions or comments. I would love to hear your feedback.
Sorry about the screen captures, I need a better way to capture screen images (i.e. from a GPIB port into Linux). Let me know what solution you have?
This design is intended as an Open Source Hardware. I still need to add the appropriate disclaimers ….
The Surf has really sucked this week. 🙂